Scolaris Content Display Scolaris Content Display

Intervenciones individuales, familiares y escolares para conductas de riesgo múltiple en los jóvenes

Contraer todo Desplegar todo

Antecedentes

La participación en conductas de riesgo múltiple puede tener consecuencias adversas para la salud durante la niñez, durante la adolescencia y en etapas posteriores de la vida; no obstante, se sabe poco acerca de la repercusión de diferentes tipos de intervenciones dirigidas a las conductas de riesgo múltiple en los niños y los jóvenes, o la repercusión diferencial de enfoques universales versus específicos. Los resultados de las revisiones sistemáticas son contradictorios y los efectos de estas intervenciones no se han calculado de forma cuantitativa.

Objetivos

Examinar los efectos de las intervenciones implementadas hasta los 18 años de edad para la prevención primaria o secundaria de las conductas de riesgo múltiple entre los jóvenes.

Métodos de búsqueda

Se realizaron búsquedas en 11 bases de datos (Australian Education Index; British Education Index; Campbell Library; Cumulative Index to Nursing and Allied Health Literature [CINAHL]; Registro Cochrane Central de Ensayos Controlados [CENTRAL] en la Cochrane Library; Embase; Education Resource Information Center [ERIC]; International Bibliography of the Social Sciences; MEDLINE; PsycINFO; y en Sociological Abstracts) en tres ocasiones (2012, 2015 y 14 de noviembre de 2016). Se realizaron búsquedas manuales en las listas de referencias, se estableció contacto con expertos en el tema, se realizaron búsquedas de citas y se buscó en sitios web de organizaciones relevantes.

Criterios de selección

Se incluyeron los ensayos controlados aleatorios (ECA), incluidos los ECA con asignación al azar grupal, que tuvieron como objetivo considerar al menos dos conductas de riesgo. Los participantes fueron niños y jóvenes de hasta 18 años de edad y los padres, los tutores o los cuidadores, siempre que la intervención se orientara a considerar la participación en conductas de riesgo múltiple en niños y jóvenes de hasta 18 años de edad. Sin embargo, los estudios podían incluir datos de resultado sobre niños > 18 años de edad en el momento del seguimiento. Específicamente, se incluyeron los estudios que obtuvieron resultados de participantes de ocho a 25 años de edad. Además, solo se incluyeron los estudios con un período combinado de intervención y de seguimiento de seis meses o más. Se excluyeron las intervenciones dirigidas a los individuos con trastornos diagnosticados clínicamente junto con las intervenciones clínicas. Las intervenciones se categorizaron de acuerdo a si se realizaron a nivel individual; a nivel familiar; o a nivel escolar.

Obtención y análisis de los datos

Se identificaron 34 680 títulos, se examinaron 27 691 artículos y se evaluaron 424 artículos de texto completo para determinar su elegibilidad. Dos o más autores de la revisión, de forma independiente, evaluaron los estudios para su inclusión en la revisión, extrajeron los datos y evaluaron el riesgo de sesgo.

Los datos se agruparon en los metanálisis mediante el modelo de efectos aleatorios (DerSimonian and Laird) en RevMan 5.3. Para cada resultado, se incluyeron subgrupos relacionados con el tipo de estudio (nivel individual, familiar o escolar y enfoque universal o específico) y se examinó la efectividad hasta los 12 meses de seguimiento y a más largo plazo (> 12 meses). La calidad y la certeza de la evidencia se evaluaron mediante el enfoque Grades of Recommendation, Assessment, Development and Evaluation (GRADE).

Resultados principales

En la revisión se incluyeron 70 estudios elegibles, de los cuales una proporción significativa eran estudios universales basados en la escuela (n = 28; 40%). La mayoría de los estudios se realizó en EE.UU (n = 55; 79%). En promedio, los estudios procuraron prevenir cuatro de las conductas primarias. Las conductas consideradas con mayor frecuencia incluyeron el consumo de alcohol (n = 55), el consumo de drogas (n = 53) o el comportamiento antisocial (n = 53), seguidas del hábito de fumar (n = 42). Ningún estudio tuvo como objetivo prevenir la autolesión o el juego junto con otros comportamientos.

La evidencia indica que para las conductas de riesgo múltiple, las intervenciones universales basadas en la escuela tuvieron efectos beneficiosos con respecto al hábito de fumar (odds ratio [OR] 0,77; intervalo de confianza [IC] del 95%: 0,60 a 0,97; n = 9 estudios; 15 354 participantes) y el consumo de alcohol (OR 0,72; IC del 95%: 0,56 a 0,92; n = 8 estudios; 8751 participantes; ambos con evidencia de calidad moderada) en comparación con un comparador y que dichas intervenciones pueden ser efectivas para prevenir el consumo de drogas ilícitas (OR 0,74; IC del 95%: 0,55 a 1,00; n = 5 estudios; 11 058 participantes; evidencia de baja calidad) y la participación en cualquier comportamiento antisocial (OR 0,81; IC del 95%: 0,66 a 0,98; n = 13 estudios; 20 756 participantes; evidencia de muy baja calidad) hasta los 12 meses de seguimiento, aunque hubo evidencia de heterogeneidad moderada a significativa (I² = 49% al 69%). Evidencia de calidad moderada también mostró que las intervenciones escolares universales para las conductas de riesgo múltiple mejoraron las probabilidades de realizar actividad física (OR 1,32; IC del 95%: 1,16 a 1,50; I² = 0%; n = 4 estudios; 6441 participantes). Se consideró que los efectos observados eran de importancia para la salud pública al aplicarlos a nivel poblacional. La evidencia fue de menor certeza para los efectos de dichas intervenciones en las conductas de riesgo múltiple en cuanto al consumo de cannabis (OR 0,79; IC del 95%: 0,62 a 1,01; P = 0,06; n = 5 estudios; 4140 participantes; I² = 0%; evidencia de calidad moderada), las conductas sexuales de riesgo (OR 0,83; IC del 95%: 0,61 a 1,12; P = 0,22; n = 6 estudios; 12 633 participantes; I² = 77%; evidencia de baja calidad), y la dieta poco saludable (OR 0,82; IC del 95%: 0,64 a 1,06; P = 0,13; n = 3 estudios; 6441 participantes; I² = 49%; evidencia de calidad moderada). Es importante señalar que alguna evidencia apoyó los efectos positivos de las intervenciones escolares universales sobre tres o más conductas de riesgo.

Para la mayoría de los resultados de intervenciones específicas y universales a nivel individual y familiar, evidencia de calidad baja a moderada indica poco o ningún efecto, aunque se debe proceder con cuidado en la interpretación debido a que hubo pocos de estos estudios disponibles para la comparación (n ≤ 4 estudios para cada resultado).

Siete estudios informaron los efectos adversos e incluyeron evidencia que indica un aumento de la participación en una conducta de riesgo entre los participantes que recibieron la intervención en comparación con los participantes que recibieron las intervenciones control.

La calidad de la evidencia se consideró moderada o baja para la mayoría de los resultados, principalmente debido a las inquietudes en cuanto al sesgo de selección, de realización y de detección y la heterogeneidad entre los estudios.

Conclusiones de los autores

La evidencia disponible es más sólida para las intervenciones escolares universales dirigidas a conductas de riesgo múltiple, lo que demuestra que pueden ser efectivas para prevenir la participación en el hábito de fumar, el consumo de alcohol, el consumo de drogas ilícitas y el comportamiento antisocial, y para mejorar la actividad física entre los jóvenes, pero no para prevenir otras conductas de riesgo. Los resultados de esta revisión no aportan evidencia sólida de efectos beneficiosos de las intervenciones familiares o individuales a través de las conductas de riesgo estudiadas. Sin embargo, el informe deficiente y las inquietudes en cuanto a la calidad de la evidencia destacan la necesidad de estudios de alta calidad de intervenciones para las conductas de riesgo múltiple con el objetivo de fortalecer aún más la base de evidencia en esta área.

Resumen en términos sencillos

Intervenciones para la prevención de las conductas de riesgo múltiple en los jóvenes

Antecedentes

Las conductas de riesgo para la salud, como el hábito de fumar y el consumo de drogas, pueden confluir durante los años de la adolescencia, y la participación en dichas conductas de riesgo múltiple puede dar lugar a problemas de salud como lesiones y abuso de sustancias durante la niñez y la adolescencia, así como a enfermedades no transmisibles en etapas posteriores de la vida. Actualmente, no se sabe que intervenciones son efectivas para prevenir o reducir estas conductas peligrosas entre los niños y los jóvenes.

Métodos de búsqueda y selección de estudios

Se realizaron búsquedas minuciosas en bases de datos científicas múltiples para identificar estudios que analizaran formas de prevenir o reducir la participación en dos o más conductas de riesgo, incluido el hábito de fumar, el consumo de alcohol, el consumo de drogas ilícitas, el juego, la autolesión, la conducta sexual de riesgo, el comportamiento antisocial, el comportamiento de riesgo al conducir, la inactividad física y la nutrición deficiente, en jóvenes de ocho a 25 de años edad. Estos estudios se dividieron en grupos (estudios a nivel individual, a nivel familiar y a nivel escolar) en dependencia de si los investigadores trabajaron con individuos, familias o niños y jóvenes en las escuelas, respectivamente. Se consideraron específicamente los estudios "de referencia (gold standard)" ‐ ensayos controlados aleatorios que tuvieron como objetivo examinar dos o más conductas de interés.

Resultados principales

En total, 70 estudios fueron elegibles para su inclusión en esta revisión. La mitad incluyó a poblaciones en las que no se consideró el estado de riesgo y la mitad se centró en grupos de mayor riesgo. La mayoría se realizaron en los Estados Unidos o en países de altos ingresos. En promedio, los estudios examinaron los efectos de las intervenciones en cuatro comportamientos, los más frecuentes fueron el consumo de alcohol, el hábito de fumar, el consumo de drogas y el comportamiento antisocial.

Se encontró que para las conductas de riesgo múltiple, los estudios basados en la escuela para todos los jóvenes son más beneficiosos que un comparador para prevenir el hábito de fumar, el consumo de alcohol y la inactividad física, y que también pueden ser beneficiosos con respecto al consumo de drogas ilícitas y el comportamiento antisocial. Los resultados fueron más débiles para el consumo de cannabis, el comportamiento sexual de riesgo y las dietas poco saludables. La evidencia indica que determinados programas escolares podrían tener una repercusión beneficiosa sobre más de un comportamiento. Por el contrario, no se encontró evidencia sólida de los efectos beneficiosos de las intervenciones en las familias o los individuos con respecto a las conductas de interés, aunque se debe proceder con cuidado al interpretar estos resultados debido a que se identificó una cantidad menor de estos estudios. Por último, se encontraron siete estudios que informaron niveles mayores de participación en las conductas de riesgo entre los que recibieron la intervención en comparación con los que recibieron el control.

En términos generales, los autores de la revisión consideraron que la calidad de la evidencia era moderada o baja para la mayoría de las conductas examinadas al utilizar los criterios estandarizados, y se encontró que un comportamiento tuvo evidencia de calidad muy baja. En parte, lo anterior se debió a las inquietudes en cuanto a cómo se realizaron algunos estudios, lo que podría haber introducido sesgo.

Conclusiones

Los resultados indican que las intervenciones escolares ofrecidas a todos los niños dirigidas a considerar la participación en las conductas de riesgo múltiple pueden desempeñar una función en la prevención del hábito de fumar, el consumo de alcohol, la consumo de drogas ilícitas y el comportamiento antisocial, así como en la mejoría de la actividad física entre los jóvenes, pero no en las otras conductas examinadas. No se encontró evidencia sólida del efecto beneficioso de las intervenciones en las familias o los individuos. Las inquietudes en cuanto al informe de los estudios y la calidad de los estudios destacan la necesidad de estudios adicionales consistentes y de alta calidad para fortalecer aún más la base de evidencia en esta área.

Authors' conclusions

Implications for practice

For this review, we have conducted quantitative syntheses to obtain the best available estimate of the effectiveness of multiple risk behaviour interventions among young people. We report that school‐based programmes provided universally without consideration of individual risk are likely to be effective in preventing tobacco use, alcohol use, and physical inactivity (moderate‐quality evidence) and may also be beneficial in relation to illicit drug use (low‐quality evidence). We identified that such interventions may also be effective in relation to antisocial behaviour, although the evidence was of very low quality, but low‐quality evidence related to sexual risk behaviour was less certain. For these outcomes, we considered the size of the effect to represent potential public health importance at the population level. Because such interventions show promise, there is scope to consider adaptation of universal school‐based models to particular contexts and implementation more widely, although we note that some caution is warranted in their interpretation owing to the low or very low quality ratings for evidence related to certain outcomes and the proximity of upper 95% confidence interval (CI) levels to one. In contrast to universal school‐based interventions, and in the context of identifying fewer such studies, we identified that individual‐ and family‐level interventions may have little or no benefit in relation to the outcomes considered.

Findings reported in this review provide the foundation for research that will assist with commissioning and decision‐making around investment or disinvestment in different types of interventions that aim to prevent engagement in multiple risk behaviours among young people. In this way, the review may contribute to shaping future service delivery and the nature of family‐ or school‐based preventive programmes and curricula. Our findings may also play a role in informing national and international guidance around public health interventions and approaches to behaviour change, such as guidelines for preventing smoking, improving physical activity, and preventing drug misuse (such as those of the UK National Institute for Health and Care Excellence (NICE); https://www.nice.org.uk).

Implications for research

Conduct of this review has highlighted a number of implications for future research and evaluation. First, interventions were heterogeneous in relation to the age of participants, intervention domains, duration, and outcomes assessed. Although such heterogeneity is to be expected with multiple risk behaviour interventions implemented throughout childhood and adolescence (up to age 18), replication studies of promising models would help to strengthen the evidence base around particular components or intervention characteristics that are effective. We also found a lack of consistency among outcomes assessed and note that greater consistency, or use of a core outcome dataset reflecting outcomes that pose the greatest harm to public health, would assist with quantitative analyses of the effects of interventions.

Furthermore, rigorously conducted and adequately powered randomised controlled trials and replication studies are clearly needed in this field to strengthen available evidence. Interventions must be characterised to a greater extent by adequate sample sizes, published protocols, and clear reporting and robust methods, including adjustments for clustered data and imputation for missing data, when necessary. The conduct of comprehensive process evaluations would also be useful to enable detailed exploration and analysis of whether interventions were conducted as planned, any changes that were introduced, mechanisms of action, and the impact of contextual factors on outcomes. In this way, it would be possible to examine how implementation affected outcomes and the potential causal pathways of different interventions (Moore 2015).

Last, given that most interventions were conducted in high‐income countries, notably the USA, further research is needed to adapt interventions to wider geographical contexts, enabling the development of tailored and culturally appropriate interventions that are effective in a range of sociodemographic, educational, and geographical environments.

Summary of findings

Open in table viewer
Summary of findings for the main comparison.

Summary of findings table for the effectiveness of targeted individual‐level multiple risk behaviour interventions compared to usual practice for outcomes up to 12 months post intervention

Patient or population: children and young people aged 0 to 18 years

Settings: varied settings (home, kindergarten, primary school, secondary school, clinic, community)

Intervention: multiple risk behaviour interventions

Comparison: no intervention/usual practice

Outcomes

Risk with usual practice

Risk with intervention

(95% CI)

Relative effect
(95% CI)

No. of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Tobacco use

156 per 1000

191 per 1000

(122 to 288)

OR 1.28
(0.75 to 2.19)

521
(2 RCTs)

⊕⊕⊕⊝
Moderatea

Alcohol use

613 per 1000

618 per 1000

(559 to 675)

OR 1.02
(0.80 to 1.31)

1204
(4 RCTs)

⊕⊕⊕⊝
Moderatea

Cannabis use

110 per 1000

120 per 1000

(79 to 179)

OR 1.10
(0.69 to 1.76)

126
(2 RCTs)

⊕⊕⊕⊝
Moderatea

Illicit drug use

32 per 1000

30 per 1000

(23 to 400)

OR 0.94
(0.71 to 1.25)

638
(3 RCTs)

⊕⊕⊕⊝
Moderatea

Antisocial behaviour

145 per 1000

170 per 1000

(135 to 213)

OR 1.21
(0.92 to 1.60)

764
(4 RCTs)

⊕⊕⊕⊝
Moderatea

Vehicle‐related risk behaviour

81 per 1000

49 per 1000

(12 to 179)

OR 0.59

(0.14 to 2.48)

94
(2 RCTs)

⊕⊝⊝⊝
Very lowb

Sexual risk behaviour

610 per 1000

533 per 1000

(434 to 628)

OR 0.73
(0.49 to 1.08)

494
(2 RCTs)

⊕⊕⊕⊝
Moderatea

Physical activity

134 per 1000

N/A

No studies in meta‐analysis

aDowngraded owing to high risk of bias due to lack of blinding and/or unclear risk of bias across additional domains.

bDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains, as well as imprecision related to width of the 95% confidence interval of the summary estimate and inconsistency between effect estimates (I² = 81%).
Note that variation was evident in measures of risk with usual practice. Baseline risk measures were calculated at follow‐up. When no data were reported for any study in that meta‐analysis, baseline measures were used.

CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Open in table viewer
Summary of findings 2.

Summary of findings table for the effectiveness of universal individual‐level multiple risk behaviour interventions compared to usual practice for outcomes up to 12 months post intervention

Patient or population: children and young people aged 0 to 18 years

Setting: varied settings (home, clinic, community)

Intervention: multiple risk behaviour interventions

Comparison: no intervention/usual practice

Outcomes

Risk with usual practice

Risk with intervention

(95% CI)

Relative effect
(95% CI)

No. of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Tobacco use

32 per 1000

33 per 1000

(10 to 98)

OR 1.03
(0.32 to 3.27)

1549
(2 RCTs)

⊕⊕⊝⊝
Lowa

Alcohol use

41 per 1000

33 per 1000

(24 to 45)

OR 0.80
(0.58 to 1.11)

1911
(4 RCTs)

⊕⊕⊕⊝
Moderateb

Cannabis use

264 per 1000

198 per 1000

(142 to 272)

OR 0.69
(0.46 to 1.04)

362
(2 RCTs)

⊕⊕⊕⊝
Moderateb

Illicit drug use

‐‐

N/A

No studies in meta‐analysis

Antisocial behaviour

131 per 1000

133 per 1000

(85 to 203)

OR 1.02

(0.62 to 1.69)

200

(1 RCT)

⊕⊕⊕⊝
Moderateb

Sexual risk behaviour

396 per 1000

216 per 1000

(84 to 450)

OR 0.42
(0.14 to 1.25)

162
(1 RCT)

⊕⊕⊕⊝
Moderateb

Physical activity

No data available to estimate risk

N/A

OR 1.11
(0.74 to 1.67)

1,530
(2 RCTs)

⊕⊕⊕⊝
Moderateb

aDowngraded owing to high risk of bias in relation to blinding and incomplete outcome data. We also downgraded the certainty of evidence owing to inconsistency because between‐study variance was high and variability was evident in the effect estimates of each study. The 95% CIs of one of the studies were wide, but researchers reported very few events, so certainty of evidence was not downgraded on this basis.

bDowngraded owing to high risk of bias due to lack of blinding and/or unclear risk of bias across additional domains.

Note that variation was evident in measures of risk with usual practice. Baseline risk measures were calculated at follow‐up. When no data were reported for any study in that meta‐analysis, baseline measures were used.

CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Open in table viewer
Summary of findings 3.

Summary of findings table for the effectiveness of targeted family‐level multiple risk behaviour interventions compared to usual practice for outcomes up to 12 months post intervention

Patient or population: children and young people aged 0 to 18 years

Setting: varied settings (home, community)

Intervention: multiple risk behaviour interventions

Comparison: no intervention/usual practice

Outcomes

Risk with usual practice

Risk with intervention

(95% CI)

Relative effect
(95% CI)

No. of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Tobacco use

176 per 1000

143 per 1000

(79 to 246)

OR 0.78
(0.40 to 1.53)

313
(2 RCTs)

⊕⊕⊕⊝
Moderatea

Alcohol use

269 per 1000

234 per 1000

(147 to 349)

OR 0.83
(0.47 to 1.46)

417
(3 RCTs)

⊕⊕⊕⊝
Moderatea

Cannabis use

180 per 1000

183 per 1000

(102 to 307)

OR 1.02
(0.52 to 2.02)

380
(3 RCTs)

⊕⊕⊝⊝
Lowb

Illicit drug use

265 per 1000

211 per 1000

(132 to 321)

OR 0.74
(0.42 to 1.31)

69
(1 RCT)

⊕⊕⊕⊝
Moderatea

Antisocial behaviour

291 per 1000

256 per 1000

(190 to 337)

OR 0.84
(0.57 to 1.24)

772
(5 RCTs)

⊕⊕⊕⊝
Moderatea

Sexual risk behaviour

750 per 1000

728 per 1000

(623 to 812)

OR 0.89
(0.55 to 1.44)

371
(3 RCTs)

⊕⊕⊕⊝
Moderatea

Physical activity

No data available to estimate risk

N/A

OR 0.72
(0.29 to 1.79)

61
(1 RCT)

⊕⊕⊕⊝
Moderatea

aDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains.

bDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains. The quality of the evidence was also downgraded on the basis of inconsistency because between‐study variance was high, and although I² was moderate, inconsistency was evident in effect estimates of individual studies, two of which had small sample sizes.

Note that variation was evident in measures of risk with usual practice. Baseline risk measures were calculated at follow‐up. When no data were reported for any study in that meta‐analysis, baseline measures were used.

CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Open in table viewer
Summary of findings 4.

Summary of findings table for the effectiveness of targeted school‐level multiple risk behaviour interventions compared to usual practice for outcomes up to 12 months post intervention

Patient or population: children and young people aged 0 to 18 years

Setting: school

Intervention: multiple risk behaviour interventions

Comparison: no intervention/usual practice

Outcomes

Risk with usual practice

Risk with intervention

(95% CI)

Relative effect
(95% CI)

No. of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Tobacco use

‐‐

‐‐

No data in meta‐analysis

Alcohol use

‐‐

‐‐

No data in meta‐analysis

Cannabis use

‐‐

‐‐

No data in meta‐analysis

Illicit drug use

50 per 1000

38 per 1000 (27 to 53)

OR 0.75
(0.53 to 1.06)

2454
(3 RCTs)

⊕⊕⊕⊝
Moderatea

Antisocial behaviour

No data available to estimate risk

N/A

OR 0.78
(0.59 to 1.05)

1,531
(3 RCTs)

⊕⊕⊕⊝
Moderatea

Sexual risk behaviour

‐‐

‐‐

No data in meta‐analysis

Physical activity

‐‐

‐‐

No data in meta‐analysis

aDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains.

Note that variation was evident in measures of risk with usual practice. Baseline risk measures were calculated at follow‐up. When no data were reported for any study in that meta‐analysis, baseline measures were used.

CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Open in table viewer
Summary of findings 5.

Summary of findings table for the effectiveness of universal school‐level multiple risk behaviour interventions compared to usual practice for outcomes up to 12 months post intervention

Patient or population: children and young people aged 0 to 18 years

Setting: school

Intervention: multiple risk behaviour interventions

Comparison: no intervention/usual practice

Outcomes

Risk with usual practice

Risk with intervention

(95% CI)

Relative effect
(95% CI)

No. of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Tobacco use

54 per 1000

42 per 1000

(33 to 52)

OR 0.77
(0.60 to 0.97)

15,354
(9 RCTs)

⊕⊕⊕⊝
Moderatea

Alcohol use

163 per 1000

123 per 1000

(98 to 152)

OR 0.72
(0.56 to 0.92)

8751
(8 RCTs)

⊕⊕⊕⊝
Moderatea

Cannabis use

110 per 1000

89 per 1000

(71 to 111)

OR 0.79
(0.62 to 1.01)

4140
(5 RCTs)

⊕⊕⊕⊝
Moderatea

Illicit drug use

41 per 1000

30 per 1000

(21 to 44)

OR 0.73
(0.50 to 1.07)

10,266
(5 RCTs)

⊕⊕⊝⊝
Lowb

Antisocial behaviour

172 per 1000

141 per 1000

(117 to 168)

OR 0.79
(0.64 to 0.97)

17,722
(11 RCTs)

⊕⊝⊝⊝
Very lowc

Sexual risk behaviour

131 per 1000

112 per 1000

(87 to 146)

OR 0.84
(0.63 to 1.13)

12,633
(6 RCTs)

⊕⊕⊝⊝
Lowd

Physical activity

276 per 1000

335 per 1000

(307 to 364)

OR 1.32
(1.16 to 1.50)

6,441
(4 RCTs)

⊕⊕⊕⊝
Moderatea

aDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains.

bDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains. Downgraded an additional level on the basis of inconsistency because substantial heterogeneity was evident (I² = 69%, Chi² = 15.88, P = 0.007), between‐study variance was moderate, and inconsistency between effect estimates of individual studies was apparent, with absence of overlap between 95% CIs of certain studies in the subgroup.

cDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains. The quality of evidence was also downgraded on the basis of inconsistency because heterogeneity was substantial (I² = 68%, Chi² = 36.95, P = 0.0002), between‐study variance was moderate, and lack of overlap was apparent between 95% CIs for certain studies with large sample sizes. Last, evidence was downgraded on the basis of possible publication or small‐study bias.

dDowngraded owing to high risk of bias in relation to blinding and/or other domains. Certainty of the evidence was also downgraded owing to substantial heterogeneity (I² = 84%, Chi² = 25.07, P < 0.0001) and high between‐study variance, with lack of overlap between the 95% CIs of certain studies in the subgroup. Although there may be plausible explanations for such heterogeneity, these reasons could not be further investigated in this review.

Note that variation was evident in measures of risk with usual practice. Baseline risk measures were calculated at follow‐up. When no data were reported for any study in that meta‐analysis, baseline measures were used.

CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Background

Description of the condition

Adolescence and young adulthood represent critical periods in the life course in relation to current and future health and wellbeing (Lancet 2012; Lancet Commission on Adolescent Health & Wellbeing; Patton 2012; World Health Organization 2014). Many of the health risk behaviours that give rise to chronic non‐communicable diseases (NCDs) later in life, such as tobacco use, alcohol use, consumption of calorific foods, and physical inactivity, are initiated during adolescence (Resnick 2012; Sawyer 2012), and they can continue into late adolescence and young adulthood (Mahalik 2013; McCambridge 2011;Ortega 2013;Resnick 2012; Sawyer 2012; Wanner 2006). Engagement in such behaviours can increase risks of low educational attainment, antisocial behaviour, sexually transmitted infections, injury, and substance use dependence during adolescence and young adulthood, and can influence morbidity later in life (Chen 1995; Djousse 2009;Hall 2016; Mason 2010;McCambridge 2011;Ortega 2013;Silins 2015), thus affecting health throughout the life course (World Health Organization 2014). Globally, for instance, alcohol use (7% of disability‐adjusted life‐years (DALYs)), unsafe sex (4%), and illicit drug use (2%) are among the main risk factors for the incidence of DALYs among young people aged 10 to 24 years (Gore 2011; Mokdad 2013).

Estimates of the prevalence of concurrent tobacco smoking, drinking of alcohol, and recent illicit drug or cannabis use for adolescents in the UK, the USA, and Canada range from 6% to 13% (Connell 2009; de Winter 2016; Fuller 2015; Leatherdale 2010; McVie 2005; NHS 2008), and a recent UK report estimated that over 20% of young people aged 16 engage in two or more substance use and delinquent behaviours (Hale and Viner 2016). Critically, risk behaviours such as smoking, antisocial behaviour, alcohol consumption, and sexual risk behaviour have been shown to cluster in adolescence (Basen‐Engquist 1996; Burke 1997; de Looze 2014; Junger 2001; Mistry 2009; Pahl 2010; van Nieuwenhuijzen 2009), and engagement in one risk behaviour increases the likelihood of engagement in others. For example, both smoking and low vegetable intake at age 13.5 increased the odds of engagement in multiple health risk behaviours at age 16 by over twofold (de Winter 2016), and odds ratios for associations between use of individual substances and sexual risk behaviours range between 1.4 and 4.7 (Jackson 2012; Meader 2016).

Engagement in multiple risk behaviours therefore can be viewed as supporting the syndemic concept, whereby synergistic involvement in risk behaviours may worsen the outcomes of engagement in risk behaviours and associated comorbidities later in life (Mendenhall 2017). Given that adolescents comprise a quarter of the world’s population worldwide, and often more than a fifth of a country's population, engagement in multiple health risk behaviours and the impact of such engagement represent a significant public health concern. In recognition of the importance of investment in adolescent health as the foundation of health and wellbeing across the life course, recent literature has highlighted the need for greater focus on adolescent health worldwide and the global application of preventive interventions and policies (Lancet 2012; Patton 2014; Resnick 2012; World Health Organization 2014). Evidence has also highlighted the health, economic, and social returns that could be realised from greater global investment in adolescent health (Catalano 2012; Lancet Commission on Adolescent Health & Wellbeing; Sheehan 2017).

Description of the intervention

This review examines evidence for interventions that are universal in their approach (i.e. that address whole populations with the aim of preventing the onset or advancement of risk behaviours, as well as those that target particular groups who may be at higher risk (e.g. those identified through screening or other assessment of risk factors such as following referral from the criminal justice system)). Interventions provided at individual, family, and school levels, as well as those that encompass more than one of these domains, are considered. Thus, the interventions considered in this review are wide‐ranging in design and may be implemented in a range of settings by providers such as nurses, teachers, or peers, with the goal of impacting behaviours of young people up to 18 years of age.

Interventions focused at the individual level include mentoring, coaching, Internet‐level education, conditional cash transfers, development of prosocial networks, and motivational interviewing. Family‐focused interventions may involve group sessions or home visits and support, and they aim to improve child‐parent interactions, communication, the family environment (e.g. through conflict resolution and problem‐solving), parenting skills, parental support, resilience and wellbeing, and knowledge and awareness. Such programmes may incorporate components for children or adolescents, including adolescent skills‐building and decision‐making curricula, goal‐setting, or practice and reinforcement of skills and behaviours. Targeted family‐based interventions may be targeted to adolescents at higher risk, such as those who are homeless or are experiencing parental substance abuse.

School programmes aim to target normative beliefs, bonding to school, behavioural goals, and commitments to not engage in risk behaviours and knowledge. They do so by utilising a range of diverse strategies, including formal classroom curricula, peer delivery, behaviour management practices, role‐play, goal‐setting, and whole‐school approaches that aim to change the school climate or ethos. Such domains can be implemented either alone or alongside additional parent or community components, such as parent leaflets, parent‐child homework exercises, extracurricular activities, and community engagement activities. Targeted interventions delivered at the school level may focus on particular higher‐risk groups, such as those in lower socioeconomic groups, those demonstrating aggressive behaviour, and those identified as being at high risk of school dropout.

How the intervention might work

The goals of multiple risk behaviour interventions are to prevent engagement in two or more behaviours, to reduce the frequency of engagement in these behaviours, and to reduce the prevalence and impact of short‐ and long‐term negative consequences associated with engagement in those behaviours.

Interventions at individual, family, and school levels may have distinct hypotheses regarding mechanisms of effect, as discussed below. For instance, individual‐level interventions may focus more exclusively on improving motivation to act, identifying goals, obtaining normative feedback, coaching, and modelling positive behaviour, with some models such as mentoring based on the underlying hypothesis that providing positive role models, support, and prosocial aspirations can change behaviour and reduce risk.

Family‐based interventions may focus on provision of skills, knowledge, and support; frequency and quality of parent‐child communications; and reinforcement of shared values and behaviours. Grounding of several interventions in social development theory, with a focus on family as one 'unit of socialisation', suggests that sufficient engagement and involvement with family and subsequent positive reinforcement can enhance family attachment, thus helping to underpin strong bonds to school, increased likelihood of involvement with prosocial peers, and reduced likelihood of risk behaviour (Hawkins and Weis 1985). Building on cognitive models, interventions may also work by influencing perceptions of risk, behavioural intentions, and self‐regulation via recognition that risk behaviours may result from a reaction to circumstances conducive to risk‐taking, depending on the intention and willingness of the individual to engage in the behaviour and perceptions of risk associated with the behaviour (Gibbons 2007). In addition, interventions may act by improving parental monitoring and providing support. In this way, parents serve as a source of socialisation regarding norms and behavioural expectations,but also provide feedback that can influence attitudes and behaviour (Brody 2005; Murry 2009;Murry 2014).

School‐based programmes may aim to enhance knowledge, social and emotional skills, resilience, and social competence, thereby improving self‐esteem and self‐control and reducing the impact of negative peer, family, and/or social influences ‐ all of which can increase the risk of engagement in risk behaviours (Biglan 2004; Chen 1995; Hawkins 2005; Jackson 2010; Mason 2010). Alternatively, programmes may seek to reinforce engagement in healthy behaviours by providing positive role‐modelling, addressing perceptions of behaviours and their consequences, and considering social influences and norms.

Theories that seek to explain why risk behaviours cluster during adolescence are relevant to consideration of how interventions might work. First, Moffitt's theory of adolescence‐limited antisocial behaviour highlights two distinct categories of individuals with differing natural histories and etiologies of antisocial behaviour: the 'adolescence‐limited' group, whose behaviour is limited to adolescence and whose behaviour is normative but in whom risk behaviours may be temporarily sustained via mimicry of antisocial behaviour observed in antisocial peers; and the second, smaller group ‐ the 'life course persistent' group, whose members progress to become lifelong offenders (Moffitt 1993). Second, Jessor's problem behaviour theory (PBT) proposes that clustering of behaviours results from a complex web of interrelated predisposing and protective factors involving interaction between individual and environment (Jessor 1991;Jessor 1992). To date, studies have highlighted shared predisposing or protective factors at individual, intermediate, and structural levels, such as positive mental health, family attachment, peer relationships, socioeconomic status, social environment, and connection with school and religion (Beyers 2004; Catalano 2012; de Looze 2014; Hale and Viner 2016; Jackson 2010; Kipping 2014;Sawyer 2012; Viner 2006).

Determinants of engagement in risk behaviours during adolescence are therefore complex, and it is noteworthy that their antecedents may originate before birth or during the early years of life, and may accumulate early in the life course (Biglan 2004; Catalano 2012; Jessor 1991; Kuh 2003). Early adverse experiences and stressors such as violence, disease, and poor nutrition in infancy and early childhood can affect growth, health, and developmental milestones such as school readiness, literacy, and healthy peer relationships. Interventions that influence early determinants of risk are central to a life course approach and may have a greater impact on an individual's propensity to engage in risk behaviours during adolescence than those that focus on reducing behaviours or mitigating harms once the risk behaviours have become established, as outlined in the logic model (Figure 1). Interventions that provide support to mothers during pregnancy, for instance, may enhance maternal skills, promote healthy behaviours, and enhance emotional well‐being, which may increase mother‐child interaction and reduce environmental stressors (Biglan 2004; Eckenrode 2010). Interventions provided during the preschool years, which comprise training in parenting or increased preschool attendance, may prevent multiple risk behaviours later in life by reducing stressors within the family environment and by enhancing maternal and child skills (Biglan 2004; Hawkins 2005, Reid 1999; Shepard and Dickstein 2009; Tremblay 1995; Webster‐Stratton and Taylor 2001). If unchecked, however, risk can continue to accumulate from early life to adolescence, increasing the likelihood of peer rejection, lack of engagement with school, low academic achievement, and a trajectory towards engagement with risk behaviours (Catalano 2012; Sawyer 2012). Thus, interventions implemented during adolescence can build on investment in the early years and target those at higher risk, or can be implemented with the aim of protecting young people from normative increases in engagement in risk behaviours (Catalano 2012; World Health Organization 2014).


Logic Model: interventions to prevent multiple risk behaviours in individuals aged 8 to 25 years.

Logic Model: interventions to prevent multiple risk behaviours in individuals aged 8 to 25 years.

Why it is important to do this review

Whilst many health interventions aim to prevent single behaviours, and several Cochrane reviews have focused on specific types of interventions to address single behaviours (Carney 2016; Faggiano 2014;Fellmeth 2011; Foxcroft 2011;Livingstone 2010;Mytton 2006), less is known about the effectiveness of interventions that aim to simultaneously prevent a wide range of multiple risk behaviours (Biglan 2004; Jackson 2010). Given that risk behaviours cluster, and that determinants of engagement in these behaviours may overlap, it is possible that multiple‐behaviour interventions may be both efficient and effective. A recent scoping review examined characteristics of interventions to prevent multiple risk behaviours but focused on adult populations (King 2015). Of the two systematic literature reviews that have focused on prevention of multiple risk behaviour in young people, one focused on the impact of interventions to target substance use and sexual risk behaviour (Jackson 2010; Jackson 2011), and the other focused on interventions that target substance use, antisocial behaviour, and sexual risk behaviour (Hale 2014), while including only interventions that reported statistically significant effects. To date, therefore, no single Cochrane review has systematically examined evidence relating to the impact of interventions that address multiple behaviours. Critically, there remains no quantitative estimate of effect to guide public health decision‐making.

This review considers the effectiveness of individual‐, family‐, and/or school‐level interventions that aim to address tobacco use, alcohol use, illicit drug use, gambling, self‐harm, vehicle‐risk behaviours, antisocial behaviour, sexual risk behaviour, physical inactivity, and poor nutrition. This review is therefore broader with respect to the number of behaviours, settings, and populations of focus.

Given limited opportunities and resources to prevent risk behaviours, it is important to explore whether targeting multiple behaviours may be more efficient than targeting single behaviours. Greater understanding of the effects of multiple risk behaviour interventions in the context of tightening budgets has substantial potential to influence decisions around commissioning and/or de‐commissioning of risk prevention interventions for children and young people.

Objectives

Primary research objective

  • To examine the effects of interventions implemented up to 18 years of age for the primary or secondary prevention of multiple risk behaviours among individuals aged eight to 25 years (see MacArthur 2012 for the protocol of this review)

Secondary research objectives

  • To explore whether effects of the intervention differ within and between population subgroups

  • To examine whether effects of the intervention differ by risk behaviours and by outcomes

  • To investigate the influence of the setting of the intervention on design, delivery, and outcomes of the interventions

  • To investigate the relationship between numbers and/or types of component(s) of an intervention, intervention duration, and intervention effects

  • To evaluate whether the impacts of interventions differ according to whether behaviours are addressed simultaneously or sequentially and/or whether behaviours are addressed in a particular order

  • To explore the association between clustering of particular behaviours and effects of the interventions

  • To assess the cost‐effectiveness of the interventions

  • To consider the implications of the findings of this review for further research, policy, and practice

In this review, we aim to examine the effects of interventions on each of the studied behaviours, in turn, and through further analyses to ascertain the effects of these interventions on multiple risk behaviours.

Methods

Criteria for considering studies for this review

Types of studies

We included only randomised controlled trials (RCTs), including cluster RCTs, that aimed to address at least two risk behaviours of interest. We included only RCTs because studies using this design provide the most reliable type of evidence for assessing effects of interventions in that they minimise the risk that findings may have been influenced by confounding (Akobeng 2005). We included RCTs that primarily assessed effectiveness of interventions but also reported findings of a full or partial economic evaluation, and those that reported resource use or costs associated with an RCT intervention. We included only studies with a combined intervention and follow‐up period of six months or longer, to enable identification of the impact of interventions over the shorter term without exclusion of studies that were not able to monitor outcomes over a longer period.

Types of participants

Participants were children and young people aged up to 18 years. Studies were also included in which participants receiving the intervention were parents, guardians, carers, peers, and/or members of a school, as long as the intervention aimed to impact involvement in multiple risk behaviours among children and young people aged up to 18 years. We included interventions targeting participants in subgroups of the population, but we excluded interventions aimed at individuals with clinically diagnosed disorders.

Types of interventions

Interventions included in this review comprised interventions that aimed to address at least two risk behaviours from among regular tobacco use; alcohol consumption; recent cannabis or other regular illicit drug use; risky sexual behaviours; antisocial behaviour and offending; vehicle‐related risk behaviours; self‐harm (without suicidal intent); gambling; unhealthy diet; high levels of sedentary behaviour; and low levels of physical activity. We excluded interventions that addressed just two risk behaviours including unhealthy diet, low levels of physical activity, and/or high levels of sedentary behaviour, to avoid overlap with a previous Cochrane systematic review (Waters 2012). In addition, we excluded interventions that address two or more risk behaviours from among tobacco use, alcohol consumption, and/or drug use; a separate review will examine these interventions (Hickman 2014). In this way, we excluded interventions that target only healthy eating and physical activity, or only tobacco use, alcohol consumption, and drug use, for example, but included interventions that target healthy eating, physical activity, and risky sexual behaviour; or alcohol use, tobacco use, and antisocial behaviour. Last, we included RCTs delivered at the individual, family, or school level; another Cochrane review will include studies conducted at the community or population level, such as media campaigns or policy, regulatory, or legislative interventions, owing to their distinct study design (Campbell 2012). We classified studies as 'individual' if they recruited participants from the general community setting (but not from the school or family), and if they delivered the bulk of the intervention component(s) in one of the following settings: criminal justice (i.e. prisons or youth offending institutions), general practice surgeries, accident and emergency departments, or community‐based settings (for mentoring‐only interventions delivered to individuals within a community setting). We classified studies as 'school‐based' if researchers recruited participants from schools and delivered most of the intervention components in a school setting, and as 'family‐level' if investigators recruited parent(s) or child(ren) from the community and delivered most of the intervention components to the family within the home or in a neutral centre‐based environment.

Researchers compared those receiving the intervention versus those receiving usual practice, no intervention, or placebo or attention control. Interventions could be conducted at the individual, family, or school level and could include psychological, educational, parenting, or environmental approaches. As described above, interventions could be provided universally, without regard for the young people's level of risk, or they could be targeted to particular young people or families. Thus, for example, studies could be conducted at an individual level without regard for risk status (universal individual‐level interventions), or they could target particular groups of students in schools (targeted school‐level interventions). We classified studies as 'universal' in their approach if all school children within a school (or those in a particular year group), all individuals within a community/organisation, or all families within a community were eligible to participate in those studies. This contrasts with interventions classified as 'targeted', usually defined by participant characteristics (e.g. ethnicity, gender, pre‐existing behavioural problems/issues). However, for studies implementing an intervention for individuals/families in an area with a high crime rate, a high percentage of social deprivation, or a high percentage of black minority ethnic individuals, or for schools specially selected to include a certain percentage of students with a specific student ethnic population, we viewed interventions as 'universal', as not all participants would be subject to these characteristics. Interventions could start before the onset of behaviours (primary prevention), or they could target those currently engaged in risk behaviours (secondary prevention). We excluded stand‐alone clinical interventions (e.g. cognitive‐behavioural therapy).

Types of outcome measures

Primary outcomes

The primary outcome was the primary or secondary prevention of two or more risk behaviours in individuals aged eight to 25 years. This age range was chosen owing to the likelihood of engagement in risk behaviours over this age range and the impact(s) and public health importance of engaging in risk behaviours during this period of the life course. Relatively few studies have examined the epidemiology of multiple risk behaviours; therefore the review includes behaviours that have an adverse impact on health, whether or not the behaviour involves an active desire for 'risk‐taking' or immediate gratification. We excluded from this review risk behaviours such as lack of ultraviolet (UV) protection, disordered eating, disordered sleep, and the choking game based on available evidence regarding prevalence, adverse impact on health, or relatedness to included behaviours; or we did so to avoid overlap with, or incorporation of, clinically diagnosed disorders. Consultation with the Centre for the Development and Evaluation of Complex Interventions for Public Health ImpRovement (DECIPHer) Public Involvement Advisory Group ALPHA (Advice Leading to Public Health Advancement) and the advisory group for the Avon Longitudinal Study of Parents and Children (ALSPAC) has supported inclusion of the range of behaviours outlined below.

We categorised risk behaviours as follows.

  • Tobacco use: regular tobacco use.

  • Alcohol consumption: binge drinking (alcohol); heavy/hazardous drinking; regular or problem drinking.

  • Drug use: recent cannabis use; recent illicit drug use (other than cannabis); regular illicit drug use.

  • Antisocial behaviour and offending: murder; aggravated assault; sexual assault; violence (including domestic or sexual violence); assault with or without injury; gang fights; hitting a teacher, parent, or student; racist abuse; criminal damage; robbery; burglary/breaking and entering; vehicle‐related theft; prostitution; selling drugs; joy‐riding; carrying a weapon; engaging in petty theft or other theft; pan‐handling (begging); buying stolen goods; being noisy and rude; exhibiting disorderly conduct; being a nuisance to neighbours; graffiti (Biglan 2004; Hales 2010).

  • Self‐harm: self‐harm without suicidal intent.

  • Gambling: gambling; regular/uncontrolled gambling.

  • Vehicle‐related risk behaviours: cycling without a helmet; not using a car seatbelt; driving under the influence of alcohol, cannabis, or illicit substances.

  • Risky sexual behaviours: unprotected sexual intercourse; early sexual debut experience.

  • Activity levels: low levels of physical activity; high levels of sedentary behaviour.

  • Unhealthy diet: low levels of fruit and vegetable consumption; low‐fibre diet; high‐fat diet; high‐sugar diet.

We excluded behaviours reported as clinical disorders (e.g. substance use disorder representing a clinical diagnosis). We included studies that addressed behaviours via upstream precursors for which a hypothesis or a clear rationale for the pathway of effect from the precursor to the subsequent behaviour was reported. This was particularly relevant for studies targeting young children (e.g. in primary school settings).

Secondary outcomes

Secondary outcomes include potential medium‐ and longer‐term outcomes that interventions are aiming to effect.

  • Education and employment: educational qualifications; truancy and school exclusion; employment; not being in education, employment, or training (NEET).

  • Crime: criminal record/offending; re‐offending.

  • Long‐term addictive behaviours: smoking, alcohol, drugs, gambling.

  • Health outcomes: teenage pregnancy or parenthood; sexually transmitted infections; injuries; morbidity (e.g. hepatitis C, HIV, anxiety and depression, obesity, type 2 diabetes, fatty liver disease, liver cirrhosis); suicide/self‐harm; premature mortality.

  • Harm associated with the process or outcomes of the intervention: for instance, if the extent of engagement in risk behaviours or adverse health outcomes increases as a result of the intervention.

  • Cost‐effectiveness of the intervention: measures of resource use; costs; or cost‐effectiveness of the intervention (e.g. incremental cost‐effectiveness ratios (ICERs), incremental cost per quality‐adjusted life‐year (QALY); cost‐benefit ratio).

Given the longer‐term adverse consequences of engagement in multiple risk behaviours and the importance of sustained outcomes, we used a primary endpoint for outcome data at the longest follow‐up point post intervention, up to a period of 12 months. We grouped outcome data from interventions with longer duration of follow‐up into a longer‐term category, which included any outcome data collected after 12 months post intervention. When data from more than one time point were reported, we took data from the furthest time point from the end of the intervention for each group.

Search methods for identification of studies

Electronic searches

We searched the following databases in May 2012. We conducted updated searches in 2015 ‐ beginning 6 May 2015 and ending 15 May 2015 ‐ and a third updated search, which commenced 10 November 2016 and ended 14 November 2016.

We did not apply any date or language restrictions to our searches. We did not exclude studies on the basis of their publication status. We included abstracts, conference proceedings, and other 'grey literature' if they met the inclusion criteria.

Several of the databases and most of the websites that we searched in May 2012 yielded no or very few studies eligible for inclusion. The few eligible studies identified via these databases or websites were also available through searches of Cochrane CENTRAL, Embase, MEDLINE, and PsycINFO. We therefore chose to exclude the following from our updated searches in 2015 and 2016: Bibliomap, Dissertation Express, Clinicaltrials.gov, DoPHER, and EThOS.

The search strategies that we used to search databases can be found in Appendix 1.

Searching other resources

We carried out handsearches of reference lists of relevant articles to identify additional relevant studies. We contacted experts in the field to identify ongoing research. We carried out citation searches for key studies identified. We also searched the following websites of organisations actively involved in prevention of risk behaviours.

  • World Health Organization.

  • UNICEF; United Nations.

  • World Bank.

  • Centers for Disease Control and Prevention.

  • National Institutes of Health.

  • National Youth Agency.

  • Foundations: Joseph Rowntree, Nuffield Trust.

  • National Criminal Justice Reference Service.

  • Policy organisations ‐ Evidence for Policy and Practice Information and Co‐ordinating Centre (EPPI Centre), National Institute for Health and Clinical Excellence (NICE), Scottish Intercollegiate Guidelines Network (SIGN), Department of Health, University of York Centre for Reviews and Dissemination, The King's Fund, Institute for Public Policy Research.

Data collection and analysis

Selection of studies

Two review authors independently carried out the initial screening process for the first 500 publications retrieved to ensure quality and accuracy of the process. We selected a further 10% of studies at random and double‐screened them to ensure that the screening process was consistent and accurate throughout. In May 2012 and May 2015, we conducted this process, which yielded an overall Kappa statistic of 0.83, reflecting high agreement between study authors. We obtained full‐text articles if we required additional information to assess eligibility for inclusion.

We obtained the full texts of eligible articles and, when necessary, grouped together multiple publications arising from a single study. Two review authors screened full‐text papers using a prespecified set of criteria for inclusion. We resolved disagreements by discussion; when disagreements persisted, we consulted a third review author to enable a consensus to be reached.

Data extraction and management

Two review authors independently used a data extraction form created for this review to extract data from included studies. Two review authors had piloted the data extraction form to ensure that it captured study data and could be used to assess study quality effectively. Data extracted from full text studies included the following.

  • Lead author, review title, or unique identifier and date.

  • Study design.

  • Study location.

  • Study setting.

  • Year of study.

  • Theoretical underpinning.

  • Context.

  • Equity (using PROGRESS Plus (see below for details)).

  • Interventions (content and activities, numbers/types of behaviours addressed, duration of interventions, and details of any intervention offered to the control group).

  • Participants in the intervention (including number randomised and number included in each intervention group; age at the start of the intervention; and demographic data when possible (e.g. ethnicity, gender, socioeconomic status).

  • Scope of the interventions (universal or targeted to high‐risk or vulnerable groups).

  • Methods of measurement of risk behaviour (self‐report or objective measure).

  • Duration of follow‐up(s).

  • Attrition rate.

  • Randomisation.

  • Allocation concealment.

  • Outcome measures post intervention at each stage of follow‐up (including unit of measurement).

  • Effect size and precision (e.g. 95% confidence interval).

  • Whether clustering was taken into account in cluster RCTs and intracluster correlation coefficient (ICC).

  • Methods of analysis.

  • Process evaluation (including fidelity, acceptability, reach, intensity, and context of interventions).

  • Cost‐effectiveness data when provided (e.g. estimates of resource use, source of resources used, estimates and sources of unit costs, price year, currency, incremental resource use and costs, point estimate and measure of uncertainty for incremental resource use, costs and cost‐effectiveness, economic analytical viewpoint, time horizon for costs and effects, and discount rate).

  • Any other comments.

We used the PROGRESS Plus checklist to collect data relevant to equity. This includes place of residence, race/ethnicity, occupation, gender, religion, education, social capital, and socioeconomic status, with Plus representing the additional categories of age, disability, and sexual orientation. We collected PROGRESS Plus factors reported at baseline and follow‐up when reported. We resolved disagreements between review authors around data extraction by discussion, or by consultation with a third review author when consensus was not reached by discussion alone. We contacted study authors to obtain additional information or data not available from published study reports, when necessary.

Assessment of risk of bias in included studies

We assessed risk of bias of included studies using the Cochrane 'Risk of bias' tool (Higgins 2008). For each domain, two review authors rated studies as having 'high', 'low', or 'unclear' risk of bias. We resolved disagreements by discussion and, when necessary, by referral to a third review author. Selection bias included assessment of both adequate sequence generation and allocation concealment. We assessed studies as having low risk of selection bias when study authors reported a clearly specified method of generating a random sequence; and as having low risk of bias associated with lack of allocation concealment when study authors clearly described methods of concealment, such as use of opaque envelopes. We assessed studies as having high risk of performance bias unless study authors explicitly stated that students were blinded to group allocation, although participants could rarely be blinded to the fact that they were participants in an intervention owing to the nature of the studies. When studies clearly stated that outcome assessors were blinded, we judged them as having low risk of bias. When outcomes were assessed by self‐report, we rated studies as having high risk of bias when students were unlikely to have been adequately blinded. To assess attrition bias, we considered rates of attrition both overall and between groups, and we assessed whether this was likely to be related to intervention outcomes. We assessed studies as having low risk of reporting bias when a published protocol or study design paper was available and all prespecified outcomes were presented in the report; or when all expected outcomes were reported. If we had additional concerns, such as baseline imbalance between groups, we noted this in the ‘other bias’ domain.

Measures of treatment effect

For dichotomous (binary) data, we used odds ratios (ORs) with 95% confidence intervals (CIs) to summarise results within each study. When ORs were not provided, we calculated ORs and their standard errors (SEs) using reported outcome data. When studies reported ORs that represented the opposite measure (e.g. wearing a condom vs not wearing a condom), we took the inverse of the value.

For continuous outcomes, we extracted or calculated mean differences (MDs) based on final value measurements, ensuring that baseline mean values were sufficiently comparable (i.e. both lay within the standard deviations (SDs) for intervention and control). When this was not the case for baseline mean values in each study arm, we excluded data from the meta‐analysis and included them in a table. We calculated a pooled standard deviation from intervention and control SDs at follow‐up and standardised results to a uniform scale by calculating standardised mean differences (SMDs).

When studies reported an outcome as dichotomous and others provided a continuous measure, we converted results to dichotomous data, assuming that the underlying continuous measurement had an approximate logistical distribution, using the methods described in Borenstein 2009 (see Chapter 7). We conducted sensitivity analyses to assess the impact of this on study findings.

Unit of analysis issues

Several interventions that were randomised at the school level did not appear to take clustering of participants into account, for instance, by using a multi‐level model or generalised estimating equations. When clustering was not taken into account, and when study authors could not provide adjusted data, we followed the approach suggested in Chapter 16.3.5 of the Cochrane Handbook for Systematic Reviews of Interventions, to conduct an 'approximately correct analysis' (Bush 1989; Fearnow‐Kenney 2003; Griffin 2009; Ialongo 1999; Ialongo 1999b; Kellam 2014; Li 2011; Lochman 2003a; Lochman 2004a; McNeal 2004; Nader 1999; O'Neill 2011; Sanchez 2007; Shek 2011). We imputed intracluster correlation coefficients (ICCs) for each outcome, which provide a measure of the relative variability within and between clusters, from other included studies that reported an ICC for the same outcome, to enable the design effect to be calculated. For all analyses, we selected the most conservative ICC for that behaviour. When no ICC was available for that behaviour, we used the largest available ICC for any behaviour to be conservative. We conducted sensitivity analyses, which utilised the lowest reported ICC for the same behaviour. When no ICC was reported, we calculated an average of available ICCs and used this value. A list of the ICCs used in the data analyses is provided in Additional Table 1.

Open in table viewer
Table 1. Intracluster correlation coefficients

Study

Country

Age

Outcome variable

Reported intracluster correlation coefficient

Published or correspondence (comment)

ICCs used in primary analyses

Gatehouse Study (Bond 2004)

Australia

13‐14

Substance use

0.06

Published

All Stars 2 (Gottfredson 2010)

USA

11‐14

Aggression

0.025

Published

Fourth R (Wolfe 2012)

USA

14‐15

Violence

0.01

Published

All Stars 2 (Gottfredson 2010)

USA

11‐14

Delinquency

0.025

Published

Fourth R (Wolfe 2012)

USA

14‐15

Sexual risk behaviour

0.01

Published

Gatehouse Study (Bond 2004)

Australia

13‐14

Diet/physical activity

0.06

Publisheda

Positive Action (Chicago) (Li 2011)

USA

8‐13

Education

0.1

Published

Gatehouse Study (Bond 2004)

Australia

13‐14

Mental illness

0.01

Published

ICCs used in sensitivity analyses

LIFT/All Stars 2 (DeGarmo 2009; Gottfredson 2010)

USA

10/11‐14

Substance use

0.0

Published

All Stars 2 (Gottfredson 2010)

USA

11‐14

Aggression

0.0

Published

Fourth R (Wolfe 2012)

USA

14‐15

Violence

0.01

Published

All Stars 2 (Gottfredson 2010)

USA

11‐14

Delinquency

0.0

Published

Fourth R (Wolfe 2012)

USA

14‐15

Sexual risk behaviour

0.01

Published

All Stars 2, Gatehouse Study, Fourth R,

LIFT, Positive Action (Chicago) (Bond 2004; Gottfredson 2010; Wolfe 2012; DeGarmo 2009; Li 2011)

USA,

Australia

10‐15

Diet/physical activity

0.0263

Publishedb

Gatehouse Study (Bond 2004)

Australia

13‐14

Education

0.01

Publishedc

Gatehouse Study (Bond 2004)

Australia

13‐14

Mental illness

0.01

Published

ICC: intracluster correlation coefficient.

aThe highest ICC value was used to be conservative.

bAverage ICC value used from across these studies.

cICC related to school engagement.

A very small number of trials did not report the number of participants in each study arm. If it was reported that attrition was comparable between study arms, we divided the total N by two to yield an approximate number for each arm. When we found interventions with multiple study arms, we split the control group to avoid double‐counting, as outlined in Section 16.5.4 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008).

Dealing with missing data

When we encountered missing or unclear data related to participants or outcomes, we contacted study authors via email. We noted missing data on the data extraction form and took them into account when judging the risk of bias of each study. We excluded from quantitative analyses studies for which insufficient data were available (e.g. in study reports, and when missing data could not be obtained) and included data from study reports in Additional Table 2.

Open in table viewer
Table 2. Outcomes not included in meta‐analysis

Author and year

Study name

Categorisation

Outcome

Authors' conclusions

1. Tobacco use

Bonds 2010

New Beginnings

Family‐Targeted

Tobacco use disorder (including nicotine withdrawal and dependence)

No difference between study arms in the proportion of participants meeting criteria for nicotine use disorder (6.7% in each arm)

Bush 1989

Know Your Body

School‐Universal

Serum thiocyanate (micromoles/L)

Mean difference from baseline to 1 year follow‐up was ‐9.87 (SE 2.5) in the intervention group, and 20.03 (SE 2.68) in the control group (P < 0.001). These data were based on a 50% subsample stratified at baseline, based on measurement after 1 year of intervention.

Connell 2007

Family Check‐Up

Family‐Universal

Nicotine abuse/dependence

Across treatment and control groups, no significant differences were found for nicotine abuse/dependence (Chi² (1, 998) = 3.09, P > 0.05). No significant correlation between assignment to experimental condition(s) and tobacco use over time

DeGarmo 2009

LIFT

School‐Universal

Initiation of tobacco use

With controls for parental drinking and deviant peer association, the intervention was associated with reduced risk of initiation of tobacco use (beta = ‐0.10, P < 0.01). The effect translated to odds ratios of a 10% reduction in risk for tobacco use.

Estrada 2015

Brief Familias Unidas

Family‐Targeted

Tobacco use in past 90 days

Brief Familias Unidas was not significantly efficacious in reducing tobacco use (beta = ‐0.09, P = 0.85) in the past 90 days.

Gonzales 2012

Bridges to High School

Family‐Targeted

Substance use

Study authors report that substance use at follow‐up was less in the intervention group than in the control group for adolescents who engaged in high levels (85th percentile) of baseline substance use (d = 3.65).

LoSciuto 1999

Woodrock Youth Development Project

School‐Universal

Substance use in past month (tobacco, alcohol, drugs)

Mean substance use in the past month was 1.1 for the intervention group and 1.15 for the control group (SMD 0.18)

McNeal 2004

All Stars

School‐Universal

Tobacco use in past 30 days

The teacher‐delivered All Stars programme was associated with reduced rate of growth in 30‐day usage of cigarettes (7.4% to 7.8%) compared to the specialist condition (11.0% to 13.8%) and the control group (15.1% to 17.9%).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Mean cigarettes per day

15‐year follow‐up: incidence of cigarettes smoked per day in past 6 months among those who received nurse visitation through pregnancy (group 3) was 0.91 compared to 1.30 among control participants (P = 0.49). Among a subgroup of women from low socioeconomic status (SES) households who were unmarried, the comparison was 1.32 vs 2.50 among control participants (P = 0.07). Incidence of cigarettes smoked per day in the past 6 months among those who received nurse visitation until the child's second birthday was 1.28 compared to 1.30 among control participants (P = 0.76). Subgroup analysis of women from low SES households who were unmarried showed that incidence was 1.50 among the intervention group compared to 2.50 among controls (P = 0.1).

Perry 2003

DARE and DARE‐Plus

School‐Universal

Current smoker (growth rate)

Growth curve analysis showed that for boys: the growth rate of tobacco use was 0.31 (0.05) in the control group, 0.28 (0.05) in the DARE group, and 0.18 (0.05) in the DARE Plus group (DARE vs control P = 0.28; DARE Plus vs control P = 0.02; DARE Plus vs DARE P = 0.08). Among girls: the growth rate was 0.28 (0.07) in the control group, 0.25 (0.07) in the DARE group, and 0.22 (0.07) in the DARE Plus group (DARE vs control P = 0.38; DARE Plus vs control P = 0.25; DARE plus vs DARE P = 0.35).

Piper 2000

Healthy for Life

School‐Universal

Tobacco use in past 30 days

The age‐appropriate condition showed no benefit over the control condition at 12‐month follow‐up (prevalence 24% in both arms; HLM coefficient 0.18, SE 0.12, P > 0.1) or at 24‐month follow‐up, where prevalence was higher in the intervention group (prevalence 36% vs 30% in the control group, coefficient 0.41, SE 0.2, P < 0.1). Among those receiving the intensive condition, prevalence was similar in both study arms (12 months: 22% vs 24% in the control group; coefficient ‐0.3, SE 0.17, P > 0.1; 24 months: 28% vs 30% in the control arm; coefficient ‐0.38, SE 0.15, P < 0.05).

Saraf 2015

(none given)

School‐Universal

Tobacco use

Current smoking (in the past month) changed from 13.1% (95% CI 10.2% to 15.9%) to 3.1% (95% CI 0.2% to 5.9%) in the intervention group; and from 7.7% (95% CI 5.0% to 10.4%) to 5.4% (95% CI 2.6% to 8.2%) in the control group (overall difference between groups in pre‐ to post‐change ‐7.7 (‐10.7 to ‐4.7); P < 0.01.

Schweinhart 1980

High/Scope Perry Preschool Study

School‐Targeted

Tobacco use

No impact of the intervention on smoking cigarettes 22 years after the end of the programme: 45% of those in the intervention group smoked compared to 56% of those in the control group (P = 0.231). Effect size 0.22

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Likelihood of smoking

Those receiving the intervention were reported to be 19.7% less likely to start smoking compared to controls (males receiving Big Brothers Big Sisters were 24.5% less likely to start smoking, and females 9.9%). Males from an ethnic minority receiving Big Brothers Big Sisters had a 29.9% increased likelihood of smoking compared to controls, but among females there was a 1.9% reduction. White males and females receiving the intervention had a 47.9% and 14.7% reduced likelihood of smoking, respectively.

Walter 1989

Know Your Body

School‐Universal

Smoking

Among the schools in Westchester, results showed a beneficial impact of the intervention: the school mean at the end of the intervention was 3.5% (SD 4.3%) compared to 13.1% (SD 5.2) among control schools; P < 0.005. This is equivalent to a 73% reduction in the rate of initiation of smoking.

2. Alcohol use

Bonds 2010

New Beginnings

Family‐Targeted

Alcohol use, binge drinking, age commencing drinking

15‐year follow‐up: alcohol use in the past month higher in the intervention arm than in the control arm (d = 0.23, 95% CI ‐0.26 to 0.72). Intervention arm commenced drinking at a mean age 0.47 years younger than the control group (95% CI ‐1.31 to 0.23 years). Binge drinking in the past year higher in the intervention group than in the control arm (d = 0.16, 95% CI ‐0.14 to 0.46).

Conduct Problems Prevention Research Group 2010

Fast Track

School‐Targeted

Binge drinking problem

The intervention marginally decreased binge drinking at 10‐year follow‐up (adjusted OR 0.75, 95% CI 0.55 to 1.01, P = 0.057).

Connell 2007

Family Check‐Up

Family‐Universal

Alcohol use

No significant association was noted between assignment to experimental condition(s) and alcohol abuse/dependence over time (Chi² (1, 998) = 0.98, P > 0.05), with the exception of Time 2, when a correlation between treatment assignment and alcohol use was observed (r = 0.09, P ≤ 0.05).

Cunningham 2012

SafERteens

Individual‐Targeted

Alcohol use

Reduction in the proportion of participants scoring ≥ 3 on AUDIT‐C from 50% at baseline to 34.4% at 3 months and 37.3% at 12 months (‐12.7% change at 12 months; OR 1.09, 95% CI 0.77 to 1.56) for those in the therapist intervention arm; and a reduction from 45.6% at baseline to 32.7% at 3 months and 28.9% at 12 months (‐16.7% change at 12 months; OR 0.95, 95% CI 0.66 to 1.37) for those in the computer arm . For controls, a reduction from 47.7% to 38.1% at 3 months and 34.7% at 12 months was evident (‐13% change at 12 months).

Cunningham 2012

SafERteens

Individual‐Targeted

Binge drinking

Reduction in the proportion of participants reporting any binge drinking from 52.8% at baseline to 34.4% at 3 months and 38.7% at 12 months (‐14.1% reduction at 12 months; OR 0.95, 95% CI 0.66 to 1.36) among those in the therapist group; and a reduction from 48.5% to 28.8% at 3 months and 30.3% at 12 months (‐18.2% reduction; OR 0.83, 95% CI 0.58 to 1.19) among those in the computer group. Similar reductions were seen in the control group: a reduction from 54% at baseline to 34.6% at 3 months and 36.1% at 12 months (‐17.9% reduction at 12 months).

Estrada 2015

Familias Unidas – Brief

Family‐Targeted

Alcohol use

Brief Familias Unidas was not significantly efficacious in reducing alcohol use (beta = 0.17; P = 0.51) in the past 90 days.

Friedman 2002

Botvin Life Skills Training and Anti‐violence

Individual‐Targeted

Degree of alcohol use

Alcohol use was decreased among intervention participants compared to controls (t = ‐1.24, P > 0.05).

Gonzales 2012

Bridges to High School

Family‐Targeted

Substance use

Study authors report that substance use was less at follow‐up in the intervention group compared to the control group for adolescents who engaged in high levels (85th percentile) of baseline substance use (d = 3.65).

Jalling 2016

Comet 12‐18

Family‐Targeted

Alcohol use (AUDIT score)

No significant difference was found between groups: at T2, mean AUDIT score was 7.59 (SD 7.60) in the intervention group vs 6.26 (SD 6.79) in the control group.

Jalling 2016b

ParentSteps

Family‐Targeted

Alcohol use (AUDIT score)

No significant difference was found between groups: at T2, mean AUDIT score was 5.10 (SD 6.38) in the intervention group vs 6.26 (SD 6.79) in the control group.

Kellam 2008

Good Behaviour Game

School‐Universal

Lifetime alcohol abuse/ dependence

The Good Behaviour Game (GBG) was associated with a reduction in lifetime alcohol abuse/dependence disorders compared to control: 13% for GBG vs 20% for controls (P = 0.08). The effect was similar for males and females.

Murry 2014

SAAF

Family‐Targeted

Escalation of alcohol use

Study authors report through structural equation modelling analysis that youth avoidance of risk opportunity situations served a role in delaying initiation and escalation of use of alcohol and other substances as they transitioned from early to late adolescence.

Monti 1999

Alcohol Screening and Brief Intervention

Individual‐Targeted

Alcohol use score

With a 2 × 2 (group × time) repeated measures analysis of variance, time effect showed reductions in alcohol scores (F(1,79) = 24.55, P < 0.001) with no group differences or interactions.

Olds 1998

Nurse Family Partnership

Family‐Targeted

Alcohol use

15‐year follow‐up: incidence of days drunk alcohol in past 6 months among those who received nurse visitation through pregnancy (group 3) was 1.81 compared to 1.57 among control participants (P = 0.97). Among a subgroup of women from low socioeconomic status (SES) households who were unmarried, the comparison was 1.84 vs 2.49 among control participants (P = 0.41). Incidence of days drunk alcohol in past 6 months among those who received nurse visitation until the child's second birthday was 1.87 compared to 1.57 among control participants (P = 0.96). Subgroup analysis of women from low SES households who were unmarried show the incidence was 1.09 among the intervention group compared to 2.49 among controls (P = 0.03).

Perry 2003

DARE vs DARE Plus

School‐Universal

Alcohol consumption in past month

Growth curve analysis showed that for boys: the growth rate in alcohol use in the past month (mean, SE) was 0.14 (0.02) for those in the control group, 0.11 (0.02) for the DARE group (P = 0.12), and 0.08 (0.02) for the DARE Plus group (P = 0.01) (DARE Plus vs DARE, P = 0.12). Among girls: values were 0.12 (0.03) for controls, 0.13 (0.02) for those in the DARE group (P = 0.40), and 0.08 (0.03) for those in the DARE Plus group (P = 0.15) (DARE Plus vs DARE, P = 0.10).

Piper 2000

Healthy for Life

School‐Universal

Alcohol use in past 30 days

Results showed a negative treatment effect at 12 months and 24 months of follow‐up: in the age‐appropriate intervention, prevalence of alcohol use in the past month was 33% in the intervention group and 28% in the control group at 12 months (hierarchical linear modelling (HLM) coefficient 0.34, SE 0.19, P < 0.1). At 24 months, the prevalence of alcohol use in the past month was 48% in the intervention group and 41% in the control group at 24 months (HLM coefficient 0.3, SE 0.14, P < 0.05). In the intensive version of the intervention, the prevalence of alcohol use at 12 months was 33% vs 28% in the control arm (HLM coefficient 0.2, SE 0.09, P < 0.05), and at 24 months, prevalence was 45% vs 41% in the control arm (HLM coefficient 0.27, SE 0.1, P < 0.05).

Schweinhart 1980

High/Scope Perry Preschool Study

School‐Targeted

Alcohol use

No impact of the intervention on drinking alcoholic beverages several or more times a week 22 years after the end of the programme: 16% of those in the intervention group drank alcohol several or more times a week compared to 26% of those in the control group. Effect size for drinking alcoholic beverages was 0.27 (P = 0.141).

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Likelihood of initiating alcohol use

Those receiving the intervention were 27.4% less likely to start using alcohol than those in the control group (19.2% reduction in likelihood among males and 38.8% among females). The reduction in likelihood was 11.4% among males from an ethnic minority, 53.7% among females from an ethnic minority; 34.5% among white males, and 8.4% among white females.

3. Illicit drug use

Connell 2007

Family Check‐Up

Family‐Universal

Marijuana use

Across treatment and control groups, no significant differences were found for marijuana abuse/dependence (Chi² (1, 998) = 0.74, P > 0.05). No significant correlation was noted between assignment to experimental condition(s) and marijuana use over time, with the exception of Time 2 (r = 0.10, P ≤ 0.05).

Bonds 2010

New Beginnings

Family‐Targeted

Marijuana use, polydrug use, other drug use

6‐year follow‐up: results showed no significant group effects for drug dependence, drug symptom count, or polydrug use (all P > 0.05).

15‐year follow‐up: intervention group displayed lower past year polydrug use (d = ‐.44, 95% CI ‐.88 to .00) and past year other drug use (d = ‐.06, 95% CI ‐.11 to ‐.00) compared to control group. No difference was observed for marijuana use between intervention and control groups (d = .00, 95% CI ‐.47 to .47).

DeGarmo 2009

LIFT

School‐Universal

Percentage of participants who have not used marijuana

One year post intervention, 2.2% had not used marijuana in the past year compared to 2.3% in the control group.

Estrada 2015

Brief Familias Unidas

Family‐Targeted

Illicit drug use (past 90 days)

Brief Familias Unidas was not significantly efficacious in reducing illicit drug use (beta = 0.03; P = 0.93) in the past 90 days.

Friedman 2002

Botvin Life Skills Training and Anti‐violence

Individual‐Targeted

Degree of drug use and involvement in selling of drugs

Among intervention participants compared to controls, data showed a greater reduction in drug use (t = ‐2.58, P < 0.01) and a greater reduction in the frequency of involvement in the selling of drugs (t = ‐1.99).

Freudenberg 2010

REAL MEN

Individual‐Targeted

Daily marijuana use in past 90 days

Intervention associated with reduced odds of daily marijuana use (OR 0.751). No 95% confidence interval or standard error was provided.

Freudenberg 2010

REAL MEN

Individual‐Targeted

Hard drug use tried in past 90 days

Intervention was associated with reduced odds of trying hard drugs (OR 0.166, P < 0.05). No 95% confidence interval or standard error was provided.

Griffin 2006

Life Skills Training

School‐Universal

High‐risk substance use

32.4% of participants in the intervention group engaged in high‐risk substance use at the young adult follow‐up compared to 37.1% of those in the control group 11 years following completion of the intervention.

Jalling 2016

Comet 12‐18 and Parent Steps

Individual‐Targeted

Any illicit drug use (%)

Higher odds of illicit drug use were evident among those whose parents took part in the study, although 95% CIs were wide. Comet 12‐18: OR 3.52, 95% CI 1.23 to 10.10. ParentSteps OR 3.23, 95% CI 1.06 to 9.08

McNeal 2004

All Stars

School‐Universal

Marijuana use in past 30 days

Marijuana use in the past 30 days for those in the specialist arm increased from 3.2% to 4.1% in the intervention group and from 5.0% to 8.7% in the control group (standardised B coefficient = 0.02, P > 0.05). For those in the teacher‐delivered arm, the increase was 3.2% at baseline and follow‐up compared to a change from 5.0% to 8.7% in the control group (standardised B coefficient ‐0.01, P > 0.05).

Piper 2000

Healthy for Life

School‐Universal

Marijuana use in past 30 days

In the age‐appropriate condition, prevalence of marijuana use was similar in the intervention and control groups at 12‐month and 24‐month follow‐up (prevalence 4% vs 5% in the control group; OR 0.77, P > 0.1; and 12% vs 10% in the control group; OR 1.28, P > 0.1, respectively). Among those receiving the intensive version of the programme, findings suggested benefit of the intervention: prevalence 5% in both arms at 12 months (OR 0.56, P < 0.05) and prevalence 8% vs 10% in the control condition (OR 0.56, P < 0.05).

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Likelihood of initiating drug abuse

Overall, receiving the intervention was associated with a 45.8% reduction in the likelihood of initiating drug abuse (55% among males and 26.6% among females). The impact was greatest among males and females from an ethnic minority, among whom results showed a 67.8% and 72.6% reduced likelihood of initiating drug use, respectively. White males in the intervention group were 32.7% less likely to start using drugs compared to white males in the control group, but white females were 49.5% more likely to start using drugs compared to white females in the control group.

4. Substance misuse (composite)

Beach 2016

ProSAAF

Family‐Targeted

Substance use in lifetime (self‐reported use of cigarettes, alcohol, and/or marijuana)

At 9 months, young people in the intervention group reported lower levels of substance use initiation compared to those in the control group (coefficient ‐2.25, SE 0.64, t = 3.54, P < 0.01).

Berry 2009

Coaching for Communities

Individual‐Community

Alcohol and drug use

At the end of the intervention, the mean use of alcohol and drugs in the past 30 days was 0.83 in the intervention group and 2.55 in the control group.

Estrada 2015

Brief Familias Unidas

Family‐Targeted

Substance use (alcohol, tobacco, and/or drugs)

Growth curve analyses showed a non‐significant difference in past 90‐day substance use between brief Familias Unidas and CPC (beta = 0.24; P = 0.37).

Freudenberg 2010

REAL MEN

Individual‐Targeted

Alcohol or drug dependence in the past year

Reduced odds of alcohol or drug dependence in the past year following receipt of intervention (OR 0.519, P < 0.05). No 95% confidence interval or standard error was provided.

Gonzales 2014

Bridges to High School (Bridges/ Puentes)

Family‐Targeted

Substance use

Intervention status was associated with a reduction in substance use at 2 years and 5 years post‐test (unstandardised regression coefficients ‐0.3 and ‐0.13, respectively).

Griffin 2006

Life Skills Training

School‐Universal

High‐risk substance use

32.4% of participants in the intervention group engaged in high‐risk substance use at the young adult follow‐up compared to 37.1% of those in the control group.

LoSciuto 1999

Woodrock Youth Development Project

School‐Universal

Substance use in past month

Participation in the programme was associated with higher average scores for lifetime substance use (F(1,711) = 6.10, P = 0.01, Cohen's d = 0.19) and past month substance use (F(1,712) = 5.93, P = 0.02, Cohen's d = 0.18). The data could not be adjusted for clustering owing to insufficient information reported.

Olds 1998

Nurse Family Partnership

Family‐Targeted

Drug use

At 15‐year follow‐up, data showed no significant difference in the incidence of days of drug use in the past 6 months between intervention and control groups. Among those who received nurse visitation during pregnancy, incidence was 3.55 vs 2.28 among controls (P = 0.49) (low SES, unmarried subgroup: 9.38 vs 4.04, P = 0.01). Among those who received nurse visitation until the child's second birthday, incidence was 2.04 vs 2.28 in the control group (P = 0.54) (low SES, unmarried subgroup: 2.5 vs 4.04 among controls, P = 0.24).

5. Antisocial behaviour and offending

Averdijk 2016

Triple P

Family‐Targeted

Delinquency

No substantial effect of the intervention was found at long‐term follow‐up (age 15 years, beta = 0.004, 95% CI ‐0.15 to 0.15; ES = 0.002).

Averdijk 2016

PATHS

School‐Universal

Delinquency

No substantial effect of the intervention was found at long‐term follow‐up (age 15 years, beta = ‐0.04, 95% CI ‐0.19 to 0.11; ES = ‐0.022).

Beach 2016

ProSAAF

Family‐Targeted

Conduct problems

Follow‐up revealed a beneficial effect of the intervention on conduct problems: coefficient for conduct problems ‐0.54, SE 0.22, t = 2.42, P = 0.05.

Berry 2009

Coaching for Communities

Individual‐Targeted

Variety and volume of offending

For variety of offending, the mean in the intervention group was 3.5 vs 5.95 in the control group at the end of intervention; and for volume of offending, the mean in the intervention group was 18.1 vs 23.9 in the control group.

Conduct Disorders Prevention Research Group 2010

Fast Track

School‐Targeted

Antisocial personality disorder (ASPD)

10 years post intervention, the prevalence of being in the DSM‐IV clinical range for ASPD was lower in the intervention group than in the control group (OR 0.60, 95% CI 0.39 to 0.93, P = 0.022).

Connell 2007

Family Check‐Up

Family‐Universal

Antisocial Behaviour

Across treatment and control groups, no significant differences were found for marijuana abuse/dependence (Chi² (1, 781) = 0.69, P > 0.05). No significant correlation between assignment to experimental condition(s) and antisocial behaviour over time

Cunningham 2012

SafERteens

Individual‐Targeted

Any peer aggression

A reduction of 34.3% in the proportion reporting any severe peer aggression at 3 months (from 82.7%), increasing to a 43.3% reduction at 12 months (OR 1.36, 95% CI 0.87 to 2.12) for the therapist group. For the computer group, a reduction of 21.3% was evident at 3 months, and 26.2% at 12 months (OR 0.88, 95% CI 0.57 to 1.34). For controls, a 16.4% reduction was evident at 3 months, increasing to 25.9% at 12 months.

Cunningham 2012

SafERteens

Individual‐Targeted

Any peer victimisation or peer violence

Reduction of 10.4% at 3 months and 22.7% at 12 months for those in the therapist group (baseline 47.6%) (OR 1.25, 95% CI 0.87 to 1.79); and reduction of 2.5% at 3 months and 17.4% at 12 months for the computer group (OR 1.06, 95% CI 0.73 to 1.52). Among those in the control group, results showed a 4.7% increase at 3 months but a 12.3% reduction in reported experience of peer violence at 12 months.

DeGarmo 2009

LIFT

School‐Universal

Percentage arrested or detained

At initial follow‐up, 300 days post intervention, 0.6% of those in the intervention group had been detained or arrested vs 4.1% in the control group. 2.5 years post intervention (900 days), 5.1% had been arrested/detained in the intervention group vs 10.3% in the control group.

Friedman 2002

Botvin Life Skills Training and Anti‐violence

Individual‐Targeted

Degree of illegal offences

Among intervention participants vs controls, there was a slight reduction in the degree of illegal offences (t = ‐1.53).

Freudenberg 2010

REAL MEN

Individual‐Targeted

Criminal justice outcomes (re‐arrest, re‐incarceration, problematic behaviour)

Intervention associated with reduced odds of re‐arrest (OR 0.871). No 95% confidence interval or standard error was provided. Odds of re‐incarceration 1.019; the intervention was associated with reduced odds of often engaging in problematic behaviour (OR 0.789)

Gonzales 2014

Bridges to High School (Bridges/ Puentes)

Family‐Targeted

Externalising symptoms

Intervention associated with small reduction in externalising symptoms at 2 and 5 years post‐test (unstandardised regression coefficients ‐0.02 and ‐0.01, respectively)

Kellam 2008

Good Behaviour Game (GBG)

School‐Universal

Lifetime antisocial personality disorder (ASPD)

At ˜ 12 years' follow‐up (participants were aged 19 to 21 years), overall rates of ASPD were lower for those in the GBG groups (17%) vs internal controls (25%) (P = 0.07).

LoSciuto 1999

Woodrock Youth Development Project

School‐Universal

Aggression

No strong evidence showed a greater reduction in aggression in experimental vs control groups at post‐test (F(1, 342) = 2.95, P = 0.09, Cohen's d = 0.19). Insufficient data were available to adjust these findings for clustering of participants by classroom.

Olds 1998

Nurse Family Partnership

Family‐Targeted

Major delinquent acts

At 15‐year follow‐up, results showed no difference between intervention and control groups in the mean number of major delinquent acts committed: mean 2.79 among those who received nurse visitation through pregnancy vs 3.02 in the control group (P = 0.93). Among a subgroup of women from low socioeconomic status (SES) households who were unmarried, the comparison was 3.45 vs 4.09 (P = 0.60).

Among those receiving nurse visitation through to the child's second birthday, the comparison was 3.57 vs 3.02 (P = 0.48). Among a subgroup of women from low SES households who were unmarried, the comparison was 3.99 vs 4.09 (P = 0.77).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Mean number of arrests

Differences between groups were evident regarding the incidence of arrests at 15‐year follow‐up. For those visited during pregnancy, the incidence of arrests among children was 0.16 vs 0.36 (P = 0.005); among a subgroup of women from low SES households who were unmarried, the comparison was 0.15 vs 0.45 (P = 0.02).

Among those visited through pregnancy and infancy, their children were arrested a mean of 0.17 times vs 0.36 times among controls (P = 0.005); and among a subgroup of women from low SES households who were unmarried, the comparison was 0.20 vs 0.45 (P = 0.03).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Externalising problems

At 15‐year follow‐up, results showed no difference between intervention and control groups in the mean number of externalising problems: mean 13.65 among those who received nurse visitation through pregnancy vs 13.73 in the control group (P = 0.95). Among a subgroup of women from low socioeconomic status (SES) households who were unmarried, the comparison was 15.63 vs 14.18 (P = 0.42).

Among those receiving nurse visitation through to the child's second birthday, the comparison was 13.88 vs 13.73 (P = 0.89) Among a subgroup of women from low SES households who were unmarried, the comparison was 11.85 vs 14.18 (P = 0.17).

Perry 2003

DARE vs DARE Plus

School‐Universal

Physical victimisation

Among boys, those in DARE‐Plus schools were less likely than those in control schools to show increases in victimisation (growth rate ‐0.1, SE 0.04, P = 0.02); there was no difference between DARE and control (growth curve rate, mean ‐0.03, SE 0.04, P = 0.18). No differences were evident between groups among girls.

Schweinhart 1980

High/Scope Perry Preschool Program

School‐Targeted

Carried a gun or knife once or more often

At 10‐year follow‐up (when participants were ˜ age 15), 13 of 44 (29.5%) in the intervention group had carried a gun or knife once or more compared to 15 of 55 (27%) in the control group.

Shetgiri 2011

[No study name]

School‐Targeted

Been in trouble with the police in the past 12 months

Eighteen per cent of those in the intervention group had been in trouble with the police in the past 12 months at follow‐up post intervention (21% at baseline) compared to 26% of those in the control group at follow‐up (32% at baseline) (P = 0.41).

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Hitting, stealing, and damaging property

Participation in the intervention was associated with a 32% reduction in the number of times participants hit someone compared to control. The percentage reduction was greater in females than in males (43% vs 25%). Results showed a greater reduction among females from an ethnic minority than among white females (48% reduction vs 2% reduction), and a greater reduction was observed in white males (45%) than in males from an ethnic minority (4%). Data show a 19% reduction in the number of times participants in the intervention group vs the control group stole something and little change (0.15% reduction) in the number of times participants damaged property. Findings show a 16% reduction in the number of times participants in the intervention group took something from a store compared to controls, and a 17% reduction in the number of times participants did risky things. Little change was evident in relation to behavioural conduct (1% reduction in intervention vs control) and the number of times participants were involved in a fight (1% reduction in intervention vs control).

6. Vehicle‐related risk behaviour

Schweinhart 1980

High/Scope Perry Preschool Study

School‐Targeted

Wearing seatbelt

Among those in the intervention group, 24 of 56 (43%) wore a seatbelt sometimes or never 22 years after the end of the programme compared to 40/61 (66%) of those in the control group. Effect size for wearing a seatbelt was 0.37 (P = 0.052).

D'Amico 2002

Risk Skills Training vs DARE

School‐Universal

Driving under the influence/riding with a drinking driver

No differences were observed at 6‐month follow‐up in relation to driving after drinking and riding with a drinking driver (mean values for baseline and 6‐month follow‐up: risk skills training programme group: mean 1.25 (SD 3.30) to 0.95 (SD 2.20); DARE‐A group: mean 0.75 (SD 1.42) to 0.67 (SD 1.26); control group: mean 1.58 (SD 5.32) to 1.32 (4.42).

Nirenberg 2013

ROAD

Individual‐Targeted

Speeding and distracted driving

Scores for speeding and distracted driving were lower in the control group (community service) than in the combined motivational interviewing study arms 6 months post intervention (t(607) = ‐2.32; P = 0.02) (i.e. the control group reported less of the behaviour) (Log+1 transformed mean values: control 2.49 (SD 1.57); combined MI 2.81 (SD 1.53)). No difference between groups was evident in relation to dangerous driving factor scores (t(607) = ‐0.21, P = 0.84) (Log+1 transformed means: control 1.39 (SD 1.46); combined MI 1.34 (SD 1.39)) or scores for alcohol, drugs, and driving (Log+1 mean values: control 0.58 (SD 1.14); combined MI 0.60 (SD 1.14)).

7. Sexual risk behaviour

Bonds 2010

New Beginnings

Family‐Targeted

Number of sexual partners

Significant group effect for number of sexual partners (control mean = 1.65, intervention mean = 0.68, P = 0.01, d = 0.49)

Estrada 2015

Brief Familias Unidas

Family‐Targeted

Inconsistent condom use in past 90 days

Growth curve analyses showed no significant differences in unsafe sexual intercourse, defined as inconsistent condom use, during the past 90 days between brief Familias Unidas and CPC (beta = 0 .26, P = 0 .25).

Freudenberg 2010

REAL MEN

Individual‐Targeted

Engaged in risky sexual behaviour in past 90 days

No difference was observed between the intervention arm and the control arm in relation to the proportion of participants engaging in risky sexual behaviour in the past 90 days (OR 0.856, no 95% CI given, but P > 0.05).

Griffin 2006

Life Skills Training

School‐Universal

Multiple sexual partners

21.3% of those in the intervention group had multiple sex partners at age 24 years (˜ 11 years following the end of the intervention) vs 24.5% of those in the control group.

Griffin 2006

Life Skills Training

School‐Universal

Condom use

Results showed no difference across experimental conditions in relation to condom use at age 24 years (˜ 11 years following the end of the intervention): 78.7% of the intervention group reported that they did not always use condoms vs 78.2% of controls (Chi² (1) = 0.05, P < 0.82).

McBride Murry 2014

SAAF (Stronger African American Families)

Family‐Targeted

Sexual behaviour

The effect size of the intervention on post‐test sexual behaviour was 0.01, although study authors state that detecting a substantial effect size was unlikely with a sample of < 1000 and owing to the length of time between the programme and longer‐term (65‐month) follow‐up. Using structural equation modelling, study authors also report that participation in SAAF led to protection in engagement in sexual risk behaviour through an indirect mechanism involving increased intervention‐targeted parenting practices (beta = 0.35, P < 0.01), which were associated in turn with increased youth self‐pride (beta = 0.25, P < 0.05), in turn associated with increased protective sexual norms (beta = 0.7, P < 0.01), in turn associated with reduced sexual risk behaviour (beta = ‐0.22, P < 0.01).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Lifetime incidence of sex partners in past 6 months

At 15‐year follow‐up: among those visited during pregnancy, the mean number of sex partners was 1.10 vs 1.56 (P = 0.48); and among a subgroup of women from low SES households who were unmarried, the mean number of sex partners in the past 6 months was 2.23 vs 2.48 (P = 0.73). Among those visited during pregnancy and infancy, mean incidence of sex partners was 1.16 vs 1.56 (P = 0.90); and for the subgroup of women from low SES households who were unmarried, mean was 0.92 vs 2.48 (P = 0.003).

Piper 2000

Healthy for Life

School‐Universal

Sexual intercourse in past 30 days

Students were followed up in the ninth and 10th grades, at 12 and 24 months. Students in the age‐appropriate condition reported higher rates of intercourse than those in the control group (13% vs 11%; HLM coefficient 0.4, SE 0.16, P < 0.05) at 12 months; at 24 months, prevalence was 23% vs 19% (HLM coefficient 0.32, SE 0.2, P > 0.1). The intensive condition had no effect on rates of intercourse in the 2 groups at 12 months (prevalence 15% vs 11% in the control arm, HLM coefficient 0.25, SE 0.21, P > 0.1) nor at 24 months (prevalence 21% vs 19% in the control arm; HLM coefficient ‐0.07, SE 0.15, P > 0.1).

8. Physical inactivity

Bush 1989

Know Your Body

School‐Universal

Fitness score

Study authors highlight that significant changes were observed in a favourable direction in relation to fitness. The observed difference between intervention and control group mean change after 2 years of intervention was ‐0.28 (SE 0.19); and ‐0.38 (SE 0.15) after adjustment for baseline value, age, sex, and socioeconomic status.

O’Neill 2016

Michigan Model for Health

School‐Universal

Physical activity skills

Six weeks following the intervention, results showed a significant intervention effect for physical activity skills: F[53,590.79] = 4.42, P = 0.001.

Saraf 2015

(none given)

School‐Universal

Total time spent watching TV (minutes)

Weak evidence for a reduction in time spent watching television in the intervention group: reported reduction from 70.4% (95% CI 67.0% to 73.8%) at pre‐test to 56.1% (95% CI 53.9% to 58.4%) at post‐test (P < 0.05). In comparison, a slight increase in time spent watching TV was observed in the control group: 56.4% (95% CI 53.9% to 58.9%) at pre‐test increasing to 57.9% (95% CI 55.2% to 60.8%) at post‐test; overall difference 15.8 (95% CI 15.7 to 16.9) (P < 0.01).

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Number of times participated in an outdoor activity

Overall, researchers reported a 23% reduction in the number of times participants participated in an outdoor activity. The effect was greater for males than for females (25% vs 18% reduction). Data show a greater reduction among females from a minority ethnic group (43%) than among males from an ethnic minority group (14%); and a greater reduction among white males (29%) than among males from an ethnic minority (14%). Data for white females were not available.

Walter 1989

Know Your Body

School‐Universal

Recovery index score

In Westchester, the recovery index in the intervention group changed by ‐0.7 per year vs ‐1.4 in the control group (overall difference in school means 0.7 (95% CI ‐0.1 to 1.5)). Among schools in the Bronx, the rate of change per year in the intervention group was ‐2.5 vs ‐2.5 in the control group (difference in school means 0.0, 95% CI ‐1.3 to 1.3).

9. Nutrition

O’Neill 2016

Michigan Model for Health

School‐Universal

Nutritional behaviours

Six weeks following the intervention, results show a significant effect on nutritional behaviours: F[53,213.47] = 2.32, P = 0.04.

Walter 1989

Know Your Body

School‐Universal

BMI

In Westchester, data showed no change per year among intervention schools (mean 0.0 (SD 0.1)) vs a change of 0.1 per year (SD 0.1) among control schools (difference ‐0.1, 95% CI ‐0.3 to 0.1). In the Bronx, the rate of change per year among intervention schools was 0.1 (SD 0.1) vs 0.2 (SD 0.1) among control schools (difference ‐0.1, 95% CI ‐0.3 to 0.1).

Walter 1989

Know Your Body

School‐Universal

Plasma total cholesterol (mg/dL)

In Westchester, the rate of change in total cholesterol was ‐2.1 mg/dL/y (SD 1.0) among intervention schools but ‐0.4 mg/dL/y (SD 0.7) among control schools ‐ equivalent to a net mean change in total cholesterol of ‐1.7 mg/dL/y (‐2.7 to ‐0.7 mg/dL). Among intervention schools in the Bronx, the rate of change was ‐2.6 mg/dL/y (SD 1.5) vs ‐1.6 (SD 1.8) among control schools ‐ equivalent to a difference of ‐1.0 mg/dL/y (95% CI ‐2.3 to 0.3 mg/dL).

Walter 1989

Know Your Body

School‐Universal

Total fat (% of total kcal)

In Westchester, the net mean reduction in total fat intake between intervention and control schools was ‐3.6% (95% CI ‐7.1 to ‐0.1%); in the Bronx, the net mean reduction in total fat intake was ‐1.9% (95% CI ‐7.1 to 3.3%). Data are presented from a random subsample of the total study population.

Walter 1989

Know Your Body

School‐Universal

Systolic blood pressure (mmHg)

Among schools in Westchester, systolic blood pressure changed by 0.6 mmHg (SD 0.8) vs 0.8 mmHg (SD 0.6) in the control group, for an overall difference of ‐0.2 mmHg (‐1.0 to 0.6 mmHg).

10. Mental health

Bonds 2010

New Beginnings

Family‐Targeted

Internalising disorder, externalising disorder

6‐year follow‐up: the MPCP intervention arm had significantly fewer externalising problems (‐0.11, SE 0.11) compared to the control group (0.08, SE 0.14) (P = 0.02). There was no difference between intervention and control in the number of internalising problems nor in the mental disorder symptom count (P ≥ 0.05).

15‐year follow‐up: lower proportion of intervention group participants with (1) internalising disorder diagnosed in past 9 years; intervention: 4.55% (SD 2.69), control: 16.7% (SD 3.25, OR 0.26), and (2) externalising disorder diagnosed in past 9 years; intervention: 0% (SD 0), control: 3.64% (SD 0.04).

Gonzales 2014

Bridges to High School (Bridges/ Puentes)

Family‐Targeted

Internalising symptoms

Intervention was associated with slight increase in internalising symptoms at 2 years post‐test (unstandardised regression coefficient 0.42) but a small reduction in internalising symptoms at 5 years post‐test (unstandardised regression coefficient ‐0.02).

Kellam 2008

Good Behavior Game

School‐Universal

Lifetime major depressive disorder and generalised anxiety disorder

At ˜ 12 years following intervention, when participants were aged 19 to 21 years, unadjusted rates of lifetime major depressive disorder were lower for the GBG group (10%) than for the control group (15%) (P = 0.27). The difference was slightly larger for males than for females (males: 9% for GBG, 14% for controls; females: 12% for GBG, 15% for controls). Overall rates of generalised anxiety disorder were small and did not differ by intervention condition (2% for GBG, 3% for control; P = 0.37).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Internalising problems

Results showed no difference between study arms in the mean number of internalising problems at 15‐year follow‐up: for those visited during pregnancy, mean 11.19 vs 10.58, P = 0.46; and among a subgroup of women from low SES households who were unmarried, mean 11.15 vs 10.82, P = 0.8.

For those visited through pregnancy and infancy, no difference between groups was evident: mean 11.66 vs 10.58, P = 0.19; among the subgroup of women from low SES households who were unmarried, mean 9.85 vs 10.82, P = 0.44.

Cho 2005 (Sanchez 2007, Hallfors 2006)

Reconnecting Youth

School‐Targeted

Anger

Findings regarding depression and anxiety were not reported. However, study authors report that at 6‐month follow‐up, a negative outcome was observed for those in the experimental arm compared to those in the control arm: main programme effect: F= ‐3.62, P = 0.058 (i.e. those in the intervention arm showed greater frequency of anger compared to those in the control arm).

Walker 2002

(none given)

Individual‐Universal

Mental health score

Data show no difference in change in mental health score between intervention and control participants at 3 or 12 months. However, among young people who scored 16 or more on the depression scale (indicating probable depression), there was a greater reduction in mental health score than among those in the control group (‐8.1 intervention, ‐1.4 control, 95% confidence interval (CI) for mean difference ‐0.3 to ‐13.3, P = 0.04 at 3 months; ‐1.6 intervention, 4.4 control, 95% CI ‐0.5 to ‐11.5, P = 0.03).

11. Educational attainment

Berry 2009

Coaching for Communities

Individual‐Targeted

In education/employment

At follow‐up (post intervention), 85% of those in the intervention group were in education or employment vs 59% of those in the control group (P < 0.05).

Bond 2004

Gatehouse Project

School‐Universal

Low school attachment

Two years following the intervention, the OR for low school attachment was 1.21 (95% CI 0.93 to 1.57).

Conduct Problems Prevention Research Group 2010

Fast Track

School‐Targeted

Graduated from high school or graduate equivalency diploma

At long‐term follow‐up, the adjusted OR for graduation from high school or a graduate equivalency diploma was 0.93 (95% CI 0.68 to 1.27, P = 0.654).

Freudenberg 2010

REAL MEN

Individual‐Targeted

Enrolled in educational or vocational programmes in the past year

Receipt of the intervention was associated with increased odds of being enrolled in educational/vocational programmes (OR 1.330). No 95% CI or standard error was provided.

Friedman 2002

Life Skills Training and Anti‐violence Program

Individual‐Targeted

School problems

No difference was evident between groups in relation to school problems (t = 0.91, P > 0.05).

Gonzales 2014

Bridges to HIgh School (Bridges/Puentes)

Family‐Targeted

High school dropout (no high school degree or equivalent and no attendance at high school at 12th grade assessment)

The path coefficient estimate for high school dropout at 5‐year follow‐up was not statistically significant (unstandardised regression coefficient ‐0.16), but an indirect effect of the intervention was identified through school engagement (unstandardised regression coefficient ‐0.062, 95% CI ‐0.517 to ‐0.001).

Kellam 2014

Good Behavior Game

School‐Universal

High school graduation

High school graduation rates were slightly higher for those in the GBG arm (72%) than for those in the control arm (64%), and this effect was larger for males than for females (68% vs 54%, respectively). However, these data were not adjusted for clustering.

Kitzman 2010

Nurse Family Partnership

Family‐Targeted

Academic achievement (grade point average ‐ GPA)

The GPA for grades 1 to 6 for those in the intervention group was 2.39 (0.04) vs 2.48 (0.05) for those in the control group (P = 0.19, mean difference 0.09 (‐0.04 to 0.22)). For PIAT scores (reading and maths) at 12 years, the mean difference was 1.27 (‐0.44 to 2.98) (P = 0.14). Among families of lower socioeconomic status, those in the intervention group had higher PIAT scores in reading and math at age 12 (ES 0.25, P = 0.009), higher GPAs and group‐based achievement test scores in reading and math in grades 1 through 6 (ES 0.18, P = 0.03; ES 0.22, P = 0.02, respectively), and higher GPAs in reading and math in grades 4 through 6 (ES 0.18, P = 0.047).

Li 2011

Positive Action

School‐Universal

Suspension from school

No difference between study arms was observed at follow‐up in relation to suspensions from school (IRR 0.58, 95% CI 0.15 to 2.26).

LoSciuto 1999

Woodrock Youth Development Project

School‐Universal

School attendance

Participants in the intervention group reported better average scores for self‐reported school attendance (F(1,705) = 12.18, P < 0.01, Cohen's d = 0.26). Insufficient data were available to adjust these findings for clustering of participants by classroom.

Melnyk 2003

COPE

School‐Universal

Academic competence

Academic competence was slightly higher in the intervention group than in the control group (adjusted mean 97.97, 95% CI 96.35 to 99.59; vs 95.69, 95% CI 94.21 to 97.18), respectively. F = 4.03, P = 0.05.

Morris 2003

Self‐Sufficiency Project

Family‐Targeted

Dropped out of school (aged 15 to 18)

At 36‐month follow‐up, math score at age 12 to 14 was 0.45 in the intervention group compared to 0.46 in the control group (ES ‐0.03); and average achievement was 3.43 compared to 3.54 in the control group (ES ‐0.11). Child‐reported average achievement was 3.50 in the intervention group vs 3.57 in the control group (ES ‐0.09).

Morris 2003

Self‐Sufficiency Project

Family‐Targeted

Average achievement (self‐reported)

Self‐reported average achievement was similar between intervention and control groups at 36 months of follow‐up (effect size ‐0.09).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Mean long‐term school suspensions

Data show no difference between study arms in the mean number of long‐term school suspensions at 15‐year follow‐up. For those visited during pregnancy: mean 0.0 vs 0.04, P = 1.0; among a subgroup of women from low SES households who were unmarried: mean 0.01 vs 0.15, P = 0.97.

For those visited through pregnancy and infancy, no difference between groups was evident: mean 0.01 vs 0.04, P = 1.0; among the subgroup of women from low SES households who were unmarried, mean 0.04 vs 0.15, P = 0.25.

Schweinhart 1993

High/Scope Perry Preschool Program

School‐Targeted

Total school achievement

Those in the intervention group had higher total achievement (mean 122.2, SD 41.6) than those in the control group (mean 94.5, SD 35.5) at 9‐year follow‐up.

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Grade point average

Educational impacts of the intervention were more pronounced among girls than among boys. Overall, those receiving the intervention overall had a higher grade point average (GPA) than those given control (average 2.71 vs 2.63). The difference was particularly marked among girls from an ethnic minority (average GPA 2.83 vs 2.62 for those in the control group).

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Truancy/skipping school

Participants of Big Brothers Big Sisters showed a 52% reduction in the number of times they skipped a day of school and a 37% reduction in the number of times they skipped class. The impact was greater among girls than boys, for instance those in the intervention group showed 84% reduction compared to 4% reduction for skipping a day of school. The reduction was greater for white females than for females from an ethnic minority (92% reduction vs 78% reduction, respectively, for skipping a day of school; 72% vs 46% reduction for skipping class). Among white males compared to males from an ethnic minority, the reduction was similar for skipping a day of school, but a greater impact was evident among ethnic minority males than among white males for skipping class (22% vs 12% reduction).

12. Teenage pregnancy

Olds 1998

Nurse Family Partnership

Family‐Targeted

Ever pregnant or made someone pregnant in the previous 6 months

At age 15 (˜ 13 years following the intervention), 5 of 176 (2.8%) in the intervention group had ever been pregnant or made someone pregnant compared to 4 of 148 (2.7%) in the control group (OR 1.04, SE 0.65).

Conduct Problems Prevention Research Group 2010

Fast Track

School‐Targeted

Pregnancy by age 18

At age 18, the proportion of participants reporting pregnancy were as follows: girls: urban African American 40%, urban European American 21%, rural European American 17%; boys: urban African American 27%, urban European American 12%, rural European American 11%.

Schweinhart 1993

High/Scope Perry Preschool Program

School‐Targeted

At least 1 pregnancy by age 19 (females only)

At 14‐year follow‐up, 12 of 25 (48%) in the intervention group had had at least 1 pregnancy by age 19 compared to 16 of 24 (67%) in the control group (effect size 0.5).

13. Health problems

Schweinhart 1993

High/Scope Perry Preschool Study

School‐Targeted

Long‐term health problems

22 years following completion of the intervention, 36% of those in the intervention group had had health problems treated in the previous 5 years compared to 38% in the control group (effect size 0.04, P = 0.823). 30% of those in the intervention group had been hospitalised in the previous 12 months compared to 15% of those in the control group (effect size 0.38; P = 0.043).

ASPD: antisocial personality disorder.

CI: confidence interval.

COPE: Creating Opportunities for Personal Empowerment.

CPC: Community Practice Condition.

DARE: Drug Abuse and Resistance Education.

DSM: Diagnostic and Statistical Manual of Mental Disorders.

ES: Effect size.

GBG: Good Behaviour Game.

GPA: grade point average.

HLM: hierarchical linear modelling.

LIFT: Linking the Interests of Families and Teachers.

MPCP: Mother program plus child program.

OR: odds ratio.

PATHS: Promoting Alternative Thinking Strategies.

ProSAAF: Protecting Stronger African American Families program.

ROAD: Reducing Offenses of Adolescent Drivers.

SafERteen: brief intervention aimed at reducing and preventing violence and alcohol use.

SD: standard deviation.

SE: standard error.

SES: socioeconomic status.

SMD: standardised mean difference.

Assessment of heterogeneity

We anticipated that the studies included in this review would be heterogeneous with respect to settings, participants, interventions, and outcomes, and so conducted random‐effects meta‐analyses. We examined heterogeneity via visual inspection of the forest plot, the Chi² test, and the I² statistic to assess whether observed variability was compatible with chance. For each outcome, we included subgroups for study setting and focus (individual, family, or school level and universal or targeted). Data were insufficient for examination of further subgroupings and/or for exploration of further reasons for heterogeneity.

Assessment of reporting biases

If we identified sufficient studies (minimum of 10), we used funnel plots to examine the study effect size against the sample size to look for publication bias or small‐study effects (Sterne 2011).

Data synthesis

We applied a random‐effects (DerSimonian and Laird) model for meta‐analysis using RevMan 5.3, so we could allow for the assumption that different studies are estimating different, yet related, intervention effects (DerSimonian 1986). Therefore the pooled effect estimates described in the review should be interpreted as the average treatment effect. For each outcome, we grouped studies according to study type (school, family, or individual level, and whether they were universal or targeted). We obtained the overall effect estimate for each subgroup. We did not calculate a summary estimate of subgroups together owing to the distinct nature of each group. However, we presented data for each group on a single forest plot for simplicity in presentation.

When studies presented results of multi‐arm trials, review authors (JR, GJM, DMC RC, MH) agreed on the extent to which additional study arms included a component with a distinct mode of delivery. When the additional study arm included an intervention component delivered in a distinct way, we treated these as separate comparisons, as described in Higgins 2008 (see Chapter 16 (16.5.4)). For studies in which the additional study arm was similar, we combined data from different study arms, as described in Higgins 2008 (see Chapter 7 (7.7.3.8)). When data were presented separately by gender or by ethnic group, we also combined them using methods described in Higgins 2008 (see Chapter 7 (7.7.3.8)). We combined measures that were highly related (e.g. overt and covert delinquency), using methods described in Borenstein 2009 (see Chapter 24). We used the value for the correlation between measures, when provided. Otherwise, we used a value of 0.5, as recommended.

When study authors reported multiple measures of behaviours (e.g. condom use, number of sexual partners), we extracted all data and we selected as the main outcome measure the outcome that represented the behaviour leading to greatest harm for public health, as demonstrated by evidence regarding consequences for mortality or morbidity. For studies with long follow‐up that reported multiple repeated measurements, or that measured the outcome at multiple points in time, we extracted data from each time point. We included two follow‐up periods: up to 12 months (short‐term) and over 12 months (long‐term), which included follow‐up periods up to our maximum of 15 years post intervention. Quantitative analyses included data related to the longest point of follow‐up within the follow‐up category (e.g. for short‐term follow‐up) and used the time point closest to 12 months' post intervention as the primary endpoint.

Subgroup analysis and investigation of heterogeneity

We planned to conduct subgroup analyses; however data within each subgroup for each behavioural outcome were insufficient for further analyses to examine the impact of interventions according to our prespecified characteristics of population subgroups, settings, and intervention components.

Sensitivity analysis

We conducted sensitivity analyses to examine the impact of using conservative versus low imputed ICCs when accounting for clustering in school‐based RCTs. For analyses that demonstrated a beneficial effect, we also conducted sensitivity analyses to examine whether statistical transformations had an impact on the findings. Last, given that a small number of studies were conducted in middle‐income countries, we conducted sensitivity analyses to test whether findings were affected by their exclusion. We did not conduct sensitivity analyses around risk of bias of included studies because exclusion of those at high or unclear risk of bias in key domains of selection and performance bias left insufficient studies in each study type subgroup.

Summarising and interpreting results

We assessed the overall quality of the body of evidence for each outcome using the Grades of Recommendation, Assessment, Development and Evaluation (GRADE) approach (Guyatt 2008), in keeping with standards for reporting of Cochrane Intervention Reviews, specifically, MECIR (Methodological Expectations of Cochrane Intervention Reviews) conduct standards 74 and 75. GRADE is also the most widely used approach for summarising confidence in effects of interventions by outcome; it is used by over 20 organisations internationally (Higgins 2008; Higgins 2018). The GRADE approach defines the quality of the evidence, involving consideration of within‐study risk of bias (methodological quality), directness of evidence, inconsistency (heterogeneity), precision of effect estimates, and risk of publication bias. We downgraded evidence depending on the presence of these factors. We summarised findings for seven key outcomes (alcohol use, tobacco use, cannabis use, illicit drug use, sexual risk behaviour, antisocial behaviour, and physical activity) in summary of findings Table for the main comparison, which includes the number of participants and studies for each outcome, the intervention effect for the relevant subgroup, and a measure of the quality of the body of evidence, using the GRADE approach.

Although we note that blinding is not possible in the interventions included in this review, we downgraded all studies on the basis of high risk of bias related to lack of blinding and, in some cases, unclear risk of bias related to allocation concealment and/or selective reporting, which were largely due to lack of clarity in reporting. We downgraded studies on the basis of inconsistency if we found evidence of substantial heterogeneity (I²), as well as poor overlap in 95% confidence intervals between studies and large between‐study variance (tau²). We took into account the extent of consistency in the direction of point estimates of individual studies. Further to subgroups by study type, we were unable to conduct additional subgroup analyses to explore possible explanations for observed heterogeneity, such as variation in intensity or duration of interventions, or age at implementation; these could not be examined further.

Most of the studies included in this review were conducted in high‐income countries (n = 67; 96%), one (1.4%) in a lower‐middle‐income country (Saraf 2015), and one (1.4%) in an upper‐middle‐income country (Matthews 2016); one (1.4%) was a joint study including an upper‐middle‐income country and a high‐income country (Lana 2014). Although this fact limits generalisability to these settings, we did not downgrade the quality of the evidence on the basis of indirectness because the objective of the review was to examine the overall effectiveness of interventions that aimed to prevent engagement in multiple risk behaviours, rather than the impact of these interventions in particular settings or geographical regions. Furthermore, although it could be argued that generalisability for certain targeted family‐level studies (e.g. the Nurse Family Partnership) could be limited owing to contextual factors related to service provision, this was relevant only to a small proportion of studies; thus we did not consider it of sufficient concern to justify downgrading the quality of the evidence on this basis. We downgraded the quality of evidence on the basis of possible small‐study or publication bias if we identified at least ten studies assessing that outcome, and if asymmetry was evident in the funnel plot.

Results

Description of studies

Results of the search

During the course of this review, we conducted three database searches (in 2012, 2015, and 2016). The initial search in 2012 produced 19,220 records, yielding 18,706 reports after removal of duplicate records. We obtained a further 9302 records in May 2015, leaving 5847 reports after removal of duplicates; we identified 5944 articles in November 2016, for a total of 3138 articles after removal of duplicates. Therefore, in total we identified 34,680 titles, screened 27,691 unique articles (2012: n = 18,706; 2015: n = 5847; 2016: n = 3138), and we obtained 424 full‐text articles. From these full‐text articles, we identified 70 studies for inclusion in this review (Figure 2). We have provided a description of each study in the Characteristics of included studies table. We will include ongoing studies and those awaiting classification in future updates of this review.


Study flow diagram (searches conducted in 2012, 2015, and 2016).

Study flow diagram (searches conducted in 2012, 2015, and 2016).

Included studies

We included 70 studies in this review. We have provided summaries of each of the 70 included studies in the Characteristics of included studies table, and in Additional Table 3, we have provided details about each study and behaviours targeted by study type (e.g. universal family‐level interventions, targeted school‐level interventions).

Open in table viewer
Table 3. Summary of studies by intervention type

Primary authors

Trial

Study type

Country

Duration

Theory

Follow‐up (post intervention)

Components

Age targeted

Behaviour targeted

N behaviours targeted

Process evaluation

Targeted individual‐level interventions

Bernstein 2010

Reaching Adolescents for Prevention

IT

USA

< 3 months (motivational intervention, referral to services, telephone conversation)

N/S

12 months

Motivational interview, referral to community resources and drug treatment services

14‐21

Alcohol use, vehicle‐related risk, antisocial behaviour, sexual risk behaviour

4

Assessed adherence to intervention

Berry 2009

Coaching for Communities

IT

UK

10 months

Distinction‐based learning

Post intervention

1‐week residential programme and 9 months of mentoring

15‐18

Alcohol use, drug use, antisocial behaviour, educational attainment

4

Assessed impact of ‘quality’ of the mentor and examined impact of dose of the intervention

Bodin 2011

A Mentoring Intervention

IT

Sweden

12 months

Rhodes model regarding role modelling on healthy relationships with adults

12 months

Mentoring

13‐18

Alcohol use, drug use, academic development, mental health

5

Yes (partial) ‐ assessed dropout, adherence, meetings, positive views of programme, intensity

Cunningham 2012

SafERteens

IT

USA

< 1 month (MI)

Traditional motivational interviewing model

12 months

Traditional MI using computer or therapist

14‐18

Antisocial behaviour, alcohol use

2

No

Dolan 2010

BBBS Ireland

IT

Ireland

12 months

Rhodes model of youth mentoring

2 years

Individual (mentoring)

10‐16 (93% < 14 years)

Alcohol use, tobacco use, cannabis use, antisocial behaviour, educational attainment, mental health

6

Yes ‐ full

Freudenberg 2010

REAL MEN

IT

USA

1‐2 months

‐‐‐

12 months

Individual components (jail‐based intervention and intervention within community setting)

17‐18

Drug use, sexual risk behaviour, antisocial behaviour, education, and employment

4

No

Friedman 2002

Botvin LST and Anti‐violence

IT

USA

6‐12 months (55 sessions for classroom programme, 20 sessions for violence programme)

Life Skills Training model; social cognitive procedures

6 months

Individual (triple‐modality classroom programme)

13‐18

Alcohol use, drug use, antisocial behaviour, educational attainment

4

Yes ‐ partial (feedback from participants, adherence, intensity)

Monti 1999

Alcohol Screening and brief intervention

IT

USA

1 day

MI

6 months

1 individual component: brief motivational interviewing

18

Alcohol use, vehicle risk behaviour

2

Yes (partial) ‐ rating of adherence, videotaping of interventionist to rate fidelity

Nirenberg 2013

ROAD

IT

USA

1 day (BI)

No

6 months

2 individual components: motivational interviewing and placement in hospital emergency department or in the community

18

Alcohol use, vehicle risk behaviour

2

No

Redding 2015

Step‐by‐Step

IT

USA

9 months

Transtheoretical model (TTM)

18 months

2 individual components: TTM‐tailored feedback via a computer‐based system and personalised stage‐targeted counselling

16

Tobacco use, sexual risk behaviour

2

Yes (full) ‐ counsellors reported on what was covered in sessions and activities they used. Teen feedback was obtained.

Tierney 1995

Big Brothers Big Sisters

IT

USA

12 months

N/S

6 months

Individual component: mentoring (3 meetings per month over a period of nearly 12 months on average)

10‐16

Alcohol use, drug use, antisocial behaviour, educational attainment

5

Yes (partial). Participants’ feedback regarding volunteer impact. Adherence.

Wagner 2014

Guided Self‐Change

IT

USA

2 months

Guided Self‐Change

6 months post intervention

Individual component: brief motivational interview via cognitive‐behavioural approach

14‐18

(mean 16.2, SD 1.2)

Alcohol use, drug use, antisocial behaviour

3

Yes (partial) ‐ recorded sessions and reviewed to assess adherence

Universal individual‐level interventions

Johnson 2015

Healthy Futures

IU

USA

3 months

Social learning theory

6 months

Individual component: 3 sessions of motivational interviewing (once per month) with follow‐up in‐between

14‐21

(mean 16)

Alcohol use, illicit drug use, antisocial behaviour

3

No

Lana 2014

Prevencanadol

IU

Spain, Mexico

9 months (1 school year)

Transtheoretical model of behaviour change; ASE model

Post intervention

Individual component – website regarding prevention and treatment of cancer

12‐15

Alcohol use, tobacco use, diet, physical activity

4

No

Minnis 2014

Yo Puedo

IU

USA

6 months

Social learning theory; behavioural economics

Post intervention

Individual components: cash payments and life skills sessions

16

Alcohol use, drug use, antisocial behaviour, sexual risk behaviour

5

Yes (partial) ‐ adherence, attendance, cash earned, meeting goals

Walker 2002

[No study name]

IU

UK

1 day (BI)

Self‐efficacy theory

12 months post intervention

Individual component: 20‐minute consultation with practice nurse to discuss health and health‐related behaviour

14‐15

Alcohol use, tobacco use, physical activity, nutrition, mental health

5

Yes (full) ‐ attendees: acceptability; observation of nurses to assess adherence

Targeted family‐level interventions

Beach 2016

ProSAAF

FT

USA

2‐3 months

Based on previous interventions and stress‐spillover theory

˜ 9 months

Family component: six 2‐hour home‐delivered sessions that focused on couple‐based issues and development of protective couple and parenting processes

10‐13

Tobacco use, alcohol use, illicit drug use, antisocial behaviour

4

No

Bonds 2010

New Beginnings

FT

USA

3 months

Cascading pathway model

3 months, 6 months, 6 years, 15 years

Family components only (mother‐only, mother‐plus‐child, and child‐only programmes)

9‐12 (average 10.4, SD 1.1)

Substance use (marijuana and alcohol), antisocial behaviour, sexual risk behaviour, mental health

5

Yes: adherence, feedback, manuals, training, supervision, scores, rating

Brody 2012

SAAF‐T

FT

USA

2‐3 months

N/S

22 months

Family components only (5 meetings for caregivers and adolescents separately, followed by a 1‐hour session for families together)

16

Drug use, antisocial behaviour, mental health

3

Adherence to curriculum, video of sessions and feedback

Catalano 1999

Focus on Families

FT

USA

1 year (approximately)

Social development model

8 months

Family components: parent skills training sessions and case management

3‐14

Alcohol use, tobacco use, cannabis use, antisocial behaviour

4

No ‐ adherence only

Estrada 2016

Brief Familias Unidas

FT

USA

2‐3 months

NS

24 months

5 weekly 2‐hour parent group sessions, 3 homework assignments for parents, and one 1‐hour family visit

15

(mean 15.3 years, SD 0.89)

Tobacco use, alcohol use, illicit drug use, sexual risk behaviour

4

Yes (partial) ‐ attendance data only

Gonzales

Bridges to High School

FT

USA

3 months

Social development model

5 years

Family components (weekly group sessions with separate and joint sessions and home visits)

12‐13

Alcohol use, tobacco use, drug use, antisocial behaviour, educational attainment, mental health

6

Yes (full) ‐ post‐test interview ‐ rating by parents; attendance; video recording of intervention sessions ‐ adherence

Jalling 2016

Comet 12‐18

FT

Sweden

3‐4 months

Operant learning and social learning principles

6 months

Family component: 9 weekly group sessions of 2 to 2.5 hours and 1 optional booster session. Sessions involved role‐play, home assignments, and use of video as a basis for discussion.

12‐18

(mean 14)

Alcohol use, illicit drug use, antisocial behaviour

3

Yes ‐ attendance and self‐assessment by group leaders of extent to which programme manual was fulfilled in sessions

Jalling 2016b

ParentSteps

FT

Sweden

3‐4 months

Resilience model

6 months

Family component: 6 weekly parent sessions of 1.5 to 2 hours

12‐18

(mean 14)

Alcohol use, illicit drug use, antisocial behaviour

3

Yes (partial) ‐ attendance and leader self‐assessment of extent to which programme manual was fulfilled in sessions

Kim 2011

Middle School Success

FT

USA

3 months with ongoing support for 1 school year

‐‐‐

2 years

Family components: curriculum to parents through 6 group sessions for parents plus follow‐up sessions; 6 skills‐based sessions for girls; ongoing training and support for parents and adolescents

11‐12

Antisocial behaviour, tobacco use, alcohol use, marijuana use, mental health

5

No

Kitzman 2010

Nurse Family Partnership 2

FT

USA

2.5 years

Theories of child development, behaviour change, human ecology, self‐efficacy, and attachment

12 years

Family components: free transportation to prenatal visits, screening, referral; prenatal, infant, and child home visitation

0 ‐2

Alcohol use, tobacco use, drug use, antisocial behaviour, mental health, educational attainment

6a

No

Li 2002

imPACT

FT

USA

90 minutes (1 day)

None

12 months

1× individual (video)

14

Alcohol use, tobacco use, drug (cannabis) use, antisocial behaviour, sexual risk behaviour

5

No

Milburn 2012

STRIVE

FT

USA

2 months

Cognitive‐behavioural theories

12 months post intervention

Family component: 5 sessions delivered to young person and parent

14

Drug use, antisocial behaviour, sexual risk behaviour

3

Yes (partial) ‐ attendance, satisfaction of parents/ adolescents; manual ‐ assessed fidelity of session delivery

Morris 2003

Self‐Sufficiency Project

FT

Canada

Up to 3 years

Economics and psychology theories

Post intervention

Individual component: earnings supplement to single parents who left welfare for full‐time employment

0‐2, 3‐8, or 9‐15

Alcohol use, tobacco use, drug use, antisocial behaviour, mental health educational attainment

6

No

Murry 2014

SAAF (Stronger African American Families)

FT

USA

7 weeks

Social learning theory, problem behaviour theory, Gibbons and Gerrard’s cognitive model of adolescent behaviour

65 months

1 family component: separate 1‐hour caregiver and adolescent session followed by joint session to practice skills

Mothers and children aged 11 years

Alcohol use, sexual risk behaviour

2

Yes ‐ attendance measured, fidelity assessed using video

Olds 1998

Nurse Family Partnership

FT

USA

2 years

Human ecology, self‐efficacy, human attachment

15 years

1 family component

Mothers aged < 19, children aged 0‐2

Drug use, antisocial behaviour, sexual risk behaviour, educational attainment

6a

No

Pantin 2009

Familias Unidas

FT

USA

N/S

N/S

2.5 years post intervention

2 family components: nine 2‐hour group sessions, ten 1‐hour family visits, four 1‐hour booster sessions.

13‐14

Alcohol use, tobacco use, drug use, antisocial behaviour, sexual risk behaviour

5

Yes (partial) ‐ sessions video recorded and rated on adherence and quality

Schwinn 2014

[No study name]

FT

USA

1 month

N/S

5 months

1 family component: 3‐session online health promotion programme

11‐12

Drug use, physical activity, nutrition

3

Minimal adherence data only

Universal family‐level interventions

Averdijk 2016

Triple P

FU

Switzerland

3 months

N/S

9 years

1 family component: a group‐based course with 4 units of 2 to 2.5 hours and 4 follow‐up telephone contacts with each participant

7.5 years

Tobacco use, alcohol use, illicit drug use, antisocial behaviour

4

Yes ‐ attendance at sessions, satisfaction with programme, provider competency, and delivery of course material

Connell 2007

Family Check‐Up

FU

USA

2.5 years

Life skills training programme

3.5 years (6 years including intervention)

School programme including a universal classroom intervention; parenting practice component with assessment and feedback, family management treatment

11‐12

Alcohol use, tobacco use, drug use, anti‐social behaviour

4

Assessed adherence only

Haggerty 2007

Parents Who Care

FT

USA

2‐3 months

Social development model

2 years post intervention

7‐session group intervention for parent and adolescent or a 7‐session self‐administered intervention for adolescent and parent with telephone support

13‐14

Alcohol use, tobacco use, drug use, antisocial behaviour, sexual risk behaviour

5

Yes (full) ‐ parents; satisfaction; student satisfaction; adherence, quality

Targeted school‐level interventions

Conduct Problems Prevention Research Group 2014

Fast Track

ST

USA

10 years

Developmental model of conduct disorders

19 years

School and family components (family group programme, parent training groups, parent‐child interaction groups, tutoring; school curriculum; youth groups, youth forums)

Kindergarten to 12th grade

Antisocial behaviour, educational attainment, mental health

3

Yes ‐ training, supervision, fidelity ‐ rating of quality of implementation, observation, quality of teaching, quality of classroom management

Lochman 2003

Coping Power 1

ST

USA

12 months

Social learning theory

12 months post intervention

Parent and child components: parent group meetings; teacher meetings; group school‐based sessions for children

10‐11

Alcohol use, tobacco use, drug use, antisocial behaviour

4

Yes (partial) ‐ adherence to intervention (manuals, guidelines, training); attendance

Lochman 2004

Coping Power 2

ST

USA

15 months

Social learning theory

12 months post intervention

Parent and child components: parent group sessions and child school‐based group sessions

9‐10

Alcohol use, drug use, antisocial behaviour

3

Yes (full) ‐ meetings with target children; staff rated accomplishment of objectives, audio/video taping; observation

Sanchez 2007

Reconnecting Youth

ST

USA

2 years

Theoretical framework based on strain, social learning and control theories

1 year post intervention

1 school component: 55 school lessons and 24 booster lessons

15

Drug use, antisocial behaviour, mental health, educational attainment

4

Yes (full) ‐ teacher logs, attendance records, observations in classroom, student questionnaires, implementation

Schweinhart 1993

High/Scope Perry Preschool Project

ST

USA

2 school years

Piaget's constructivist theory of child development

36 years

Preschool and parent components: preschool for 2.5 hours each weekday morning, home visits by teachers for 1.5 hours per week, parent group meetings

3‐4

Antisocial behaviour, educational attainment

2

No

Shetgiri 2011

[No study name]

ST

USA

9 months (1 school year)

N/S

Post intervention

3 school components; 28 weekly peer groups facilitated by a school clinical social worker, field trips, community service activities

13‐15

Alcohol use, tobacco use, drug use, antisocial behaviour

4

No

Universal school‐level interventions

Averdijk 2016b

PATHS

SU

Switzerland

1 year

Not clear

8 years

School curriculum: 46 lessons addressing problem‐solving, social relationships, self‐regulation, emotional understanding, rules, and positive self‐esteem

8‐9

(year 2 primary school)

Tobacco use, alcohol use, illicit drug use, antisocial behaviour

4

Yes ‐ classroom observations, rating of lesson quality

Beets 2009

Positive Action (Hawaii)

SU

USA

4‐5 years

Theory of self‐concept, consistent with theories of triadic influence

Post intervention

School curriculum with school‐wide climate programme, family, community components

K‐12

Alcohol use, tobacco use, drug use, violent behaviours, sexual risk behaviour

5

No

Bond 2004

Gatehouse Project

SU

Australia

24 months

Health‐promoting schools framework, ecological approach

4 years

Whole‐school approach involving a curriculum, institutional and individual‐focused components

13‐14

Alcohol use, tobacco use, cannabis use, mental health (emotional well‐being)

6

Yes ‐ full

Bush 1989

Know Your Body 2

SU

USA

5 years

Social learning theory

Following 2 years of intervention

School curriculum, screening and feedback, parental involvement, and newsletter to families

10‐13

Tobacco use, physical activity, cardiovascular risk factors (nutrition)

3

No ‐ adherence only

D'Amico 2002

Risk Skills Training Programme vs DARE

SU

USA

< 1 month (1 hour)

Based on models ‐ not theory (DARE, alcohol skills training programme, BASICS)

4 months

School components only: school‐based group session with curriculum

14‐19

Alcohol use, drug use, vehicle‐related risk behaviour

3

Yes ‐ rating of audiotapes, adherence to protocol, student rating of acceptability and feedback

DeGarmo 2009

LIFT

SU

USA

3 months

Developmental model centred on moment‐to‐moment social interaction processes

7 years

School and family components (parent management training, child skills training, school recess component)

10‐11

Alcohol use, tobacco use, drug use, antisocial behaviour

4

Yes ‐ acceptability of intervention to parents and teachers, adherence using checklists, family participation recorded, completion of critical components

Fearnow‐Kenney 2003

All Stars Sr

SU

USA

9 months (1 school year)

No

Post intervention

School components only (All Stars activities implemented by teachers)

13‐19

Alcohol use, cannabis use, tobacco use, nutrition

4

Yes ‐ full (teacher and student focus groups re perceptions of programme, % implemented)

Flay 2004

Aban Aya

SU

USA

4 years

Theory of triadic influence, incorporation of Nguzo Saba principles to promote African American cultural values

Post intervention

School curriculum from grade 5 to 8 (SDC); or curriculum plus parental support, school climate and community components

10‐14

Antisocial behaviour, sexual risk behaviour, substance use

3

Yes ‐ full

Gottfredson 2010

All Stars 2

SU

USA

32 weeks (1 school year)

Social learning theory

Post intervention

School components only: classroom interactive sessions, homework assignments, parental attendance at graduation ceremonies

11‐14

Alcohol use, tobacco use, drug use, antisocial behaviour

4

Yes (full) ‐ adherence ‐ observed implementation with site visits, fidelity checklists; quality rating and how sessions met objectives; adherence and delivery

Griffin 2006

Life Skills Training

SU

USA

3 years

Life skills training

10 years

School components: skills‐based curriculum with interactive teaching methods

12‐13

Alcohol use, drug use, sexual risk behaviour

3

Yes (partial) ‐ monitoring of classes ‐ completeness with respect to the % of curriculum covered

Griffin 2009

BRAVE (Building Resiliency and Vocational Excellence)

SU

USA

3 school years

Social learning theory

1 year

School component and individual component: health education classroom sessions and training sessions on life skills, manhood development, or violence prevention; development of career plans, buddy system; plus mentoring

13‐14

Alcohol use, tobacco use, drug use, antisocial behaviour

4

No ‐ just reviewed and practised using material with trainers; trainers were required to have schedule for delivery of lesson

Ialongo 1999

Classroom‐Centred (CC) programme including Good Behavior Game (GBG)

SU

USA

1 school year (9 months)

Life course/social field theory

5 years

Curriculum, behaviour management using the GBG, and strategies for children who failed to respond to intervention

6‐7

Antisocial behaviour, mental health, academic achievement (antecedents of substance use)

6a

Yes (full) ‐ checked classroom setup, observed classroom sessions, visit records. Classroom record reviews completed by students were reviewed.

Ialongo 1999

FSP

SU

USA

1 school year (9 months)

Life course/social field theory

5 years

Training for teachers and staff, home‐school learning activities, 9 workshops for parents

6‐7

Antisocial behaviour, mental health, academic achievement (antecedents of substance use)

6a

Yes (full) ‐ documentation of contact with parents. Parents reported on implementation and usefulness. Recorded observations of workshops

Kellam

Good Behaviour Game (GBG)

SU

USA

2 school years

Life course/social field theory

Up to 12 years

School component: behaviour management

6‐8

Antisocial behaviour, educational achievement (antecedents of substance use)

5a

No

Lewis 2013

Positive Action (Chicago)

SU

USA

6 school years

Self‐esteem enhancement theory, social ecological theory

Post intervention

School components: classroom curriculum; school‐wide climate development; teacher, family, counsellor, and community training

8‐13 (grades 3‐8)

Alcohol use, tobacco use, drug use, antisocial behaviour

6

Yes (partial) ‐ adherence to programme; workshops for teachers; unit implementation report at the end of each unit

LoSciuto 1999

Woodrock Youth Development Project

SU

USA

2 school years

None

Post test

3 components: education through seminars, psychosocial support (mentoring, tutoring, extracurricular activities), and family and community supports (family involvement, counselling, and outreach)

6‐14

(mean 10, SD 1.7)

Alcohol use, tobacco use, drug use, antisocial behaviour, educational attainment

5

No

Mathews 2016

PREPARE

SU

South Africa

12 months

Social cognition models including the reasoned action framework and the I‐Change theoretical model

12 months

4 school components: 21 interactive and skills‐based sessions of 1‐1.5 hours once per week; a school health service delivered by a nurse from a local public clinic; safety training to school personnel and parent representatives; and a school safety programme delivered to a randomly selected group of 20 volunteers

Grade 8 (mean 13 years)

Antisocial behaviour, sexual risk behaviour

2

Yes ‐ rating by participants of the quality of sessions, attendance at lessons, visits to the school nurse, and attendance at the safety programme. Facilitator performance scores

McNeal 2004

All Stars 1

SU

USA

9 months (1 school year)

Social learning theory

12 months post intervention

School‐based components: curriculum including classroom, group, and 1‐to‐2 sessions; homework to increase parental interaction/involvement

11‐13

Alcohol use, tobacco use, drug use, antisocial behaviour, sexual risk behaviour

5

Yes ‐ rating of sessions, rating of objectives achieved

Melnyk 2013

COPE

SU

USA

4 months

Cognitive‐behavioural theory

6 months

School and parent components: educational and cognitive‐behavioural skills‐building programme, including physical activity in each of the 15 sessions, homework, and a parent newsletter

14‐16

Alcohol use, drug use, physical activity, nutrition, mental health, educational attainment

6

Yes (partial) ‐ participants completed questionnaires, parents provided feedback. Fidelity of teachers measured

Nader 1999

CATCH 3

SU

USA

3 years

Social cognitive theory, social learning theory, organisational change theory

3 years post intervention

School and family components: classroom curriculum, teacher training, consultation to teachers, physical activity sessions, school policy, home‐based tobacco use prevention programme, family activities to promote physical activity

8‐11

Tobacco use, physical activity, nutrition

3

Yes ‐ full

O'Neill 2011

Michigan Model for Health

SU

USA

2 years

Health Belief Model, social learning theory

2 months

School‐based curriculum and skills‐based learning (24 lessons in grade 4; 28 lessons in grade 5)

9‐11

Alcohol use, tobacco use, antisocial behaviour, physical activity, nutrition

5

Yes (partial) ‐ adherence to instructor, fidelity to protocol, teacher survey regarding lessons delivered, implementation fidelity assessment

Perry 2003

DARE vs DARE+

SU

USA

18 months

N/S

Post intervention

DARE: 1 school component involving 10‐session curriculum delivered by police officers. DARE‐Plus: 3 school components including 4‐session classroom‐based, peer‐led, parental involvement programme, extracurricular activities, and neighbourhood action teams organised by community leaders

12‐13

Alcohol use, tobacco use, drug use, antisocial behaviour

4

No

Piper 2000

Healthy for Life

SU

USA

Intensive: 3 months; age appropriate: 3 years

Social influences model

24 months post intervention

2 school components: 54‐lesson curriculum, use of peer leaders.

1 family component: parent orientation session, home mailings, homework involving parent/adult interviews.

1 community component: community organiser, sponsorship of health event, policy work

11

Alcohol use, tobacco use, drug (cannabis) use, sexual risk behaviour, nutrition

5

Yes (full) ‐ teacher logs re sessions; observation; interviews with staff, teachers, students; feedback surveys. Context, implementation

Saraf 2015

[No study name]

SU

India

9 months

N/S

Post intervention

2 school components: a school health committee, classroom activities, school policies.

2 community component: community outreach

12‐13

Tobacco use, physical activity, nutrition

3

No – number of schools that adopted policies is stated

Shek 2011

PATHS

SU

Hong Kong

20 hours for each of the 3 school years

N/S

Post intervention

1 school component: school curriculum

12‐14

Alcohol use, tobacco use, drug use, antisocial behaviour, sexual risk behaviour

5

No

Simons‐Morton 2005

Going Places

SU

USA

3 school years

Social development and social cognitive theory

3 months post intervention

2 school components: social skills curriculum (18 sessions in 6th grade, 12 in 7th grade, 8 in 8th grade) and school environment enhancement.

Parent component: parent education via video, booklet, periodic newsletters, and involvement in homework

11‐14

Alcohol use, tobacco use, antisocial behaviour

3

Yes (full) ‐ adherence, teacher rating of students’ participation, student feedback regarding attendance, helpfulness, knowledge. Parent interviews.

Walter 1989

Know Your Body 1

SU

USA

5‐6 years

PRECEDE health education planning model (which incorporates elements of social learning theory and the Health Belief Model)

Post intervention

School component: curriculum for 2 hours per week for each school year (grades 4‐9).

Parent component: self‐assessment, newsletters, participation in activities, seminars.

8‐15

(mean 9 years at baseline)

Tobacco use, physical activity, nutrition

3

Yes (partial) ‐ adherence to protocol, visits to classrooms

Wolfe 2009

Fourth R‐Skills for Youth Relationships

SU

Canada

8 months

N/S

2 years

School component: 21‐lesson curriculum.

14‐15

Alcohol use, tobacco use, drug use, antisocial behaviour, sexual risk behaviour

5

Yes (partial) ‐ teacher checklists re completion

ASE: Attitude ‐ social influence ‐ self‐efficacy model.

BASICS: Brief Alcohol Screening and Intervention for College Students.

BBBS: Big Brothers Big Sisters.

BI: Brief intervention.

BRAVE: Building Resiliency And Vocational Excellence.

CATCH: Coordinated approach to child health.

CC: classroom‐centred.

COPE: Creating Opportunities for Personal Empowerment.

DARE: Drug Abuse and Resistance Education.

FSP: Family Schools Partnership.

FT: family‐targeted.

FU: family‐universal.

GBG: Good Behaviour Game.

imPACT: Informed Parents and Children Together.

IT: individual‐targeted.

IU: individual‐universal.

LIFT: Linking the Interests of Families and Teachers.

LST: Life Skills Training.

MI: motivational intervention.

N/S: not stated.

PATHS: Promoting Alternative Thinking Strategies.

ProSAAF: Protecting Stronger African American Families.

ROAD: Reducing Offenses by Adolescent Drivers.

SAAF: Stronger African American Families.

SD: standard deviation.

SDC: Social Development Curriculum.

ST: school‐targeted.

STRIVE: Support to Reunite, Involve and Value Each Other.

SU: school‐universal.

TTM: transtheoretical model.

Countries

Most studies were conducted in the USA (n = 55; 79%); the remainder were conducted in Europe (n = 8), Canada (n = 2), Australia (n = 1), Hong Kong (n = 1), India (n = 1), Spain and Mexico (n = 1), and South Africa (n = 1). Based on the World Bank classification of countries by income, most studies were conducted in high‐income countries (n = 67; 96%), one (1.4%) was conducted in a lower‐middle‐income country, one (1.4%) was a joint study including an upper‐middle‐income country and a high‐income country, and one (1.4%) was conducted in an upper‐middle‐income country.

Study types

Of the 70 included studies, 12 (17%) provided targeted individual‐level interventions, and four (6%) used universal individual‐level interventions; 17 interventions (24%) were targeted to families, three (4%) were universal family‐level interventions, six (9%) were targeted in school‐based studies, and most (n = 28; 40%) explored use of universal school‐based studies.

Most individual‐level interventions were mentoring or motivational interventions targeting young people at risk (e.g. of antisocial or criminal behaviour, of alcohol‐related injury or harm). Researchers recruited participants from community (Minnis 2014), clinic (Bernstein 2010a; Cunningham 2012; Johnson 2015; Monti 1999; Nirenberg 2013; Walker 2002), and criminal justice settings (Freidman 2002; Freudenberg 2010; Redding 2015), and via relevant service providers, agencies, or charities (Berry 2009; Dolan 2010; Tierney 1995), or through schools (Bodin 2011; Lana 2014; Wagner 2014). Family‐level interventions were targeted to particular families on the basis of a variety of factors (e.g. ethnicity (African‐American (Beach 2016; Brody 2012; Murry 2014) or Mexican‐American (Pantin 2009; Sanchez 2007)); some populations included parents who were being treated for substance misuse (Catalano 1999), others included young mothers (Kitzman 2010; Olds 1998), and others included those living in deprived communities (e.g. targeting those in public housing) (Li 2002; Schwinn 2014). School‐based interventions included whole‐school approaches as well as those that were focused towards individuals of a particular age or age range, and some were implemented over multiple school years. School‐based interventions that were targeted in their focus utilised procedures such as screening on the basis of ratings of aggressive and/or disruptive behaviours (Conduct Problems Prevention Research Group 2010; Lochman 2003a; Lochman 2004a), criteria of high‐risk status (e.g. through truancy, low grade point average (GPA), disciplinary action, or referral by a teacher (Sanchez 2007; Shetgiri 2011)), or low socioeconomic status and low levels of parental education (Schweinhart 1993).

Several studies included multiple components, but additional components tended to be implemented alongside a primary component directed to a particular group (e.g. parental involvement with homework, parental leaflets in addition to a school‐based curriculum targeted to secondary school students). Most interventions were based on education, mentoring, and/or behavioural approaches to risk reduction, but two interventions involved the provision of financial support or financial incentives (Minnis 2014; Morris 2003). One intervention offered earnings supplements to single parents who left welfare for full‐time employment (Morris 2003), and the other offered small cash payments upon completion of activities such as receiving job training or education, or visiting a reproductive health clinic (Minnis 2014).

Seventeen studies were conducted and analysed as cluster RCTs (Beets 2009; Bond 2004; Conduct Problems Prevention Research Group 2010; DeGarmo 2009; Flay 2004a; Gottfredson 2010; Griffin 2006; Kellam 2014; Li 2011; Melnyk 2013; Nader 1999; O'Neill 2011; Piper 2000a; Sanchez 2007; Saraf 2015; Walter 1989; Wolfe 2012).

Age range of study participants

Interventions generally targeted children and young people across a one‐ or two‐year age range. Most interventions were implemented when children were between 9 and 14 years of age.

Among individual‐level interventions, mentoring‐style interventions tended to include young people of a range of ages in mid to late adolescence, with two mentoring interventions including young people aged 10 to 16 (Dolan 2010; Tierney 1995), and another including those aged 13 to 18 (Bodin 2011). Those targeting alcohol use and related risks tended to target those in later adolescence, reflecting levels of engagement in these risk behaviours. Among family‐level interventions, the age range of participants was variable. Broadly, four interventions targeted those across early childhood to mid‐adolescence (Averdijk 2016; Beach 2016; Catalano 1999; Morris 2003), two targeted children early in the life course (up to two years of age) in the Nurse Family Partnership (Kitzman 2010; Olds 1998), six targeted those in early adolescence (nine to 12 years of age) (Bonds 2010; Connell 2007; Gonzales 2012; Kim 2011; Murry 2014; Schwinn 2014), and eight targeted those in mid‐adolescence (13 to 16 years of age) (Brody 2012; Estrada 2015; Haggerty 2007a; Jalling 2016; Jalling 2016b; Li 2002; Milburn 2012; Pantin 2009), demonstrating a focus on primary prevention by targeting interventions at relatively earlier stages of adolescence before the rise in engagement in multiple risk behaviours.

Similarly, among school‐based interventions, although variable with regard to the age of participants, most were implemented before age 16, again demonstrating a focus on primary prevention. A small number were implemented before or during primary school, with one provided during preschool (Schweinhart 1993), three spanning kindergarten through to twelfth grade (Averdijk 2016b; Beets 2009; Conduct Problems Prevention Research Group 2010), and three targeting those aged 6 to 8 years (Ialongo 1999; Kellam 2014) (note that the Ialongo 1999 publications describe two interventions). Eighteen interventions targeted children aged eight to 14, six targeted those aged 13 to 16 (Bond 2004; Johnson 2015; Melnyk 2013; Sanchez 2007; Shetgiri 2011; Wolfe 2012), and two targeted those 13 to 19 years of age (D'Amico 2002a; Fearnow‐Kenney 2003).

Intervention duration

The duration of interventions was variable, both overall and by study type. Twenty‐two studies were less than three months in duration (31% of all studies), nine lasted three to six months (13%), 15 took place over a six‐ to 12‐month period (21%), and 24 continued for longer than one year (34%). The latter were mostly school‐based interventions, which were provided over successive school years, although the total time period of intervention, for instance in terms of total hours of exposure/lessons, would have been markedly less.

Individual‐level interventions were generally shorter, with all 12 targeted interventions of less than six months' duration. Targeted family‐level interventions, in contrast, were mixed, with seven of 17 (41%) less than three months in duration and seven of 17 (41%) over six months in duration, three of which took place over several years (the Nurse Family Partnership and the Self‐Sufficiency Project: Kitzman 2010; Morris 2003; Olds 1998). As mentioned above, most school‐based interventions were provided over a period longer than one year, with four of six targeted interventions over 12 months in duration, one of which was implemented over a 10‐year period (Conduct Problems Prevention Research Group 2010), and none with duration less than six months. Sixteen of the 28 (57%) universal school‐based interventions took place over a period longer than 12 months, several of which were provided for a total period of at least three school years (Beets 2009; Bush 1989; Flay 2004a; Griffin 2006; Griffin 2009; Li 2011; Nader 1999; Piper 2000a; Shek 2011; Simons‐Morton 2005; Walter 1989). Two studies were implemented for less than three months (D'Amico 2002a; DeGarmo 2009).

Alongside duration, the variable intensity of interventions should be noted. For instance, several individual‐focused interventions, particularly those involving motivational interviewing, were short in duration and were characterised by lower intensity compared to individual‐focused mentoring interventions, for instance, when participants meet their mentor weekly over the course of a year. Similarly, certain family‐based interventions involved, for example, five to nine weekly parent and child‐focused sessions, joint group sessions, and home visits (Beach 2016; Brody 2012; Catalano 1999; Gottfredson 2010; Kim 2011; Murry 2014; Pantin 2009), but others involved up to two years of nurse home visitation to infants up to the age of two (Kitzman 2010; Olds 1998); in contrast, another intervention involved a low‐intensity single‐session consisting of video and role‐play (Li 2002). In relation to school‐based programmes, studies included those with multiple components (e.g. family, curriculum, school policy components) provided over one or more school years (Beets 2009; Conduct Problems Prevention Research Group 2010; Flay 2004a; Johnson 2015; Li 2011; Nader 1999), others included classroom sessions provided over one or more school years (Fearnow‐Kenney 2003; O'Neill 2011; Perry 2003a; Shek 2011; Wolfe 2012), and still others applied a whole‐school approach (Bond 2004), demonstrating how the intensity of different interventions can vary between studies. We have provided further details for each study in Table 3.

Post‐intervention follow‐up

Among all studies, the duration of follow‐up was relatively evenly distributed between those that provided a post‐test follow‐up (n = 17; 24% of studies) (i.e. at completion of the intervention), those that followed participants for up to six months following the intervention (n = 14; 20%), those that followed participants for six to 12 months (n = 16; 23% of studies), and those that followed participants for over one year (n = 23; 33%). Seventeen studies conducted follow‐up after at least five years post intervention, the largest number of which were family‐targeted studies (n = 6 studies) (Bonds 2010; Gonzales 2012; Kitzman 2010; Murry 2014; Olds 1998; Pantin 2009) and studies providing universal school interventions (n = 7 studies) (Averdijk 2016b; DeGarmo 2009; Griffin 2006; Ialongo 1999; Kellam 2014; Nader 1999) (note that Ialongo 1999 describes two interventions).

Among the 16 individual‐level interventions, none conducted follow‐up over one year, although seven studies conducted six‐ to 12‐month follow‐ups. In contrast, all studies using universal family‐level interventions provided longer‐term follow‐up (over 12 months post intervention) (Averdijk 2016; Connell 2007; Haggerty 2007a), and eight of the 17 studies examining targeted family‐level interventions (47% of these studies) provided longer‐term follow‐up (> 12 months) (Bonds 2010; Brody 2012; Estrada 2015; Gonzales 2012; Kim 2011; Kitzman 2010; Murry 2014; Olds 1998). Two school‐level intervention studies provided longer‐term follow up (> 5 years) (Conduct Problems Prevention Research Group 2010; Schweinhart 1993), and three provided follow‐up data at six to 12 months (Lochman 2003a; Lochman 2004a; Sanchez 2007). Lastly, a substantial proportion of studies exploring universal school‐based interventions reported follow‐up data at completion of the intervention (n = 11 of 28 studies; 39%). Five studies (18%) reported short‐term follow‐up data at less than six months, two reported data at up to 12 months, and 10 (36%) reported longer‐term follow‐up data (> 12 months).

Theoretical framework

Most studies (n = 53 of 70 studies; 76%) reported that theory informed development of the intervention; some used more than one theory, and 17 (24%) omitted mention of a theoretical model. Eligible studies used 33 different theoretical models. Social learning theory informed 15 studies, with social‐cognitive/cognitive‐behavioural theories informing seven additional interventions. Life skills training, the social development model, and social field theory each informed the development of three interventions. Remaining theories informed one or two of the interventions and included the theory of triadic influence, the transtheoretical model, the Health Belief Model, theories of child development, problem behaviour theory, the Health Promoting Schools framework, and the life course model. Notably, some interventions were based on multiple theories. Nevertheless, few interventions provided detail about how programmes were informed by, or incorporated, such theories.

Intervention focus

On average, studies aimed to address four of the primary behaviours (range two to five). Five studies aimed to address five of the primary behaviours (Bond 2004; Haggerty 2007a; Li 2002; McNeal 2004; Pantin 2009; Piper 2000a; Shek 2011; Wolfe 2012). Most studies targeted alcohol use (n = 55), drug use (n = 53), and/or antisocial behaviour (n = 53) alongside other behaviours, followed by tobacco use (n = 42). In terms of combinations of behaviours addressed, most studies targeted substance use together with antisocial behaviour (n = 49; 70%). Relatively fewer studies addressed sexual risk behaviour (n = 21); 20 interventions in these studies also targeted tobacco, alcohol, and/or drug use, with 16 of those (76%) simultaneously addressing antisocial behaviour. Physical activity and nutrition were targeted by fewer studies again (n = 9, n = 11, respectively), and no studies targeted self‐harm or gambling alongside other behaviours (although we note that interventions that aimed to address only physical inactivity and nutrition were excluded from this review (see Types of interventions)). Education/attainment and mental health were targeted by 19 and 17 studies, respectively, alongside other risk behaviours. We have provided further details regarding the numbers and range of behaviours targeted by different interventions, by study type, in Table 4 and Table 5.

Open in table viewer
Table 4. Number of studies targeting different behaviours by study type

N studies

N outcomes targeted (average)

Primary behaviours

Secondary behaviours

Tobacco

Alcohol

Drugs

ASB

Self‐harm

Gambling

Vehicle risk

Sexual risk

Physical activity

Nutrition

Education and attainment

Mental illness

Individual Targeted

12

4

4

10

7

9

3

3

5

2

Individual Universal

3

4

2

4

2

2

1

2

2

1

1

Family Targeted

17

4

8

13

16

14

7

1

1

4

7

Family Universal

3

4

3

3

3

3

1

School Targeted

6

3

2

3

4

6

3

2

School Universal

28

4

23

22

21

19

1

9

6

8

6

5

TOTAL

70

42

55

53

53

0

0

4

21

9

11

19

17

Open in table viewer
Table 5. Behaviours targeted by interventions by study type

Author, date

Study type

Tobacco use

Alcohol use

Illicit drug use

Antisocial behaviour

Vehicle risk

Sexual risk behaviour

Physical inactivity

Nutrition

Mental illness

Education & attainment

Bernstein 2010

IT

X

X

X

X

Berry 2009

IT

X

X

X

X

Bodin 2011

IT

X

X

X

X

X

Cunningham 2012

IT

X

X

Dolan 2010

IT

X

X

X

X

X

X

Freudenberg 2010

IT

X

X

X

X

Friedman 2002

IT

X

X

X

X

Monti 1999

IT

X

X

Nirenberg 2013

IT

X

X

Redding 2015

IT

X

X

Tierney 1995

IT

X

X

X

X

X

Wagner 2014

IT

X

X

X

Johnson 2015

IU

X

X

X

Lana 2014

IU

X

X

X

X

Minnis 2014

IU

X

X

X

X

X

Walker 2002

IU

X

X

X

X

X

Beach 2016

FT

X

X

X

X

Bonds 2010

FT

X

X

X

X

X

Brody 2012

FT

X

X

X

Catalano 1999

FT

X

X

X

X

Estrada 2016

FT

X

X

X

X

Gonzales

FT

X

X

X

X

X

X

Jalling 2016

FT

X

X

X

Jalling 2016b

FT

X

X

X

Kim 2011

FT

X

X

X

X

X

Kitzman 2010

FT

Indirect

X

X

X

X

X

Li 2002

FT

X

X

X

X

X

Milburn 2012

FT

X

X

X

Morris 2003

FT

X

X

X

X

X

X

Murry 2014

FT

X

X

Olds 1998

FT

Indirect

X

X

X

X

X

Pantin 2009

FT

X

X

X

X

X

Schwinn 2014

FT

X

X

X

Averdijk 2016

FU

X

X

X

X

Connell 2007

FU

X

X

X

X

Haggerty 2007

FU

X

X

X

X

X

CPRG 2014

ST

Indirect

Indirect

Indirect

X

Indirect

X

X

Lochman 2003

ST

X

X

X

X

Lochman 2004

ST

X

X

X

Sanchez 2007

ST

X

X

X

X

Schweinhart 1993

ST

Indirect

Indirect

Indirect

X

X

Shetgiri 2011

ST

X

X

X

X

Beets 2009

SU

X

X

X

X

X

Bond 2004

SU

X

X

X

X

X

X

Bush 1989

SU

X

X

X

D'Amico 2002

SU

X

X

X

DeGarmo 2009

SU

X

X

X

X

Fearnow‐Kenney 2003

SU

X

X

X

X

Flay 2004

SU

X

X

X

Gottfredson 2010

SU

X

X

X

X

Griffin 2006

SU

X

X

X

Griffin 2009

SU

X

X

X

X

Ialongo 1999a

SU

X

X

X

X

X

X

Ialongo 1999b

SU

X

X

X

X

X

X

Kellam

SU

X

X

X

X

Indirect

X

Lewis 2013

SU

X

X

X

X

X

X

LoSciuto 1999

SU

X

X

X

X

X

Matthews 2016

SU

X

X

McNeal 2004

SU

X

X

X

X

X

Melnyk 2013

SU

X

X

X

X

X

X

Nader 1999

SU

X

X

X

O'Neill 2011

SU

X

X

X

X

X

Perry 2003

SU

X

X

X

X

Piper 2000

SU

X

X

X

X

X

Saraf 2015

SU

X

X

X

Shek 2011

SU

X

X

X

X

X

Simons‐Morton 2005

SU

X

X

X

Walter 1989

SU

X

X

X

Wolfe 2009

SU

X

X

X

X

X

TOTAL

42

55

53

53

4

21

9

11

17

19

FT: family‐targeted.

FU: family‐universal.

IT: individual‐targeted.

IU: individual‐universal.

ST: school‐targeted.

SU: school‐universal.

Process data

A large proportion of eligible studies reported some process evaluation data (n = 47 of 70; 67%). Many lacked detail about how the intervention was implemented. Among studies that reported process data, 47 (100%) provided quantitative data and 14 (30%) qualitative data, with 47 studies (100%) providing details about fidelity or intensity of the intervention. Nevertheless, few studies conducted in‐depth analyses of whether the intervention was delivered as intended, any deviations from study protocols and manuals, and/or mechanisms by which the intervention had an effect on behaviour. This is reflected in the comparative lack of qualitative data, which may complement quantitative data in evaluating fidelity and contextual factors such as barriers to delivery and/or engagement and uptake.

Economic data

Only four studies reported economic data: one targeted a school‐level intervention with 40‐year follow‐up (Schweinhart 1993), two separate studies examined the family‐level Nurse Family Partnership intervention (Kitzman 2010; Olds 1998), and one was a universal school‐level study (Wolfe 2012).

Equity

All studies provided data regarding age of participants at baseline, and most provided a breakdown by gender (63 studies; 90%), with one intervention provided to female participants only. Fifty‐eight studies (83%) also reported the proportions of participants in different ethnic groups, although three studies included individuals of a particular race/ethnicity. However, only 39 studies (56%) provided data regarding socioeconomic status of participants. Those that did highlighted factors such as parental income, education, or occupation; the proportion of participants residing in public housing; or the proportion of participants receiving free school meals. Few reports presented findings according to ethnic group (two studies), gender (nine studies), or socioeconomic status (two studies). In addition, only two studies were reported from low‐ or middle‐income countries; this limits the generalisability of our findings to these settings. Most papers reported sources of funding for studies; only six failed to provide details. Most studies were supported by government agencies or charities, with one study supported by a foundation or a private source alone or in conjunction with other sources. We have provided further details for individual studies in the Characteristics of included studies table. Further analyses have been performed by our research group to examine equity and effects of the interventions included in this review (Tinner 2018).

Adverse events and outcomes

Seven studies reported adverse events (two of which ‐ Jalling 2016, Jalling 2016b ‐ were reported from a three‐arm trial) (Conduct Problems Prevention Research Group 2010; Jalling 2016; Jalling 2016b; Morris 2003; Nirenberg 2013; Sanchez 2007). First, in the Self‐Sufficiency Project (SSP) (Morris 2003), which involved provision of financial support for parents who left welfare for employment, mothers in the intervention group reported lower average school achievement for their children than for children in the control group. However, results showed no significant programme control differences in the proportion of parents reporting that their children were performing below average in school. Nearly 19% of children in the programme group said they were below average in at least one subject in school, compared to about 14% in the control group (P < 0.05). The SSP was also associated with increased frequency of minor delinquent activity for older children (aged 15 to 18; P < 0.05), but this was not the case for their younger peers (aged 12 to 14); and the programme led to increased use of tobacco, alcohol, and drugs by approximately four percentage points (Morris 2003).

For the Fast Track study (Conduct Problems Prevention Research Group 2010), the death rate at age 25‐year follow‐up was 2.5% for the intervention group and 1.6% for the control group, and the incarceration rate for the intervention group was 6.3% compared to 5% for the control group (approximately 19 years following intervention implementation), although study authors noted that these findings were not statistically significant. Sanchez 2007 also reported that Reconnecting Youth was associated with worse outcomes in the intervention group for conventional peer bonding, high‐risk peer bonding, and prosocial weekend activities compared to the control group.

Nirenberg 2013 reported that participants in the intervention arm reported more speeding and distracted driving and higher levels of hazardous drinking compared to those in the control group. Study authors also reported reduced odds of having at least one high‐risk driving or alcohol police charge in the intervention arms and no differences between study arms in relation to dangerous driving or alcohol, drugs, and driving. Study authors postulated (1) that this finding may have resulted from discussion of behaviours among participants during intervention sessions, which might have led to greater sensitivity in reporting risky driving behaviours (response shift bias), or (2) that those in the intervention group might have reported the behaviour more readily.

Lastly, investigators reported that the odds of illicit drug use were higher in the two intervention arms (Comet 12‐18 and ParentSteps) of a three‐arm trial (odds of any drug use: Comet 12‐18: OR 3.52, 95% CI 1.23 to 10.10; ParentSteps: OR 3.23, 95% CI 3.23 to 9.08). Study authors suggested that this may be due to measurement error, in that illicit drug use (in the past six months) increased between baseline and follow‐up among intervention groups but decreased in the control group, whereas evidence suggests that illicit drug use increases during adolescence in Sweden. Study authors also highlighted that the small sample size may have given rise to uncertainty in the effect estimate.

Missing data

We contacted the authors of 34 studies to obtain additional study or outcome data. We received responses from the authors of 23 studies; data from four studies were not available owing to time elapsed since the studies were conducted (D'Amico 2002a; Griffin 2009; Ialongo 1999; Ialongo 1999b); we received additional study data from 11 studies (Conduct Problems Prevention Research Group 2010; Gottfredson 2010; Li 2011; Lochman 2003a; Lochman 2004a; McNeal 2004; Milburn 2012; O'Neill 2011; Pantin 2009; Shek 2011; Shetgiri 2011).

Ongoing studies

From updated searches in November 2016, we identified nine ongoing studies that may be eligible for inclusion in this review; we have listed these in the Characteristics of ongoing studies table. Four were school‐based, one was school‐ and community‐based, one was based in a further education setting, one at a university, one in the community, and one in general practice. Two studies were conducted in the UK, two in Australia, two in the USA, one in Italy, one in Guatemala, and one in Brazil. All were universal. Six studies addressed tobacco use, alcohol use, physical inactivity, and nutrition; one addressed substance use and sexual risk behaviour; and one addressed tobacco use, alcohol use, sexual risk behaviour, and violence. One was an online intervention study, and another a video game intervention study (for details of each study, see Characteristics of ongoing studies).

Excluded studies

We have summarised in the Characteristics of excluded studies table studies that did not meet eligibility criteria, such as addressing two or more eligible behaviours or being randomised controlled trials.

Risk of bias in included studies

We have provided in Figure 3 a summary of the overall risk of bias assessment for included studies, and we have given detailed summaries of study‐specific judgements by study in Figure 4. Overall, a large proportion of studies were at high or unclear risk of bias, and many studies lacked sufficient detail to permit a judgement around risk of different forms of bias.


Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.


Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

Most studies lacked detail with regard to reporting around the method of randomisation and the use and/or method of allocation concealment; thus we classed these studies as having unclear risk of bias. In particular, researchers rarely reported concealment of allocation and/or the method of concealment. Thus, we judged only 19 studies as having low risk of selection bias on the basis of random sequence generation (Bernstein 2010a; Bodin 2011; Bonds 2010; Cunningham 2012; DeGarmo 2009; Gonzales 2012; Gottfredson 2010; Ialongo 1999; Ialongo 1999b; Kim 2011; Lana 2014; Li 2002; Li 2011; Milburn 2012; Olds 1998; Saraf 2015; Shetgiri 2011; Wagner 2014; Wolfe 2012).

Blinding

We classed most studies as having high risk of performance bias owing to lack of blinding of participants. We judged only two interventions to present low risk of detection bias (Brody 2012; Gonzales 2012). Without adequate blinding of participants and personnel, it is possible that outcome measurements, made mostly via self‐report of the frequency or extent of engagement in risk behaviours, could have been influenced, particularly because data collection occurred in the school context for many of the included studies. It should be noted, however, that it is generally not possible to blind interventions of the kind included in this review, and so assessments of high risk of bias are somewhat inevitable.

Incomplete outcome data

Judgements around risk of attrition bias among studies were variable, with 20 studies judged to be at high risk of attrition bias (Bernstein 2010a; Berry 2009; Bush 1989; D'Amico 2002a; Flay 2004a; Freidman 2002; Griffin 2006; Lana 2014; Li 2011; Lochman 2003a; Lochman 2004a; Milburn 2012; O'Neill 2011; Piper 2000a; Saraf 2015; Schwinn 2014; Shek 2011; Simons‐Morton 2005; Walker 2002; Walter 1989) owing to high levels of attrition overall (> 30%), differential attrition, and/or lack of imputation to manage missing data. We judged 33 studies to be at low risk of attrition bias owing to low attrition in the study overall, clear descriptions related to differential attrition, and/or imputation (Bodin 2011; Bond 2004; Bonds 2010; Brody 2012; Catalano 1999; Connell 2007; Cunningham 2012; DeGarmo 2009; Dolan 2010; Gonzales 2012; Gottfredson 2010; Griffin 2009; Haggerty 2007a; Ialongo 1999; Ialongo 1999b; Kim 2011; Kitzman 2010; LoSciuto 1999; McNeal 2004; Monti 1999; Murry 2014; Nader 1999; Nirenberg 2013; Olds 1998; Pantin 2009; Perry 2003a; Redding 2015; Sanchez 2007; Schweinhart 1993; Shetgiri 2011; Tierney 1995; Wagner 2014; Wolfe 2012). Sixteen studies conducted imputation or used an appropriate method to account for missing data (Bodin 2011; Bonds 2010; Conduct Problems Prevention Research Group 2010; Connell 2007; DeGarmo 2009; Dolan 2010; Gonzales 2012; Gottfredson 2010; Haggerty 2007a; Kellam 2014; Kim 2011; Li 2011; Murry 2014; O'Neill 2011; Redding 2015; Wagner 2014).

Selective reporting

Only two studies had a published and accessible protocol for the intervention (Lana 2014; Melnyk 2013); in many cases, the extent of selective reporting bias was unclear because we were not able to judge whether all expected outcomes had been reported according to prespecified protocols. We judged seven studies to be at high risk of selective reporting bias owing to absence of data for expected outcomes or time points (Griffin 2006; Melnyk 2013; Murry 2014; Sanchez 2007; Schwinn 2014; Shek 2011; Walker 2002), with some studies lacking clarity in the presentation of data.

Other potential sources of bias

Other sources of bias identified included the possibility of contamination between study arms, for instance, when students were randomised within schools at the level of the individual or the classroom. If this were the case, the effect of the intervention would be diluted. When we encountered other issues, such as baseline imbalance, we also noted this in this domain. Studies with long‐term follow‐up measured and reported multiple outcomes; thus, there remains a possibility that positive findings might have been identified by chance, particularly when such interventions had broad aims and were implemented early in the life course (e.g. Schweinhart 1993), such that the full range of outcomes could not be clearly prespecified.

'Summary of findings'

Among the outcomes listed in the 'Summary of findings' tables (summary of findings Table for the main comparison), we judged that much of the evidence related to tobacco use, alcohol use, cannabis use, illicit drug use, sexual risk behaviour, and physical activity was of moderate or low quality, primary owing to risk of bias and inconsistency in findings between studies. For antisocial behaviour, we judged the evidence to be of very low quality owing to high risk of bias across domains, as well as identification of heterogeneity between studies and funnel plot asymmetry. We judged that the odds ratios observed among universal school‐level interventions for tobacco, alcohol, and drug use; antisocial behaviour; and sexual risk behaviour, which ranged between 0.72 and 0.83, and the size of the effect observed for physical activity reflected reductions in engagement in risk behaviours that could be of potential public health importance at the population level.

Funnel plots

Within each study type subgroup, we identified too few studies (n < 10) to examine possible publication bias or small‐study effects, with the exception of antisocial behaviour (universal school‐based interventions) (Figure 5). The funnel plot for universal school‐based interventions targeting antisocial behaviour shows an under‐representation of small studies reporting negative findings, suggesting possible small‐study effects or possible reporting biases leading to overestimation of effectiveness (Sterne 2011). This was also evident in funnel plots that included all studies of each study type addressing antisocial behaviour (Figure 6; Figure 7; Figure 8).


Funnel plot. Outcome 7: antisocial behaviour and offending (short‐term): universal school‐based interventions.

Funnel plot. Outcome 7: antisocial behaviour and offending (short‐term): universal school‐based interventions.


Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.1: antisocial behaviour and offending ‐ any (short‐term).

Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.1: antisocial behaviour and offending ‐ any (short‐term).


Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.2: violent offences (short‐term).

Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.2: violent offences (short‐term).


Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.3: school or general delinquency (short‐term).

Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.3: school or general delinquency (short‐term).

For tobacco use, alcohol use, cannabis use, illicit drug use, and sexual risk behaviour, we plotted funnel plots for each intervention type addressing these outcomes (Figure 9; Figure 10; Figure 11; Figure 12; Figure 13). The cylindrical appearance of plots for alcohol use and cannabis suggests that high levels of heterogeneity may be present, which would be expected when different study types are combined. Lastly, the funnel plots of studies targeting illicit drug use and sexual risk behaviour are asymmetrical, suggesting that small‐study bias may be present, leading to the possibility of an overestimation of effect, given the under‐representation of small studies with negative findings (Sterne 2011).


Funnel plot of comparison. Outcome 1: tobacco. Outcome 1.1: tobacco use (short‐term).

Funnel plot of comparison. Outcome 1: tobacco. Outcome 1.1: tobacco use (short‐term).


Funnel plot of comparison. Outcome 2: alcohol. Outcome 2.1: alcohol use (short‐term).

Funnel plot of comparison. Outcome 2: alcohol. Outcome 2.1: alcohol use (short‐term).


Funnel plot of comparison. Outcome 4: cannabis use. Outcome 4.1: cannabis use (short‐term).

Funnel plot of comparison. Outcome 4: cannabis use. Outcome 4.1: cannabis use (short‐term).


Funnel plot of comparison. Outcome 5: illicit drug use. Outcome 5.1: Illicit drug use (short‐term).

Funnel plot of comparison. Outcome 5: illicit drug use. Outcome 5.1: Illicit drug use (short‐term).


Funnel plot of comparison. Outcome 9: sexual risk behaviours. Outcome 9.1: sexual risk behaviour (short‐term).

Funnel plot of comparison. Outcome 9: sexual risk behaviours. Outcome 9.1: sexual risk behaviour (short‐term).

Effects of interventions

See: Summary of findings for the main comparison ; Summary of findings 2 ; Summary of findings 3 ; Summary of findings 4 ; Summary of findings 5

Although the studies included in this review addressed two or more of the behaviours of interest, we will describe effects of the interventions on each behaviour, separately, below, because this is how they were reported in the study papers. In future research, we will explore the possibility of examining effects of interventions on multiple behaviours. We note that we were not able to address the secondary objectives of the review owing to a lack of available data.

Primary outcomes

Tobacco use

Forty‐two studies (60%) targeted tobacco use among young people; most of these studies provided universal school‐based interventions (n = 23). We provide a further breakdown of these studies by intervention type and name in Table 4 and Table 5. Thirty‐seven of these studies concomitantly targeted use of another substance, five targeted tobacco use alongside physical activity and nutrition as a cardiovascular prevention intervention (Bush 1989; Nader 1999; O'Neill 2011; Saraf 2015; Walter 1989), and one targeted tobacco use and sexual risk behaviour (Redding 2015). Three of these 38 studies targeted tobacco use via proposed risk factors for later substance use (Ialongo 1999; Ialongo 1999b; Kellam 2014), and a further four studies targeted tobacco use through hypothesised indirect mechanisms (Conduct Problems Prevention Research Group 2010; Kitzman 2010; Olds 1998; Schweinhart 1993). Of studies targeting tobacco use, one study was conducted in Sweden (Bodin 2011), one in Ireland (Dolan 2010), two in Switzerland (Averdijk 2016; Averdijk 2016b), one in India (Saraf 2015), one in Hong Kong (Shek 2011), one in Spain/Mexico (Lana 2014), one in the UK (Walker 2002), two in Canada (Morris 2003; Wolfe 2012), and one in Australia (Bond 2004); the remainder were conducted in the USA.

Fifteen studies reported data up to 12 months following completion of the intervention that could be included in the meta‐analyses for tobacco use (Analysis 1.1); seven studies reported longer‐term tobacco outcomes (over 12 months following the end of the intervention, extending up to 12 and 22 years of follow‐up (Kellam 2014 and Schweinhart 1993, respectively)). Four studies reported tobacco outcomes as part of a substance use composite score or measure (Kitzman 2010; LoSciuto 1999; Pantin 2009; Shek 2011). We have recorded in Table 2 outcomes from the 12 studies that could not be included (outcomes not included in meta‐analysis). Two studies did not report data regarding tobacco use (Dolan 2010; Wolfe 2012).

Measures

Among those included in the meta‐analysis of findings up to 12 months, studies reported outcome data in terms of frequency data (e.g. smoking in the past 30 days) or ever having smoked, depending on which was more appropriate for the age at which data were collected. Among those with longer‐term follow‐up, outcome measures included smoking initiation (which was relevant when studies were initiated with very young children, e.g. in primary or elementary school) and smoking frequency.

Effectiveness over the short term

Analysis 1.1 presents results of the meta‐analysis for tobacco use by intervention type. We identified that individual‐level universal or targeted interventions may have little or no effect in relation to tobacco use at up to 12 months' follow‐up (universal: odds ratio (OR) 1.03, 95% confidence interval (CI) 0.32 to 3.27; P = 0.97; n = 2 studies; 1549 participants; I² = 38%; moderate‐quality evidence; targeted: OR 0.98, 95% CI 0.35 to 2.73; P = 0.97; n = 2 studies; 1549 participants; I² = 72%; low‐quality evidence), although each meta‐analysis included only two studies (universal: Lana 2014; Walker 2002; targeted: Bodin 2011; Redding 2015).

Among family‐based interventions (Catalano 1999; Li 2002), data show uncertainty of the effect in relation to tobacco use (OR 0.78, 95% CI 0.40 to 1.53; P = 0.47; n = 2 studies; 313 participants; I² = 0%; moderate‐quality evidence) (Analysis 1.1).

For universal school‐based interventions (Beets 2009; Bond 2004; Fearnow‐Kenney 2003; Gottfredson 2010; Griffin 2009; Li 2002; Nader 1999; O'Neill 2011; Simons‐Morton 2005), moderate‐quality evidence shows that such interventions, on average, reduced smoking among young people (OR 0.77, 95% CI 0.60 to 0.97; P = 0.03; n = 9 studies; 15,354 participants), although heterogeneity was moderate (I² = 57%). We considered the size of this effect to represent a potentially important change in engagement in this risk behaviour at the population level. In particular, the Positive Action programme (Beets 2009), the Gatehouse Study (Bond 2004), the Michigan Model for Health (O'Neill 2011), and Going Places reported positive findings (Simons‐Morton 2005). Sensitivity analysis using a different intracluster correlation coefficient (ICC) did not change this finding, and analysis of only those reporting dichotomous outcomes gave a similar result, although this was not the case for two studies reporting continuous data (Table 6).

Open in table viewer
Table 6. Sensitivity analysis

Outcome

Classification

Subgroup

N studies

N intervention

N control

Estimate, 95% CI

I²

Intracluster correlation coefficient (ICC): use of lowest ICC (studies with short‐term follow‐up only)

Tobacco use

Individual Targeted

Highest ICC

2

280

241

1.28, 0.75 to 2.19

0

Lowest ICC

2

280

241

1.28, 0.75 to 2.19

0

Individual Universal

Highest ICC

2

925

624

1.03, 0.32 to 3.27

38

Lowest ICC

2

925

624

1.03, 0.32 to 3.27

38

Family Targeted

Highest ICC

2

160

153

0.78, 0.40 to 1.53

0

Lowest ICC

2

160

153

0.78, 0.40 to 1.53

0

School Universal

Highest ICC

9

8365

6989

0.77, 0.60 to 0.97

57

Lowest ICC

9

8365

6989

0.76, 0.59 to 0.97

65

Alcohol use

Individual Targeted

Highest ICC

4

1204

840

1.02, 0.79 to 1.30

48

Lowest ICC

4

1204

840

1.02, 0.79 to 1.30

48

Individual Universal

Highest ICC

4

1105

806

0.80, 0.58 to 1.11

0

Lowest ICC

4

1105

806

0.80, 0.58 to 1.11

0

Family Targeted

Highest ICC

3

212

205

0.83, 0.47 to 1.46

29

Lowest ICC

3

212

205

0.83, 0.47 to 1.46

29

School Targeted

Highest ICC

1

615

603

1.03, 0.56 to 1.91

‐‐

Lowest ICC

1

615

603

1.03, 0.56 to 1.91

‐‐

School Universal

Highest ICC

8

4382

4369

0.72, 0.56 to 0.92

58

Lowest ICC

8

4382

4369

0.71, 0.55 to 0.91

60

Binge drinking

Individual Targeted

Highest ICC

3

130

120

0.97, 0.68 to 1.37

0

Lowest ICC

3

130

120

0.97, 0.68 to 1.37

0

School Universal

Highest ICC

5

2825

2669

0.66, 0.41 to 1.06

43

Lowest ICC

5

2825

2669

0.66, 0.45 to 0.99

49

Cannabis use

Individual Targeted

Highest ICC

2

67

59

1.10, 0.71 to 1.97

0

Lowest ICC

1

67

59

1.10, 0.71 to 1.97

0

Individual Universal

Highest ICC

2

180

182

0.69, 0.46 to 1.04

0

Lowest ICC

2

79

83

0.69, 0.46 to 1.04

0

Family Targeted

Highest ICC

3

192

188

1.02, 0.52 to 2.02

43

Lowest ICC

3

192

188

1.02, 0.52 to 2.02

43

School Universal

Highest ICC

5

1924

2216

0.79, 0.62 to 1.01

0

Lowest ICC

5

1924

2216

0.77, 0.61 to 0.97

0

Illicit drug use

Individual Targeted

Highest ICC

3

342

296

0.94, 0.71 to 1.25

0

Lowest ICC

3

342

296

0.94, 0.71 to 1.25

0

Family Targeted

Highest ICC

1

33

36

0.74, 0.42 to 1.31

‐‐

Lowest ICC

1

33

36

0.74, 0.42 to 1.31

‐‐

School Targeted

Highest ICC

4

1431

1023

0.75, 0.53 to 1.06

60

Lowest ICC

4

1431

1023

0.75, 0.53 to 1.06

60

School Universal

Highest ICC

5

4745

6313

0.74, 0.55 to 1.00

69

Lowest ICC

5

4715

6313

0.74, 0.54 to 1.03

82

Tobacco, alcohol, and/or illicit drug use

Family Targeted

Highest ICC

2

115

98

0.81, 0.50 to 1.33

0

Lowest ICC

2

115

98

0.81, 0.50 to 1.33

0

School Targeted

Highest ICC

5

244

98

0.55, 0.24 to 1.25

69

Lowest ICC

5

244

98

0.54, 0.24 to 1.21

71

School Universal

Highest ICC

2

2771

4256

1.13, 0.88 to 1.44

0

Lowest ICC

2

2771

4256

1.10, 1.01 to 1.20

0

Antisocial behaviour

Individual Targeted

Highest ICC

4

409

355

1.21, 0.92 to 1.60

19

Lowest ICC

4

409

355

1.21, 0.92 to 1.60

19

Family Targeted

Highest ICC

6

437

335

0.84, 0.57 to 1.24

42

Lowest ICC

6

437

335

0.84, 0.57 to 1.24

42

Family Universal

Highest ICC

1

208

98

0.87, 0.56 to 1.35

0

Lowest ICC

1

208

98

0.87, 0.56 to 1.35

0

School Targeted

Highest ICC

3

815

716

0.78, 0.59 to 1.05

0

Lowest ICC

3

815

716

0.82, 0.68 to 0.99

0

School Universal

Highest ICC

12

9960

10796

0.81, 0.66 to 0.98

66

Lowest ICC

12

9960

10796

0.82, 0.69 to 0.97

67

Sexual risk behaviour

Individual Targeted

Highest ICC

2

266

228

0.73, 0.49 to 1.08

45

Lowest ICC

2

266

228

0.73, 0.49 to 1.08

45

Individual Universal

Highest ICC

1

79

83

0.42, 0.14 to 1.25

‐‐

Lowest ICC

1

79

83

0.42, 0.14 to 1.25

‐‐

Family Targeted

Highest ICC

3

188

183

0.89, 0.55 to 1.44

0

Lowest ICC

3

188

183

0.89, 0.55 to 1.44

0

School Universal

Highest ICC

6

5757

6876

0.83, 0.61 to 1.12

77

Lowest ICC

6

5757

6876

0.83, 0.61 to 1.12

77

Physical activity

Individual Universal

Highest ICC

2

748

782

1.11, 0.74 to 1.67

0

Lowest ICC

2

748

782

1.11, 0.74 to 1.67

0

Family Targeted

Highest ICC

1

31

30

0.72, 0.29 to 1.79

‐‐

Lowest ICC

1

31

30

0.72, 0.29 to 1.79

‐‐

School Universal

Highest ICC

4

3547

2894

1.32, 1.16 to 1.50

0

Lowest ICC

4

3547

2894

1.33, 1.18 to 1.50

0

Nutrition (BMI)

Individual Universal

Highest ICC

1

421

158

0.80, 0.48 to 1.31

‐‐

Lowest ICC

1

421

158

0.80, 0.48 to 1.31

‐‐

School Universal

Highest ICC

3

2901

2116

0.84, 0.60 to 1.19

61

Lowest ICC

3

2901

2116

0.88, 0.62 to 1.23

69

Nutrition (unhealthy diet)

Individual Universal

Highest ICC

2

925

624

0.76, 0.42 to 1.34

51

Lowest ICC

2

925

624

0.76, 0.42 to 1.34

51

School Universal

Highest ICC

3

3608

2833

0.82, 0.64 to 1.06

49

Lowest ICC

3

3608

2833

0.85, 0.66 to 1.09

63

Educational attainment (academic performance)

Individual Targeted

Highest ICC

1

67

59

1.34, 0.71 to 2.52

‐‐

Lowest ICC

1

67

59

1.34, 0.71 to 2.52

‐‐

School Targeted

Highest ICC

3

619

628

0.91, 0.30 to 2.73

84

Lowest ICC

3

619

628

0.91, 0.39 to 2.14

85

School Universal

Highest ICC

3

602

393

0.94, 0.62 to 1.44

0

Lowest ICC

3

602

393

0.95, 0.74 to 1.22

0

Dichotomous vs continuous outcomes (studies with positive findings; 12‐month follow‐up only)

Tobacco use

School Universal

All

9

8365

6989

0.77, 0.60 to 0.97, P = 0.03

57

Dichotomous

7

7581

6275

0.72, 0.52 to 0.99, P = 0.05

60

Continuous

2

784

714

SMD ‐0.01,‐0.40 to 0.37, P = 0.95

84

Alcohol use

School Universal

All

8

4382

4369

0.72, 0.56 to 0.92, P = 0.009

58

Dichotomous

6

3598

3663

0.68, 0.51 to 0.90, P = 0.008

48

Continuous

2

784

706

SMD ‐0.12, ‐0.46 to 0.22; P = 0.49

79

Illicit drug use

School Universal

All

5

4745

6313

0.74, 0.55 to 1.00, P = 0.05

69

Dichotomous

4

2932

2808

0.67, 0.49 to 0.93

62

Continuous

1

1813

3505

SMD 0.06, 0.00 to 0.12, P = 0.04

‐‐

Cannabis use

School Universal

All

5

1924

2216

0.79, 0.62 to 1.01, P = 0.06

0

Dichotomous

4

1832

2130

0.82, 0.64 to 1.06, P = 0.13

0

Continuous

1

92

86

SMD ‐0.29, ‐0.58 to 0.01, P = 0.06

‐‐

Antisocial behaviour

School Universal

All

13

8445

9277

0.79, 0.64 to 0.97, P = 0.02

68

Dichotomous

4

4042

4339

0.55, 0.30 to 1.01, P = 0.06

87

Continuous

9

5708

6255

SMD ‐0.06,

‐0.11 to ‐0.0, P = 0.03

31

Sexual risk behaviour

School Universal

All

6

5757

6876

0.83, 0.61 to 1.12

77

Dichotomous

4

3020

2635

0.71, 0.39 to 1.30, P = 0.27

84

Continuous

2

2737

4241

SMD ‐0.03, ‐0.08 to 0.02, P = 0.2

0

Studies conducted in all countries vs high‐income countries only (studies with up to 12‐month follow‐up; meta‐analyses incorporating relevant data only)

Tobacco use

Individual Universal

All

2

925

624

1.03, 0.32 to 3.27

38

High‐income countries only

1

504

466

0.74, 0.43 to 1.28

‐‐

Alcohol use

Individual Universal

All

4

1105

806

0.80, 0.58 to 1.11

0

High‐income countries only

3

684

648

0.74, 0.54 to 1.06

0

Antisocial behaviour

School Universal

All

13

9960

10796

0.81, 0.66 to 0.98

66

High‐income countries only

12

8445

9277

0.79, 0.64 to 0.97

68

Sexual risk behaviour

School Universal

All

6

5757

6876

0.83, 0.61 to 1.12

77

High‐income countries only

5

5654

6779

0.81, 0.59 to 1.11

80

Physical activity

Individual Universal

All

2

748

782

1.11, 0.74 to 1.67

0

High‐income countries only

1

504

466

1.40, 0.67 to 2.94

‐‐

School Universal

All

4

3547

2894

1.32, 1.16 to 1.50

0

High‐income countries only

3

2533

1834

1.44, 1.20 to 1.74

0

Unhealthy diet

Individual Universal

All

2

925

624

0.76, 0.42 to 1.34

51

High‐income countries only

1

504

466

0.50, 0.23 to 1.08

‐‐

School Universal

All

3

3608

2833

0.82, 0.64 to 1.06

49

High‐income countries only

2

2594

1773

0.95, 0.76 to 1.19

0

CI: confidence interval.

ICC: intracluster correlation coefficient.

Among the studies that could not be included in the meta‐analysis with follow‐up data up to 12 months, one study examining a targeted individual‐level intervention reported positive findings (Tierney 1995). One study providing a targeted family‐level intervention that could not be included in the meta‐analysis reported positive findings (Gonzales 2012), and one study providing a universal family‐level intervention reported null findings (Connell 2007). Five universal school‐level interventions that could not be included in the meta‐analysis showed beneficial effects (Bush 1989; DeGarmo 2009; McNeal 2004; Saraf 2015; Walter 1989), two universal school‐level interventions showed no effect (LoSciuto 1999; Piper 2000a), and another showed a beneficial effect of one of two active intervention arms among boys only (Perry 2003a).

Long‐term effectiveness

Analysis 1.2 presents results of the meta‐analysis for follow‐up data over 12 months. One targeted individual‐level study (n = 397 participants), which reported findings at 18 months' follow‐up (Redding 2015), showed no benefit of the intervention.

Two targeted family‐level intervention studies, representing 1177 participants, provided longer‐term follow‐up data, reporting findings at approximately 24 months in Kim 2011 and at 36 months in Morris 2003. The two studies reported contrasting results, with one caregiver training and skills‐building programme favouring intervention (OR 0.48, 95% CI 0.23 to 0.98; Kim 2011), and the other income/employment support programme reporting an increase in tobacco use among those in the intervention group (OR 1.27, 95% CI 0.96 to 1.68; Morris 2003).

Across three school‐based universal interventions that were implemented in primary school and provided longer‐term follow‐up, evidence shows that such interventions probably have a positive effect (OR 0.60, 95% CI 0.33 to 1.09; P = 0.09; n = 3 studies; 879 participants; I² = 0%) (Ialongo 1999; Ialongo 1999b; Kellam 2014).

Alcohol use

Fifty‐five studies (79%) addressed alcohol use, three of which targeted alcohol use through early risk factors for later alcohol use (Ialongo 1999; Ialongo 1999b; Kellam 2014). Two additional studies addressed alcohol use by targeting early life experiences and thus targeted indirect pathways (Conduct Problems Prevention Research Group 2010; Schweinhart 1993). Forty‐two of 48 studies (88%) also addressed another substance together with antisocial behaviour, and four addressed alcohol use and vehicle‐risk behaviour (Bernstein 2010a; D'Amico 2002a; Monti 1999; Nirenberg 2013). We provide in Table 4 and Table 5 a summary of the studies that aimed to prevent alcohol use by intervention type and study name.

Most studies were conducted in the USA, and two were conducted in the UK (Berry 2009; Walker 2002), three in Sweden (Bodin 2011; Jalling 2016; Jalling 2016b), two in Switzerland (Averdijk 2016; Averdijk 2016b), one in Ireland (Dolan 2010), one in Spain/Mexico (Lana 2014), one in Australia (Bond 2004), one in Hong Kong (Shek 2011), and one in Canada (Wolfe 2012).

We could not include 17 studies in the meta‐analysis (Connell 2007; Cunningham 2012; D'Amico 2002a; DeGarmo 2009; Estrada 2015; Freidman 2002; Jalling 2016; Jalling 2016b; McNeal 2004; Monti 1999; Murry 2014; Nirenberg 2013; Olds 1998; Perry 2003a; Piper 2000a; Shetgiri 2011; Tierney 1995). See Table 2 for a summary of these data. Nine studies provided composite measures of substance use (Averdijk 2016; Averdijk 2016b; Beach 2016; Kitzman 2010; Lochman 2003a; Lochman 2004a; LoSciuto 1999; Shek 2011; Wolfe 2012).

Eleven studies measured binge drinking or drunkenness (Beets 2009; Bond 2004; Conduct Problems Prevention Research Group 2010; Cunningham 2012; Dolan 2010; Fearnow‐Kenney 2003; Griffin 2009; Li 2011; Bernstein 2010a; Bodin 2011; Bonds 2010), eight of which could be included in a meta‐analysis of short‐term effects (Analysis 3.1). One study could not be included (Cunningham 2012), and another two studies provided long‐term follow‐up 10 and 15 years after the intervention (Conduct Problems Prevention Research Group 2010; Bonds 2010, respectively). We have provided these data in Table 2,

Measures

For alcohol use, measures included initiation of alcohol use, frequency of alcohol use over a recent time period (appropriate to the age of participants), and measures of days drinking per week or month. For binge drinking, measures included ever having been drunk/really drunk, being drunk in the past 30 days, and frequency of binge drinking.

Effectiveness over the short term

Analysis 2.1 presents results of the meta‐analysis for alcohol use by intervention type for studies with follow‐up periods up to 12 months. We identified moderate‐quality evidence showing that universal or targeted interventions at the individual level probably have little or no effect in relation to alcohol use (universal individual‐level interventions: OR 0.80, 95% CI 0.58 to 1.11; P = 0.18; n = 4 studies; 1911 participants; I² = 0%; targeted individual‐level interventions: OR 1.02, 95% CI 0.80 to 1.31; P = 0.87; 2044 participants), although each analysis included relatively few studies (n = 4 studies; 1911 participants; and n = 4 studies; 2044 participants, respectively; moderate‐quality evidence). However, for universal individual‐level interventions, the effect estimate was in the direction of benefit.

Targeted family‐level interventions also probably have little or no effect in relation to alcohol use (OR 0.83, 95% CI 0.47 to 1.46; P = 0.52; n = 3 studies; 417 participants; moderate‐quality evidence).

On average, across eight universal school‐level interventions, evidence shows that such interventions probably reduce alcohol use (OR 0.72, 95% CI 0.56 to 0.92; P = 0.009; n = 8 studies; 8751 participants; moderate‐quality evidence). Heterogeneity was moderate (I² = 58%). Sensitivity analysis around the ICC used did not change the findings; and meta‐analysis of only those reporting dichotomous outcomes yielded a similar result, although this was not the case for two studies reporting continuous data (Table 6).

Among the studies that could not be included in the meta‐analysis with follow‐up data up to 12 months, one providing a targeted individual‐level intervention reported beneficial effects (Tierney 1995), and three exploring targeted individual‐level interventions reported no effect (Cunningham 2012; Freidman 2002; Monti 1999).

One targeted family‐level intervention showed beneficial effects (Gonzales 2012); one universal family‐based intervention (in Connell 2007) and three targeted family‐level interventions (in Estrada 2015,Jalling 2016, and Jalling 2016b) showed null effects.

Among school‐based interventions, one universal school‐level intervention showed beneficial effects (Piper 2000a), and one universal school‐level intervention showed benefit of one active intervention arm for boys only (Perry 2003a).

Long‐term effectiveness

Analysis 2.2 presents results of the meta‐analysis for alcohol use by study type for studies reporting follow‐up data over the longer term (> 12 months). Each subgroup included only two studies. One targeted school‐level study intervention produced a positive effect (Conduct Problems Prevention Research Group 2010). Other findings were inconclusive.

Binge drinking

Effectiveness over the short term

Analysis 3.1 presents results of the meta‐analysis for binge drinking for studies with follow‐up to 12 months. They show absence of evidence of an effect of targeted individual‐level interventions on binge drinking (OR 0.97, 95% CI 0.68 to 1.37; P = 0.85; n = 3 studies; 250 participants; I² = 48%). On average, across five universal school‐level interventions, evidence suggests that such interventions may have a beneficial effect (OR 0.66, 95% CI 0.41 to 1.06; P = 0.09; n = 5 studies; 5494 participants; low‐quality evidence), but the 95% CI was also consistent with the null hypothesis of no effect. Heterogeneity was moderate (I² = 43%). Sensitivity analysis around the ICC did not change the summary effect estimate, but the 95% CI was narrower (95% CI 0.45 to 0.99), and heterogeneity was slightly increased (I² = 49%) (Table 6).

Long‐term effectiveness

Analysis 3.2 presents results of the meta‐analysis for excess drinking for studies reporting data over the longer term. One targeted family‐level study showed an absence of effect (OR 1.30, 95% CI 0.79 to 2.13), and one targeted school‐level study reported an OR of 0.75 (95% CI 0.55 to 1.02).

Illicit drug use

A total of 53 interventions (76%) addressed illicit drug use (see Table 3 and Table 4). As above, three intervention studies aimed to prevent drug use by targeting early risk factors for later substance use (Ialongo 1999; Ialongo 1999b; Kellam 2014); and two additional intervention studies aimed to prevent drug use by acting on early life events and thus indirect mechanisms (Conduct Problems Prevention Research Group 2010; Schweinhart 1993). Most were conducted in the USA, with one conducted in the UK (Dolan 2010), two in Switzerland (Averdijk 2016; Averdijk 2016b), two in Sweden (Jalling 2016; Jalling 2016b), two in Canada (Morris 2003; Wolfe 2012), one in Australia (Bond 2004), and one in Hong Kong (Shek 2011).

We identified 29 studies that specifically measured cannabis use alone, all of which targeted other substances as well, some indirectly (as above). Twenty‐five of these studies also targeted antisocial behaviour. One study was conducted in the UK (Dolan 2010), two in Switzerland (Averdijk 2016; Averdijk 2016b), and the others in the USA (Table 3).

Measures

Measures of cannabis use included ever use, initiation of use (appropriate to the age of participants), frequency of use (e.g. in past month, in past six months). Similarly, measures of illicit drug use focused on use of one or more illicit substances, measuring frequency in the past month or a measure of initiation of illicit drug use.

Effectiveness over the short term

Analysis 4.1 presents results of the meta‐analysis for illicit drug use by study type for studies with follow‐up periods up to 12 months. Evidence from three studies shows that targeted individual‐level interventions probably had little or no effect in reducing illicit drug use over the short term (OR 0.94, 95% CI 0.71 to 1.25; P = 0.67; n = 3 studies; 638 participants; I² = 0%; moderate‐quality evidence).

One targeted family‐level intervention showed a null effect (OR 0.74, 95% CI 0.42 to 1.31; 69 participants; moderate‐quality evidence).

Analyses also indicated that two targeted school‐based interventions probably have little or no effect in relation to illicit drug use (OR 0.96, 95% CI 0.79 to 1.18; P = 0.72; n = 2 studies; 1299 participants; I² = 0%; moderate‐quality evidence). On average, across five universal school‐based interventions, evidence suggests that these interventions may have a positive effect in relation to illicit drug use (OR 0.74, 95% CI 0.55 to 1.00; P = 0.05; n = 5 studies; 11,058 participants; low‐quality evidence), although heterogeneity was substantial (I² = 69%). Sensitivity analyses did not markedly change these findings (Table 6).

Long‐term effectiveness

Ten studies provided data from long‐term follow‐up, ranging from two years up to 10 years (Conduct Problems Prevention Research Group 2010; Schweinhart 1993), 12 years (Kellam 2014), and 15 years post intervention (Analysis 4.2) (Bonds 2010).

Analysis 4.2 presents results of meta‐analyses for illicit drug use for studies with a follow‐up period greater than 12 months. On average, across four studies (five active intervention arms), data suggest uncertainty around the benefit of targeted family‐level interventions (OR 0.80, 95% CI 0.52 to 1.24; P = 0.32; n = 5 studies; 2032 participants; I² = 66%) due to the very low quality of the evidence, although the effect estimate was in the direction of benefit. Three of these estimates favoured intervention but the 95% CIs were consistent with the null hypothesis of no effect. Bonds 2010 reported a beneficial effect (OR 0.54, 95% CI 0.33 to 0.89) and Morris 2003 reported an adverse effect of the intervention (OR 1.24, 95% CI 1.00 to 1.55).

Evidence suggests that, on average, universal school‐level interventions may be effective in reducing illicit drug use (OR 0.73, 95% CI 0.56 to 0.95; P = 0.02; n = 4 studies; 3338 participants; I² = 0%; low‐quality evidence). Two of the interventions that were included in the meta‐analysis were conducted in the same study with one control group, which we accounted for in the analysis (see Data synthesis section).

Cannabis use
Effectiveness over the short term

Analysis 5.1 presents results of the meta‐analysis for cannabis use by study type. Ten studies reported data regarding cannabis use that could be synthesised in a meta‐analysis of effects of interventions with up to 12 months' follow‐up. Moderate‐quality evidence shows that targeted individual‐level interventions probably have little or no beneficial effects in relation to cannabis use (OR 1.10, 95% CI 0.69 to 1.76, P = 0.39; n = 2 studies; 126 participants; I² = 0%), whereas although a null effect was found for universal individual‐level interventions, the effect estimate was in the direction of benefit (OR 0.69, 95% CI 0.46 to 1.04; P = 0.08; two studies; 362 participants; I² = 0%; moderate‐quality evidence). Analysis also suggests that targeted family‐level interventions may have little or no difference in reducing cannabis use (OR 1.02, 95% CI 0.52 to 2.02; P = 0.95; n = 3 studies; 380 participants; I² = 43%; low‐quality evidence), with one study reporting an increase in cannabis use as a result of the intervention (mean frequency of marijuana use in intervention 10.8 compared to 6.4 among controls) (Milburn 2012).

On average, across five universal school‐based interventions, evidence shows that they may be beneficial (OR 0.79, 95% CI 0.62 to 1.01; P = 0.06; n = 5 studies; 4140 participants; I² = 0%; moderate‐quality evidence), and we considered the size of the effect to represent public health benefit at the population level, but the confidence interval spanned the null value. The sensitivity analysis conducted with different ICCs revealed a slightly lower effect estimate (OR 0.77, 95% CI 0.61 to 0.97) and heterogeneity that remained the same (I² = 0%) (Table 6).

Seven studies reported data regarding effects on cannabis or illicit drug use that could not be used in a meta‐analysis (Connell 2007; D'Amico 2002a; DeGarmo 2009; Freudenberg 2010; Griffin 2006; McNeal 2004; Piper 2000a); we present reported data in Table 2. In summary, two targeted individual‐level interventions showed a beneficial effect (Freudenberg 2010; Tierney 1995).

One family‐based intervention study reported an effect at one time point only (Connell 2007), and another reported no positive effect (Estrada 2015). Two family‐based intervention studies reported higher illicit drug use among those in the intervention arm compared to the control arm (Jalling 2016; Jalling 2016b).

One universal school‐based intervention study reported a beneficial effect (Griffin 2006); one universal school‐based programme reported an effect in one of the active intervention arms only (Piper 2000a); and two universal school‐based studies reported no effect (DeGarmo 2009; McNeal 2004).

Long‐term effectiveness

Analysis 5.2 presents results of the meta‐analysis for studies that aimed to address cannabis use and conducted follow‐up over the longer term. Only two studies in each study type subgroup could be included, and they provided no evidence of any effect for any study type.

Among two studies that could not be included in the meta‐analysis for cannabis or illicit drug use, one reported no effect of the intervention (Bonds 2010), and the other reported a beneficial effect (Griffin 2006)

Tobacco, alcohol, and/or illicit drug use (composite measures)

Several studies reported their findings as a composite measure of substance use (e.g. tobacco, alcohol, and/or illicit drug use) or alcohol and drug use. Six studies reported short‐term findings that could be included in meta‐analyses (Analysis 6.1). All except one study were conducted in the USA, the other in Hong Kong (Shek 2011). We have reported in Table 2 findings of an additional eight studies that could not be included in the meta‐analysis (Beach 2016; Berry 2009; Estrada 2015; Freudenberg 2010; Gonzales 2012; Griffin 2006; LoSciuto 1999; Olds 1998).

Measures

Measures included measures of substance use (tobacco, alcohol, and/or drugs; alcohol, tobacco, and/or marijuana; or alcohol and marijuana) over the past month, 6 months, or year, depending on the age of the participants.

Effectiveness over the short term

Analysis 6.1 presents results of the meta‐analysis for composite measures of substance use (i.e. tobacco, alcohol, and/or drug use). Evidence shows that targeted family‐level interventions may have little or no difference in reducing substance use (OR 0.81, 95% CI 0.50 to 1.33; P = 0.40; n = 2 studies; 213 participants; I² = 0%), although only two studies were included in each subgroup. We also found that universal school‐based interventions may show little or no difference in reducing substance use (OR 1.03, 95% CI 0.77 to 1.37; P = 0.85; n = 3 studies; 7390 participants; I² = 28%). Sensitivity analysis with a changed ICC altered the finding for the two universal school‐based interventions such that results of the meta‐analysis favoured control (OR 1.10, 95% CI 1.01 to 1.20; P = 0.03). Two targeted school‐based studies with multiple study arms were included in the meta‐analysis. Two arms of one of the studies showed benefit of the intervention (Lochman 2003a), but overall, data show uncertainty around the effects of these studies (Analysis 6.1) (Lochman 2003a; Lochman 2004a). Sensitivity analysis around the ICC did not change these findings (Table 6).

Among studies that could not be included in the meta‐analysis, two individual‐level intervention studies reported a beneficial effect (Berry 2009; Freudenberg 2010), two family‐level intervention studies reported a beneficial effect (Beach 2016; Gonzales 2012), and one family‐level intervention study reported a null effect (Estrada 2015). One school‐based intervention study reported an adverse effect (LoSciuto 1999).

Long‐term effectiveness

Analysis 6.2 presents the meta‐analysis for composite substance use by study type for studies reporting longer‐term follow‐up (> 12 months). Weak evidence from four targeted family‐based studies showed benefit in relation to substance use (OR 0.69, 95% CI 0.47 to 1.03; P = 0.07; n = 4 studies; 1622 participants; moderate‐quality evidence) along with substantial heterogeneity (I² = 82%) (Averdijk 2016; Brody 2012; Kim 2011; Kitzman 2010). Three studies were conducted in the USA, and one in Switzerland. Two universal school‐based intervention studies showed a null effect on drug and alcohol use approximately two years post intervention (OR 1.09, 95% CI 0.94 to 1.27; P = 0.24; n = 2 studies; 2145 participants; I² = 0%) (Averdijk 2016b; Wolfe 2012).

Among two studies that could not be included in the meta‐analysis, one targeting a family‐level intervention reported no effect of the intervention on this outcome (Olds 1998), and one targeting a universal school‐based intervention showed a beneficial effect (Griffin 2006)..

Antisocial behaviour and offending

Fifty‐three studies (76%) addressed antisocial behaviour, 11 of which were targeted to individuals, 17 to families, six to higher‐risk participants at schools, and 19 to participants at schools irrespective of risk. We have provided in Table 4 and Table 5 details of studies that targeted antisocial behaviour by study type.

All but one of the studies addressing antisocial behaviour concurrently aimed to prevent use of at least one substance (two of which were provided via indirect mechanisms (Conduct Problems Prevention Research Group 2010; Schweinhart 1993)); 16 studies (30%) concurrently aimed to reduce sexual risk behaviour. Twenty‐seven studies provided short‐term data that could be included in the meta‐analysis summarising impact of interventions on engagement in any antisocial behaviour (Analysis 7.1). Studies of each type were insufficient to show the impact of interventions on different forms of antisocial behaviour discussed in this review (e.g. property‐related offences vs arrests and general offences). Eleven additional studies reported long‐term data, one of which reported both short‐ and long‐term data (Lochman 2003a). We could not include 14 studies in the meta‐analysis; we have reported these data in Table 2. Most studies were conducted in the USA (n = 38 of 53; 72%), with one conducted in Canada (Wolfe 2012), one in Hong Kong (Shek 2011), one in India (Saraf 2015), one in Australia (Bond 2004), two in the UK (Berry 2009; Walker 2002), two in Switzerland (Averdijk 2016; Averdijk 2016b), one in Mexico/Spain (Lana 2014), one in Ireland (Dolan 2010), three in Sweden (Bodin 2011; Jalling 2016; Jalling 2016b), and one in South Africa (Matthews 2016). Thus, most studies included here were conducted in high‐income countries.

Measures

Measures for studies that provided shorter‐term follow‐up included aggressive or violent behaviour and school‐level or general delinquency; those with longer‐term follow‐up included measures around arrests in addition to measures of violence, aggressive behaviour, and/or conduct problems.

Effectiveness over the short term

Analysis 7.1 presents results for the meta‐analysis for antisocial behaviour by study type for studies reporting follow‐up data up to 12 months. Overall, moderate‐quality evidence shows that individual‐level interventions probably produced little or no difference in reducing antisocial behaviour (universal individual‐level: OR 1.02, 95% CI 0.62 to 1.69; one study; 200 participants; targeted individual‐level: OR 1.21, 95% CI 0.92 to 1.60; P = 0.17; I² = 19%; n = 4 studies; 764 participants; both moderate‐quality evidence).

Among family‐level interventions, one universal study with two study arms showed no effect on engagement in antisocial behaviour (OR 0.87, 95% CI 0.56 to 1.35; P = 0.53; 306 participants; I² = 0%; moderate‐quality evidence), and six targeted family‐level studies showed no effect in relation to engagement in antisocial behaviour (OR 0.84, 95% CI 0.57 to 1.24; P = 0.39; I² = 42%; n = 6 studies; 772 participants; moderate‐quality evidence). The STRIVE programme, in particular, showed a large effect (Milburn 2012).

Studies provided weaker evidence of a probable positive effect of targeted school‐based interventions in relation to antisocial behaviour (OR 0.78, 95% CI 0.59 to 1.05; P = 0.1; I² = 0%; n = 3 studies; 1531 participants; moderate‐quality evidence), although the effect was in the direction of benefit, and sensitivity analysis around the ICC slightly altered this finding in favour of the interventions (OR 0.82, 95% CI 0.68 to 0.99; P = 0.04; I² = 0%). Last, on average, across 13 universal school‐based interventions, evidence suggests that such interventions may have an impact in reducing antisocial behaviour (OR 0.81, 95% CI 0.66 to 0.98; P = 0.03; n = 13 studies; 20,756 participants; very low‐quality evidence), although heterogeneity was substantial (I² = 66%). Sensitivity analyses around the ICC, income classification of study country, and dichotomous versus continuous outcomes yielded similar results (Table 6).

Analysis 7.2 presents results for the meta‐analysis for violent offences for studies reporting data at up to 12 months' follow‐up. For individual‐ and family‐based interventions, data showed uncertainty around whether there was an effect, or evidence suggested that such interventions may have little or no effect in reducing violent behaviour, although few studies could be included in each subgroup (targeted individual‐level studies: OR 1.11, 95% CI 0.56 to 2.17; P = 0.77; n = 2 studies; 514 participants; I² = 68%; targeted family studies: OR 0.95, 95% CI 0.49 to 1.84; n = 1 study; 238 participants).

Meta‐analyses did not show benefit of school‐based targeted interventions (OR 0.60, 95% CI 0.31 to 1.16; n = 1 study with three arms; 158 participants; P = 0.13; I² = 0%).

Similarly, on average, across nine universal school‐level interventions, evidence indicated that such interventions may have little effect (OR 0.86, 95% CI 0.69 to 1.07; P = 0.18; n = 9 studies; 11,347 participants; I² = 70%), although as above, the effect estimate was in the direction of benefit, but the 95% CI was consistent with the null hypothesis. Sensitivity analyses around the ICC did not change these findings (Table 6).

Analysis 7.3 presents results for the meta‐analysis for school‐level or general delinquency (e.g. stealing things worth less than USD50) by study type. For targeted individual‐level interventions, two studies provided evidence of a null effect (OR 1.07, 95% CI 0.61 to 1.89; 250 participants; P = 0.81; I² = 37%).

For targeted family‐level interventions, evidence of benefit was insufficient (OR 0.80, 95% CI 0.54 to 1.20; P = 0.28; n = 4 studies; 598 participants; I² = 40%), although the effect estimate was in the direction of benefit.

For targeted school‐based interventions, evidence of benefit was insufficient (OR 0.79, 95% CI 0.59 to 1.06; P = 0.11; n = 3 studies; 1573 participants; I² = 0%), although the summary estimate was in the direction of benefit, and on average, evidence showed a beneficial effect from six universal school‐based interventions in relation to school or general delinquency (OR 0.88, 95% CI 0.77 to 1.00; P = 0.05; n = 6 studies; 10,113 participants; I² = 0%).

We did not have sufficient data to analyse the effects of interventions on any other individual domains of antisocial behaviour.

Among studies that could not be included in the meta‐analysis, three providing targeted individual‐level interventions reported a beneficial effect (Berry 2009; Freudenberg 2010; Tierney 1995), and two reported a null effect (Cunningham 2012; Freidman 2002). Two studies examining targeted family‐level interventions reported a beneficial effect (Beach 2016; Gonzales 2012), and a universal family‐level programme reported a null effect (Connell 2007). One targeted school‐level intervention revealed a null effect (Shetgiri 2011). Last, one universal school‐level intervention showed a beneficial effect (DeGarmo 2009), one showed an effect of one active study arm for boys only (Perry 2003a), and one revealed a null effect (LoSciuto 1999).

Long‐term effectiveness

Analysis 7.4 presents results of the meta‐analysis for antisocial behaviour by study type from studies with follow‐up over 12 months. Eleven studies reported longer‐term outcome data and could be included in the meta‐analysis. Evidence suggested that targeted family‐level interventions representing 2486 participants may be beneficial (OR 0.74, 95% CI 0.54 to 1.03; P = 0.08; n = 5 studies; 2486 participants; I² = 78%; low‐quality evidence) although heterogeneity was substantial, and weak evidence suggested that a universal family‐based programme with two active study arms may have a beneficial effect (OR 0.67, 95% CI 0.43 to 1.04; n = 1 study, 304 participants; P = 0.07), although the 95% CIs highlight that the data are inconclusive.

Evidence on two universal school‐based interventions highlighted uncertainty in their effect over the long term (OR 0.91, 95% CI 0.63 to 1.31; P = 0.6; n = 2 studies; 4146 participants; I² = 60%), and the meta‐analysis included only two studies.

Among studies that could not be included in the meta‐analysis, one providing a targeted family intervention reported a null effect (Averdijk 2016), one providing a targeted family‐level intervention showed a null effect for externalising problems and major delinquent acts but a beneficial effect for incidence of arrests (Olds 1998), and one providing a targeted family‐based intervention reported a null effect for carrying a gun or knife once or more (Schweinhart 1993).

Two targeted school‐based studies reported a beneficial effect (Conduct Problems Prevention Research Group 2010; Kellam 2014), and one universal school‐based study reported a null effect (Averdijk 2016b).

Vehicle‐related risk behaviour

Four studies addressed vehicle‐related risk behaviours (Bernstein 2010a; D'Amico 2002a; Monti 1999; Nirenberg 2013); three provided targeted individual‐level interventions (Bernstein 2010a; Monti 1999; Nirenberg 2013), and one provided a universal school‐based intervention (D'Amico 2002a). Two studies could be included in a meta‐analysis of effects of the interventions in relation to driving under the effect of alcohol and/or drugs (Bernstein 2010a; Monti 1999); we have reported in Table 2 findings from the other two studies.

Measures

Measures from these studies related to dangerous driving, alcohol‐ and/or drug‐related vehicle‐risk behaviour, and driving with a drinking driver.

Effectiveness over the short term

Analysis 8.1 presents results of the meta‐analysis for vehicle‐related risk behaviour. Two studies with targeted individual‐level interventions reported different effects, with one accident and emergency (A&E)‐based brief motivational intervention showing no effect on alcohol‐related vehicle‐risk behaviour (OR 1.11, 95% CI 0.74 to 1.69 (n of participants not provided)) (Bernstein 2010a), and a second brief intervention showing benefit for reducing drinking and driving (OR 0.26, 95% CI 0.08 to 0.83; n = 94 participants), although this was a small study and the 95% confidence intervals were wide (Monti 1999).

Among the two studies that could not be included in the meta‐analysis, one providing a targeted individual‐level intervention (Nirenberg 2013) and one providing a universal school‐based intervention (D'Amico 2002a) reported null effects.

Long‐term effectiveness

Identified studies conducted follow‐up over a period of six months post intervention (D'Amico 2002a; Monti 1999; Nirenberg 2013), or over 12 months post intervention (Bernstein 2010a). No studies provided longer‐term follow‐up.

Sexual risk behaviour

Twenty‐one studies (30%) aimed to reduce sexual risk behaviour alongside other risk behaviours (see Table 4; and Table 5), and two studies examining additional interventions measured sexual risk behaviour at long‐term follow‐up (Conduct Problems Prevention Research Group 2010; Kellam 2014), giving a rationale for an indirect effect of the interventions on sexual risk behaviour (e.g. via targeting of early aggressive and disruptive behaviour, which are antecedent risk factors for later risk behaviours including sexual risk).

Most of the studies provided universal school‐based interventions (n = 9; 43%) or targeted family‐based interventions (n = 7; 33%). Most interventions concurrently addressed antisocial behaviour and prevention of use of at least one substance (n = 15). One study was conducted in Australia (Bond 2004), one in Hong Kong (Shek 2011), one in Canada (Wolfe 2012), and one in South Africa (Matthews 2016); the remainder were conducted in the USA (n = 16 studies). Further study details can be found in Table 3; and Table 5.

Eleven studies provided short‐term data and could be included in the meta‐analysis (Beets 2009; Bernstein 2010a; Bond 2004; Flay 2004a; Li 2002; McNeal 2004; Milburn 2012; Minnis 2014; Pantin 2009; Redding 2015; Shek 2011), and eight studies provided longer‐term follow‐up (Bond 2004; Bonds 2010; Conduct Problems Prevention Research Group 2010; Haggerty 2007a; Kellam 2014; Kim 2011; Redding 2015; Wolfe 2012). Five studies could not be included in the meta‐analysis (Estrada 2015; Freudenberg 2010; Griffin 2006; Olds 1998; Piper 2000a); we have provided in Table 2 data from these studies.

Measures

Measures of sexual risk behaviour included condom use, unprotected sex, multiple partners, and composite measures of sexual risk. Studies that targeted primary school age or early adolescence reported early initiation of sexual intercourse.

Effectiveness over the short term

Analysis 9.1 presents results of the meta‐analysis for sexual risk behaviour by study type for interventions reporting follow‐up data up to 12 months. Moderate quality evidence shows that individual‐level interventions probably have little or no effect in reducing sexual risk behaviour (targeted: OR 0.73, 95% CI 0.49 to 1.08; P = 0.11; n = 2 studies; 494 participants; I² = 45%; universal: OR 0.42, 95% CI 0.14 to 1.25; n = 1 study; 162 participants). The effect estimate for targeted individual‐level interventions was in the direction of benefit. However, each subgroup of the meta‐analysis included few studies (each subgroup synthesis included fewer than 500 participants); thus findings must be interpreted with caution.

Moderate‐quality evidence shows that targeted family interventions also probably have little or no effect in reducing sexual risk behaviour (OR 0.89, 95% CI 0.55 to 1.44; P = 0.63; n = 3 studies; 371 participants).

On average, across six universal school‐based interventions, the effect was in the direction of benefit of the interventions, but the 95% CI was consistent with the null hypothesis of no effect (OR 0.83, 95% CI 0.61 to 1.12; P = 0.22; n = 6 studies; 12,633 participants; low quality evidence) and heterogeneity was substantial (I² = 77%), in part because one study showed a clear beneficial effect (OR 0.18, 95% CI 0.08 to 0.41) (Beets 2009). Sensitivity analysis did not change these findings (Table 6).

Among studies that could not be included in the meta‐analysis, one targeted individual‐level study reported null findings (Freudenberg 2010). One targeted family‐based study reported beneficial findings (Murry 2014), and one reported null findings (Estrada 2015).

Long‐term effectiveness

Analysis 9.2 presents results of the meta‐analysis for sexual risk behaviour by study type, for studies reporting longer‐term follow up. As for short‐term analyses, few interventions in each study type subgroup provided long‐term follow‐up data. One targeted individual‐level study reported a null effect (OR 0.93, 95% CI 0.64 to 1.35; n = 461 participants).

Moderate‐quality evidence shows benefit of targeted family‐based interventions on average (OR 0.47, 0.31 to 0.71; P = 0.0004; n = 2 studies; 318 participants; I² = 0%), and one universal family‐level intervention (with two study arms) reported a null effect (OR 1.12, 95% CI 0.64 to 1.96; 237 participants). One targeted school‐based intervention provided evidence of benefit (OR 0.62, 95% CI 0.47 to 0.82; P = 0.0009; 650 participants). As above, the overall effect of universal school‐based interventions was in the direction of benefit, but the 95% CI was consistent with the null hypothesis of no effect (OR 0.74, 95% CI 0.50 to 1.09; P = 0.13; n = 3 studies; 3391 participants).

Among studies that could not be included in the meta‐analysis, one targeted family‐based study reported beneficial findings (Bonds 2010), and one reported beneficial findings in one subgroup for one study arm only (Olds 1998). One universal school‐based study reported beneficial findings for number of sexual partners but not for condom use (Griffin 2006).

Physical activity

Nine studies targeted physical inactivity (Bush 1989; Lana 2014; Melnyk 2013; Nader 1999; O'Neill 2011; Saraf 2015; Schwinn 2014; Walker 2002; Walter 1989), six of which were universal school‐based interventions (Bush 1989; Melnyk 2013; Nader 1999; O'Neill 2011; Saraf 2015; Walter 1989). All concurrently aimed to prevent poor nutrition, and all simultaneously aimed to prevent use of at least one substance. One study was conducted in the UK (Walker 2002), one in Spain/Mexico (Lana 2014), and one in India (Saraf 2015); the remainder were conducted in the USA (six studies).

Measures

Studies assessed vigorous activity, fitness, and physical activity.

Effectiveness over the short term

Analysis 10.1 presents findings of the meta‐analysis for vigorous physical activity by study type for studies reporting follow‐up data up to 12 months. Seven studies could be included in the meta‐analysis. Evidence shows that universal individual‐level interventions probably have little or no effect in enhancing physical activity (OR 1.11, 95% CI 0.74 to 1.67; P = 0.62; n = 2 studies; 1530 participants; I² = 0%; moderate‐quality evidence).

One targeted family‐level intervention provided no benefit (OR 0.72, 95% CI 0.29 to 1.75; n = 61 participants). Note here that positive odds ratios demonstrate that the intervention can increase physical activity and thus represents a benefit (i.e. favouring the intervention).

In contrast to the aforementioned studies, evidence shows that on average, across four studies, universal school‐based studies improve physical activity (OR 1.32, 95% CI 1.16 to 1.50; P < 0.0001; n = 4 studies; 6441 participants; I² = 0%; moderate‐quality evidence). Sensitivity analysis around the ICC did not change this result (OR 1.33, 95% CI 1.18 to 1.50; P < 0.00001; I² = 0%) (Table 6), and including only studies conducted in high‐income countries slightly increased the odds ratio (Table 6).

Among four universal school‐based studies that could not be included in the meta‐analysis, three reported beneficial findings (Bush 1989; O'Neill 2011; Saraf 2015), and one showed a null effect (Walter 1989).

Long‐term effectiveness

No studies reported longer‐term follow‐up.

Unhealthy diet outcomes

Eleven (16%) of the 70 studies addressed unhealthy diet (Bush 1989; Fearnow‐Kenney 2003; Lana 2014; Melnyk 2013; Nader 1999; O'Neill 2011; Piper 2000a; Saraf 2015; Schwinn 2014; Walker 2002; Walter 1989), eight of which provided universal‐school based interventions (Bush 1989; Fearnow‐Kenney 2003; Melnyk 2013; Nader 1999; O'Neill 2011; Piper 2000a; Saraf 2015; Walter 1989). All simultaneously addressed substance use in addition to unhealthy diet; and nine studies concurrently addressed physical activity. One study was conducted in the UK (Walker 2002), one in Spain/Mexico (Lana 2014), and one in India (Saraf 2015); the remainder were conducted in the USA. Further details can be found in Table 4 and Table 5.

Measures

Studies reported outcomes related to unhealthy diet (e.g. dietary cholesterol, dietary fat) as well as body mass index (BMI) to provide a measure of obesity. We conducted separate meta‐analyses for these outcome measures.

Effectiveness over the short term

Analysis 12.1 and Analysis 12.2 present results of the meta‐analyses for BMI and unhealthy diet by study type for studies reporting follow‐up to 12 months.

For BMI, one universal individual‐level intervention showed a null effect (OR 0.80, 95% CI 0.48 to 1.31; 579 participants). There was no evidence that universal school‐based interventions, on average, had a positive effect (OR 0.84, 95% CI 0.60 to 1.19; P = 0.33; n = 3 studies; 5017 participants), and heterogeneity was substantial (I² = 61%).

For unhealthy diet, the evidence for individual‐level interventions was uncertain (OR 0.76, 95% CI 0.42 to 1.34; P = 0.34; n = 2 studies; 1549 participants; I² = 51%). On average, across three studies, the effect of universal school‐based interventions was in the direction of benefit in relation to unhealthy diet, but the 95% CI was consistent with the null hypothesis of no effect (OR 0.82, 95% CI 0.64 to 1.06; P = 0.13; n = 3 studies; 6441 participants; I² = 49%), and few studies were included in the meta‐analysis. Sensitivity analysis including only studies conducted in high‐income countries (n = 2 studies) slightly increased the odds ratio (OR 0.95, 95% CI 0.76 to 1.19; P = 0.68; I² = 0%) (Table 6).

Among studies for which data could not be included in the meta‐analyses, one universal school‐based study reported a null effect for BMI and a beneficial effect for total fat intake in one of two areas included in the study (Walter 1989). A second universal school‐based study reported beneficial effects (O'Neill 2011).

Long‐term effectiveness

No studies provided longer‐term follow‐up data.

Secondary outcomes

We identified data regarding mental health and educational attainment that could be included in quantitative syntheses; we have reported in Table 2 data regarding teenage pregnancy. We did not identify data that could be analysed regarding cost‐effectiveness of the interventions.

Mental health

Seventeen (27%) of 62 studies aimed to improve mental health (Bodin 2011; Bond 2004; Bonds 2010; Brody 2012; Conduct Problems Prevention Research Group 2010; Gonzales 2012; Ialongo 1999; Ialongo 1999b; Kellam 2014; Kim 2011; Kitzman 2010; Li 2011; Melnyk 2013; Morris 2003; Olds 1998; Sanchez 2007; Walker 2002), one through indirect long‐term mechanisms (Olds 1998). Six of these (35%) provided targeted family‐based interventions, and five (29%) universal school‐based interventions (see Table 5). Fifteen of these studies addressed substance use, and fourteen also aimed to prevent antisocial behaviour. Among studies not conducted in the USA, one was conducted in Sweden (Bodin 2011), one in Ireland (Dolan 2010), one in the UK (Walker 2002), and one in Australia (Bond 2004).

Seven studies could not be included in the meta‐analysis (Gonzales 2012; Ialongo 1999; Ialongo 1999b; Kellam 2014; Sanchez 2007; Conduct Problems Prevention Research Group 2010; Olds 1998), four of which reported clinical assessments (e.g. diagnosis of conduct disorder). We have reported findings from these studies in Table 2.

Measures

All studies included in the meta‐analysis measured depression or internalising behaviour/problems.

Effectiveness over the short term

Analysis 11.1 presents results of meta‐analyses for depressive symptoms by study type for studies reporting follow‐up to 12 months. Four studies provided short‐term follow‐up data and could be included in the meta‐analysis. One targeted individual‐level intervention showed a null effect (OR 1.02, 95% CI 0.54 to 1.93; n = 124 participants). On average, across three universal school‐based studies, we found no evidence of benefit in relation to prevention of depressive symptoms (OR 0.92, 95% CI 0.71 to 1.20; P = 0.56; n = 3 studies; 3907 participants), and heterogeneity was substantial (I² = 63%).

Long‐term effectiveness

Analysis 11.2 presents results of meta‐analyses for depressive symptoms by study type for studies reporting longer‐term follow‐up data (> 12 months). Five studies were included in the meta‐analysis, four of which provided targeted family‐based interventions. Overall, evidence showed benefit of targeted family‐based interventions in preventing depressive symptoms over the longer term (OR 0.88, 95% CI 0.80 to 0.98; P = 0.02; n = 4 studies; 2386 participants; I² = 0%). One targeted school‐based study reported inconclusive evidence (OR 0.68, 95% CI 0.42 to 1.09; n = 721 participants).

Two targeted family‐based studies that could not be included in the meta‐analysis reported beneficial effects for some outcomes and not for others or at different time points (Bonds 2010; Gonzales 2012); another reported a null effect (Olds 1998). One targeted school‐based intervention showed a null effect (Kellam 2014), and one universal individual‐level intervention showed benefit only among those reporting probable depression (Walker 2002).

Educational attainment

Nineteen studies (31%) aimed to improve educational attainment alongside other risk behaviours (Berry 2009; Conduct Problems Prevention Research Group 2010; Dolan 2010; Freidman 2002; Freudenberg 2010; Gonzales 2012; Ialongo 1999; Ialongo 1999b; Kellam 2014; Kitzman 2010; Li 2011; LoSciuto 1999; Melnyk 2013; Minnis 2014; Morris 2003; Olds 1998; Sanchez 2007; Schweinhart 1993; Tierney 1995), five of which were targeted at the individual level and four at the family level; seven provided universal school‐based interventions (see Table 4). One of these was conducted in the UK (Berry 2009), one in Ireland (Dolan 2010), and one in Canada (Morris 2003); the remaining 17 studies (85%) were conducted in the USA. Sixteen studies concurrently targeted substance use, and 17 targeted antisocial behaviour.

Measures

Studies reported academic achievement in terms of grade point average (GPA) or scores for particular subjects (e.g. mathematics achievement scores).

Effectiveness over the short term

Analysis 13.1 presents data for the meta‐analysis for academic achievement by study type for studies reporting follow‐up data up to 12 months. Six studies could be included in meta‐analyses of short‐term follow‐up data, one of which tested two study arms at the same time (Ialongo 1999; Ialongo 1999b). Findings were variable, and each subgroup included few studies. One targeted individual‐level intervention reported a null effect (OR 1.34, 95% CI 0.71 to 2.52; 126 participants). Three targeted school‐based studies reported null findings (OR 0.91, 95% CI 0.30 to 2.73; P = 0.86; n = 3 studies; 1247 participants; I² = 84%), and one universal school‐based study with two study arms reported null findings (OR 0.94, 95% CI 0.62 to 1.44; 579 participants; I² = 0%).

Long‐term effectiveness

We were not able to include in a meta‐analysis data regarding longer‐term educational outcomes. Consequently, we did not conduct a meta‐analysis to examine long‐term effectiveness for this outcome. Among studies that could not be included in the meta‐analysis: three targeted individual‐level interventions reported a beneficial effect for education or employment (Berry 2009; Freudenberg 2010), as well as academic achievement (Tierney 1995), and one reported no effect on school problems (Freidman 2002).

One targeted family‐level intervention reported an indirect effect of the intervention through school engagement (Gonzales 2012), and another reported benefit in relation to educational attainment but no effect on school suspensions (Kitzman 2010). For another targeted family‐level financial support intervention, findings were uncertain (Morris 2003). One targeted school‐based intervention reported benefit of preschool programme (Schweinhart 1993), and the findings of a targeted school‐based study were inconclusive (Conduct Problems Prevention Research Group 2010).

Two universal school‐level interventions reported a positive effect in relation to school attendance and academic competence (LoSciuto 1999; Melnyk 2013), one universal school‐based intervention showed no effect on school attachment (Bond 2004), another reported no effect on school suspensions (Li 2011), and another reported possible benefit in relation to high school graduation (Kellam 2014).

We have presented these results in Table 2.

Multiple behaviours

Although some studies provided insufficient evidence of a positive effect on more than one outcome, several interventions showed benefit in preventing more than one behaviour, although the strength of the evidence for an effect varied between studies.

Among targeted individual‐level interventions, the REAL MEN intervention for young males leaving jail, described in Freudenberg 2010, and the mentoring intervention, Big Brothers Big Sisters, reported by Tierney 1995, yielded effects for multiple behaviours that were in the direction of benefit, as did the individual‐level conditional cash transfer and life skills intervention, Yo Puedo (Minnis 2014). Evidence suggests that targeted family‐level interventions New Beginnings (Bonds 2010), SAAF‐T (Brody 2012), Middle School Success (Kim 2011), and the Nurse Family Partnership (Kitzman 2010) produced positive effects for at least three behaviours; and evidence shows that the targeted school‐level intervention FAST TRACK showed benefit in relation to at least three risk behaviours (Conduct Problems Prevention Research Group 2010).

Researchers provided evidence of a positive effect of six universal school‐based interventions on at least three risk behaviours (Beets 2009; Bond 2004; Li 2011; Melnyk 2013; O'Neill 2011; Saraf 2015), and they provided evidence suggestive of benefit for three others (DeGarmo 2009; Griffin 2006; Kellam 2014). Positive Action (Hawaii) showed benefit in preventing tobacco use, alcohol use, drug use, antisocial behaviour, and sexual risk behaviour (Beets 2009), and Positive Action (Chicago) showed benefit in relation to alcohol use, antisocial behaviour, depressive symptoms, and illicit drug use (Lewis 2012; Lewis 2013; Li 2011). Positive Action (a social‐emotional and character development model) involves more than 140 15‐minute, age‐appropriate lessons taught four days per week from kindergarten to grade six, and 70 lessons taught two days per week for grades seven and eight. Training for teachers, families, communities, and school climate changes is also involved. The Michigan Model for Health (MMH) (O'Neill 2011), a health education curriculum intervention for students from kindergarten to grade 12, showed beneficial results in relation to tobacco use, alcohol use, and antisocial behaviour, with a more recent study highlighting effects for physical activity and nutrition (see Studies awaiting classification). The MMH involves 24 lessons in grade four and 28 lessons in grade five, focusing on skills‐based learning. The 15‐week health course, COPE, which incorporated physical activity into skills‐building sessions, was effective in reducing alcohol use and BMI and increasing physical activity (Melnyk 2013). Saraf 2015 reported beneficial effects of a multi‐component school‐based study in relation to tobacco use, poor diet, and physical inactivity. Over the longer term, FAST TRACK, which was implemented between grades 1 and 10, and which involved parent training, tutoring, skills development curriculum, home visits, and parent‐child and parent‐youth groups, was effective in preventing alcohol use, illicit drug use, and sexual risk behaviour (Conduct Problems Prevention Research Group 2010). Last, the Gatehouse study (Bond 2004), a whole‐school intervention involving a curriculum and school‐wide changes, reduced tobacco use as well as sexual risk behaviour and antisocial behaviour over the longer term.

Investigation of the impact of interventions on combinations of behaviours will be the topic of further research to be conducted by the study team in the future.

Discussion

Summary of main results

This review has examined evidence related to the effectiveness of individual‐, school‐, and family‐level interventions that target multiple risk behaviours among young people eight to 25 years of age. We identified 70 studies, of which half were universal and half were targeted to individuals in particular ethnic or socioeconomic groups, or to those identified as being at higher risk of engagement in risk behaviours and/or consequent harms. A substantial proportion of the interventions identified were universal school‐based interventions (n = 28; 40%), and these made up the majority of school‐based programmes. Of 35 targeted studies, 17 (49%) provided family‐level interventions, six (17%) school‐level interventions, and 12 (34%) individual‐level interventions.

The included studies focused on a broad range of outcomes, targeting, on average, four behaviours. Most studies targeted tobacco, alcohol, and/or drug use and antisocial behaviour. We conducted meta‐analyses for ten primary outcomes (tobacco use, alcohol use, binge drinking, illicit drug use, cannabis use, substance use, vehicle‐related risk behaviour, sexual risk behaviour, physical inactivity, and unhealthy diet) and two secondary outcomes (depressive symptoms and educational attainment).

Overall, evidence from meta‐analyses showed that on average, universal school‐based interventions probably have a positive effect in relation to tobacco use, alcohol use, and physical activity, and that they may also have a beneficial effect in relation to illicit drug use and antisocial behaviour at up to 12 months' follow‐up, versus a comparator. Available data did not permit analysis of particular intervention components associated with effectiveness, and no single component was consistently associated with stronger effects. Nevertheless, our data suggest that interventions for which meta‐analyses showed beneficial effects in relation to at least one behaviour included additional components, such as school policy changes, school‐wide adoption of the intervention principles, or family engagement.

For instance, our meta‐analyses demonstrated a beneficial effect of the Positive Action programme on tobacco, alcohol, and drug use, as well as antisocial behaviour (Beets 2009; Li 2011). This programme commenced in primary school and involved classroom curricula each year, school‐wide climate changes to reinforce positive actions throughout the school, and family components organised around the core concepts of self‐concept, positive actions for body and mind, getting along with others, social and emotional actions for managing oneself responsibly, being honest with yourself and others, and self‐improvement. In addition to Positive Action, the meta‐analysis demonstrated that interventions that were effective in relation to tobacco use included the Gatehouse Study (Bond 2004), the Michigan Model for Health (O'Neill 2011), and Going Places (Simons‐Morton 2005), with the Michigan Model for Health being effective in relation to alcohol use and antisocial behaviour as well. As noted for Positive Action (Beets 2009; Li 2011), which is a kindergarten through grade 12 multi‐component programme, these interventions have in common an extended duration of intervention, or a multi‐component or whole‐school approach. For instance, Going Places included a social skills curriculum, parent education, and school environment enhancement and was implemented across three school years (Simons‐Morton 2005); the Gatehouse Project is a whole‐school intervention based on the Health‐Promoting Schools Framework (Bond 2004); and the Michigan Model for Health is a kindergarten through grade 12 school curriculum programme (O'Neill 2011). Additional programmes that were effective in relation to two behaviours (COPE and BRAVE) involved family or mentoring components as well as school curricula (Griffin 2009; Melnyk 2013). Thus, it is possible that interventions that have multiple components, involve school‐wide changes, and/or are extended in duration may be effective in relation to these behaviours.

In contrast to the above studies, we did not find evidence that family‐level or individual‐level interventions have a positive effect on the outcomes investigated, although we identified fewer of these studies. In addition, effect estimates for individual‐level studies were in the direction of benefit for certain outcomes (alcohol use, cannabis use, sexual risk behaviour) but not for others.

Over the longer term, evidence showed that universal school‐based interventions that took a whole‐school approach ‐ as described by Bond 2004 ‐ or that commenced early in primary school and targeted antecedent risk factors ‐ as examined by Ialongo 1999,Ialongo 1999b, and Kellam 2014 ‐ may be beneficial in relation to illicit drug use. Evidence showed that targeted family‐level interventions may be effective in reducing substance use, antisocial behaviour, sexual risk behaviour, and depressive symptoms over the longer term. However, it must be noted that comparatively few studies reported longer‐term follow‐up data (i.e. at least two years post intervention), so these findings should be interpreted with caution. Evidence suggests that multiple risk behaviour interventions conducted at the individual level may have little or no effect in preventing these outcomes, although, again, these studies are few.

Overall completeness and applicability of evidence

In our review, most studies addressed tobacco, alcohol, and/or drug use alongside antisocial behaviour, with a smaller proportion addressing sexual risk behaviour, mental illness, and educational attainment.

It is interesting to note that despite relatively high prevalence of engagement in other risk behaviours, we found few studies targeting these behaviours as a multiple risk approach alongside the other behaviours included in this review. For instance, data from cohort studies show that in the UK, 26% of females and 9% of males have ever self‐harmed by age 16 to 17 years (Kidger 2012), and self‐harm is a leading cause of mortality among young people globally (leading to 8% to 9% of deaths) (Mokdad 2013). Recent estimates of the prevalence of gambling among young people aged 10 to 24 years have ranged from 0.2% to 12.3% (Calado 2016), and road injuries remain a leading cause of disability‐adjusted life years (DALYs) and death among young people globally (accounting for 5.4% of total DALYs for young people aged 10 to 24 years and 14% to 15% of deaths among young people aged 15 to 24, respectively) (Gore 2011; Mokdad 2013). Nevertheless, we found no studies that targeted self‐harm or gambling alongside other behaviours, and only four studies targeted vehicle‐related risk behaviour in multiple risk behaviour interventions. This was also the case for physical inactivity and poor nutrition, which have high prevalence (MacArthur 2012) but were not frequently targeted alongside risk behaviours that may represent greater association with experimentation or 'thrill‐seeking'. This may reflect the view that addressing antecedent risks of low levels of activity and poor diet may require distinct approaches owing to clustering patterns of different risk behaviours (Faeh 2006; Meader 2016; Mistry 2009; van Nieuwenhuijzen 2009; Wiefferink 2006), or because of the need to focus on training in specific skills related to addressing a particular risk behaviour, such as self‐harm.

We identified a larger number of studies that compared universal school‐based interventions versus those targeted at the individual or family level; thus we note that some caution is warranted in interpreting findings regarding the latter types of interventions. School‐based programmes have tended to predominate among interventions targeted to adolescents. This may reflect the target age range of participants, thus almost universal coverage of young people, access to large numbers of adolescents, and ease of delivery (Bonnell 2016; Stockings 2016), making schools a highly efficient setting for behaviour change‐focused interventions. Most of these 34 school‐level interventions (n = 28; 82%) were universal in nature and did not target particular groups, as might be expected in this setting. Although several studies targeted individuals in early adolescence and thus focused on primary prevention, relatively few were initiated among children younger than 10 years (n = 10 of 28 studies; 36%), and so it was not possible to assess whether early intervention led to differential impact for the different outcomes assessed.

As outlined above, data suggest that interventions with beneficial effects were often characterised by multiple components including school‐wide changes or family engagement. Recent evidence highlights that combined student and parent programmes can be effective in relation to substance use outcomes (Newton 2017), and whole‐school interventions that combine multiple components such as policy changes and parental involvement can prevent smoking and sexual risk behaviour (Langford 2014; Shackleton 2016). We will be conducting further research to explore the impact of distinct components of interventions, combinations of components, or the intensity of intervention components because such additional analyses were not possible here, given the scale and complexity of this review. Similarly, we were not able to examine the specific impact of interventions that targeted particular combinations of behaviours; further analyses are required in this regard.

In contrast to school‐based interventions, 17 of 20 (85%) family‐level studies were targeted to particular populations on the basis of factors such as ethnic group, socioeconomic status, or family status (e.g. being in foster care, having recently divorced parents). Most of these interventions targeted illicit drug use (95%), antisocial behaviour (85%), and alcohol use (80%). Over half (55%) targeted tobacco use, sexual risk, and mental illness. The particular groups targeted in such interventions showed variability, and heterogeneity was evident among the interventions themselves, making conclusions about these types of interventions difficult.

Although this review provides the first quantified effect estimate for the effects of multiple risk behaviour interventions, we must note additional limitations of the review. Overall, all studies lacked a focus on equity. For instance, data were insufficient to show the impact of interventions in relation to gender, ethnicity, and/or socioeconomic group. In addition, most studies were conducted in the USA (79%), only two studies in low‐ or middle‐income countries (Matthews 2016; Saraf 2015), and one jointly in an upper‐middle‐income country and a high‐income country (Lana 2014). A scoping review of multiple risk behaviour interventions in adult populations also found a scarcity of studies conducted among minority ethnic groups and lack of studies conducted in the Middle East, Africa, and South America (Meader 2016). Thus, the generalisability of these findings to other geographical settings and educational or cultural contexts is unclear. In 2012 it was reported that 90% of the world's population of young people aged 10 to 24 years live in low‐ and middle‐income countries (LMICs), and some of the highest rates of tobacco use and overweight and lowest levels of physical activity were evident in these areas (Sawyer 2012), alongside a substantial health burden from injuries, unsafe sex, and alcohol use (Gore 2011; Mokdad 2013). These findings together highlight the need to examine the effectiveness of context‐dependent and culturally relevant interventions that may target multiple risk factors or behaviours in LMICs. As we have discussed, we were unable to explore the impact of distinct components of interventions, combinations of components, or the impact of targeting different combinations of behaviours using available data; thus further research is needed to allow a more detailed understanding of the components needed for successful prevention of engagement in multiple behaviours. Last, few studies reported long‐term follow‐up data, so the extent to which effects are sustained over the longer term remains unclear.

Quality of the evidence

We rated the overall quality of evidence for each outcome using the GRADE approach and found the quality to be low or moderate, with the exception of antisocial behaviour, for which we judged the quality of the evidence to be very low overall (see summary of findings Table for the main comparison). This was due to several factors. First, a large proportion of studies were at unclear or high risk of bias owing to lack of blinding and lack of clarity around allocation concealment or reporting. As noted above, it is not always possible to blind complex public health interventions such as those included in this review, and many studies were published before reporting guidelines were available. Nevertheless, we took such factors into account when considering the quality of included evidence. Second, we downgraded the quality of the evidence on the basis of inconsistency when we found evidence of substantial heterogeneity (I²), large between‐study variance, and poor overlap between 95% confidence intervals. It is likely that there are plausible explanations for the heterogeneity observed, but we were unable to explore possible explanations further via subgroup or meta‐regression analyses (see Assessment of heterogeneity). Lastly, for certain outcomes, funnel plots suggested possible small‐study or publication bias.

It should be noted, however, that the use of GRADE to judge quality of evidence from complex public health interventions may have limitations (Movsisyan 2016; Movsisyan 2016b; Rehfeuss 2013); it is unusual for complex interventions to be rated as 'high' quality, and ratings of very low quality compared with that of simple interventions are likely (Movsisyan 2016b). For instance, as stated above, it is generally not feasible or possible to blind study participants, and heterogeneity among these complex interventions is inevitable, owing to factors such as variability in numbers and types of intervention components, modes of intervention delivery, and intervention contexts (Movsisyan 2016b; Rehfeuss 2013).

Potential biases in the review process

We assessed the possibility of publication bias for several outcomes when we identified a sufficient number of studies (see Risk of bias in included studies). These analyses highlighted the possibility of small‐study bias or reporting bias due to an under‐representation of smaller studies reporting negative findings. However, we conducted searches in all languages, with no geographical restrictions, and we searched a large number of databases, alongside searches of grey literature, so our search for studies was extremely thorough in identifying available studies. We contacted study authors for additional data when data were missing, and we incorporated such data into our analyses.

We note that owing to the poor quality of data reported in many studies, it was necessary to manipulate the data to conduct analyses, including re‐analysis of data to account for clustered data (e.g. for school‐based studies). We were not able to address issues around randomisation of matched clusters in such randomised controlled trials. We conducted sensitivity analyses to assess the impact of using different intracluster correlation coefficients (ICCs), and findings were unchanged in most cases. In three instances in which summary effect estimates changed, they reached statistical significance, demonstrating the conservative nature of our main analyses.

Agreements and disagreements with other studies or reviews

Primary outcomes

Tobacco, alcohol, and/or drug use

Our findings support a systematic review focused on interventions that target multiple risk behaviours (tobacco use, alcohol use, drug use, and sexual risk behaviour), which reported mixed but broadly beneficial findings (Jackson 2011). Similarly, a systematic review of interventions targeting substance use, sexual risk, and antisocial behaviour reported that such interventions were broadly effective (Hale 2014), although no meta‐analysis was conducted and only trials reporting positive findings were included. A Cochrane review that examined effects of interventions on tobacco use reported that social competence curricula (odds ratio (OR) 0.52, 95% confidence interval (CI) 0.30 to 0.88) and combined social competence and social influences curricula were effective in preventing the onset of smoking (OR 0.50, 95% CI 0.28 to 0.87) (Thomas 2013); Langford 2014 also found that tobacco and multiple risk behaviour interventions within the Health‐Promoting Schools (HPS) Framework were effective in reducing tobacco use (OR 0.77, 95% CI 0.64 to 0.93; and OR 0.84, 95% CI 0.76 to 0.93, respectively). In addition, a recent systematic review identified weak evidence of benefit of social competence, social influence, and combined interventions in relation to cannabis use among young people (risk ratio (RR) 0.90, 95% CI 0.81 to 1.01; RR 0.88, 95% CI 0.72 to 1.07; and RR 0.79, 95% CI 0.59 to 1.05, respectively) (Faggiano 2014), although results showed no clear benefit in relation to hard drugs. Thus, our findings, together with a wider body of evidence, suggest that school‐based programmes targeting single or multiple behaviours may be effective in preventing smoking in adolescent populations.

Our finding regarding the impact of universal school‐based programmes in relation to alcohol use contrasts with those reported in other reviews. For instance, Foxcroft 2011 reported that psychosocial and developmental prevention programmes can be effective, but that findings were mixed overall, with some studies reporting statistically significant effects and others reporting no effect. Additional systematic reviews have similarly highlighted the mixed findings of reviews of school‐based or education‐ or skills‐based prevention interventions, with no clear pattern distinguishing effective from ineffective interventions (Martineau 2013; Stockings 2016), although Strom 2014 found a small but favourable effect among school‐based interventions reporting continuous outcomes and no effect among those reporting categorical outcomes. Langford 2014 found no effect of interventions testing the HPS framework to address alcohol use or multiple risk behaviours (OR 0.72, 95% CI 0.34 to 1.52; OR 0.75, 95% CI 0.55 to 1.02, respectively).

In addition, although we identified few studies overall, we found an absence of evidence to support the effectiveness of family‐based interventions in preventing tobacco or alcohol use. Systematic reviews of alcohol‐focused parenting interventions have reported small but consistently beneficial effects (Allen 2016; Foxcroft 2011b; Kuntsche 2016; Stockings 2016), and a systematic review of family‐based programmes reported a positive effect on prevention of smoking among children and adolescents (Thomas 2015). The contrast in these findings may reflect the combinations of behaviours addressed in this review; our companion review will further investigate the impact of family‐based interventions targeted to substance use (Hickman 2014).

Antisocial behaviour and offending

Evidence indicates that universal school‐based programmes may show benefit in reducing antisocial behaviour or offending compared to usual practice. Evidence was less certain in relation to violent offences or school delinquency only, but the summary effect estimate was in the direction of benefit. Our findings support others who have reported that school‐based programmes are effective in reducing aggressive behaviour (standardised mean difference (SMD) ‐0.41, 95% CI ‐0.56 to ‐0.26), with effects maintained at 12 months' follow‐up (Mytton 2006). Langford 2014 reported that multi‐component school‐based interventions reduce bullying victimisation and show promise in reducing perpetration of bullying. Another review has reported beneficial outcomes of parenting programmes aimed at reducing early conduct problems in children three to 12 years of age (SMD ‐0.44, 95% CI ‐0.77 to ‐0.11) (Furlong 2012), and family and parenting programmes have been reported to reduce the time spent in institutions (weighted mean difference (WMD) 51.34 days, 95% CI 72.52 to 30.16) and the risk of re‐arrest (RR 0.66, 95% CI 0.44 to 0.98) (Woolfenden 2009).

Sexual risk behaviour

Low‐quality evidence suggests that multiple risk behaviour interventions may have little or no effect in reducing sexual risk behaviour, although the average effect was in the direction of benefit and the size of the summary point estimates would be consistent with potential public health benefit at the population level. Among universal school‐based interventions, Positive Action showed a strong effect (Beets 2009), and one of two active intervention arms in the Aban Aya trial showed a beneficial effect (Flay 2004a); these study authors reported effects in boys but not girls. The Gatehouse Study, All Stars, and PATHS reported null findings (Bond 2004; McNeal 2004; Shek 2011). Our findings support those of a previous systematic review of multiple risk behaviour interventions (targeting substance use and sexual risk behaviour) (Jackson 2011), which highlighted mixed findings for sexual risk behaviour, including condom use, sexual partners, having had sexual intercourse, and teenage pregnancy.

Physical activity and unhealthy diet

We found moderate‐quality evidence showing that universal school‐based multiple risk behaviour interventions are likely to be effective in increasing vigorous activity or fitness among young people (four studies; 6441 participants), with findings suggesting a possible 32% increase in this outcome associated with such interventions compared to control or usual practice. A Cochrane review of interventions using the HPS Framework also found that physical activity and nutrition interventions are effective in increasing physical activity in students (Langford 2014); Dobbins 2013 reported that school‐based programmes could have small but positive effects on physical activity, with possible increases of just under five to 45 minutes more moderate to vigorous physical activity per week, although most studies used self‐reported measures. We did not find that interventions had a positive effect on nutrition or body mass index (BMI), but few studies addressed these outcomes and could be included in the meta‐analysis (e.g. three universal school‐based interventions).

Secondary outcome

Mental illness and educational attainment

We noted an absence of evidence to support the effectiveness of school‐based interventions in reducing depressive symptoms, although, on average, targeted family programmes appeared to have a beneficial effect over the longer term (four studies; 2386 participants). We found few studies that addressed educational attainment; therefore, although we found no benefit, this finding should be interpreted with caution.

Logic Model: interventions to prevent multiple risk behaviours in individuals aged 8 to 25 years.
Figuras y tablas -
Figure 1

Logic Model: interventions to prevent multiple risk behaviours in individuals aged 8 to 25 years.

Study flow diagram (searches conducted in 2012, 2015, and 2016).
Figuras y tablas -
Figure 2

Study flow diagram (searches conducted in 2012, 2015, and 2016).

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Figuras y tablas -
Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Figuras y tablas -
Figure 4

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Funnel plot. Outcome 7: antisocial behaviour and offending (short‐term): universal school‐based interventions.
Figuras y tablas -
Figure 5

Funnel plot. Outcome 7: antisocial behaviour and offending (short‐term): universal school‐based interventions.

Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.1: antisocial behaviour and offending ‐ any (short‐term).
Figuras y tablas -
Figure 6

Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.1: antisocial behaviour and offending ‐ any (short‐term).

Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.2: violent offences (short‐term).
Figuras y tablas -
Figure 7

Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.2: violent offences (short‐term).

Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.3: school or general delinquency (short‐term).
Figuras y tablas -
Figure 8

Funnel plot of comparison. Outcome 7: antisocial behaviour and offending. Outcome 7.3: school or general delinquency (short‐term).

Funnel plot of comparison. Outcome 1: tobacco. Outcome 1.1: tobacco use (short‐term).
Figuras y tablas -
Figure 9

Funnel plot of comparison. Outcome 1: tobacco. Outcome 1.1: tobacco use (short‐term).

Funnel plot of comparison. Outcome 2: alcohol. Outcome 2.1: alcohol use (short‐term).
Figuras y tablas -
Figure 10

Funnel plot of comparison. Outcome 2: alcohol. Outcome 2.1: alcohol use (short‐term).

Funnel plot of comparison. Outcome 4: cannabis use. Outcome 4.1: cannabis use (short‐term).
Figuras y tablas -
Figure 11

Funnel plot of comparison. Outcome 4: cannabis use. Outcome 4.1: cannabis use (short‐term).

Funnel plot of comparison. Outcome 5: illicit drug use. Outcome 5.1: Illicit drug use (short‐term).
Figuras y tablas -
Figure 12

Funnel plot of comparison. Outcome 5: illicit drug use. Outcome 5.1: Illicit drug use (short‐term).

Funnel plot of comparison. Outcome 9: sexual risk behaviours. Outcome 9.1: sexual risk behaviour (short‐term).
Figuras y tablas -
Figure 13

Funnel plot of comparison. Outcome 9: sexual risk behaviours. Outcome 9.1: sexual risk behaviour (short‐term).

Comparison 1 Tobacco, Outcome 1 Tobacco Use (short‐term).
Figuras y tablas -
Analysis 1.1

Comparison 1 Tobacco, Outcome 1 Tobacco Use (short‐term).

Comparison 1 Tobacco, Outcome 2 Tobacco Use (long‐term).
Figuras y tablas -
Analysis 1.2

Comparison 1 Tobacco, Outcome 2 Tobacco Use (long‐term).

Comparison 2 Alcohol, Outcome 1 Alcohol Use (short‐term).
Figuras y tablas -
Analysis 2.1

Comparison 2 Alcohol, Outcome 1 Alcohol Use (short‐term).

Comparison 2 Alcohol, Outcome 2 Alcohol Use (long‐term).
Figuras y tablas -
Analysis 2.2

Comparison 2 Alcohol, Outcome 2 Alcohol Use (long‐term).

Comparison 3 Binge drinking, Outcome 1 Drunkenness or Excess Drinking (short‐term).
Figuras y tablas -
Analysis 3.1

Comparison 3 Binge drinking, Outcome 1 Drunkenness or Excess Drinking (short‐term).

Comparison 3 Binge drinking, Outcome 2 Drunkenness or Excess Drinking (long‐term).
Figuras y tablas -
Analysis 3.2

Comparison 3 Binge drinking, Outcome 2 Drunkenness or Excess Drinking (long‐term).

Comparison 4 Illicit drug use, Outcome 1 Illicit drug use (short‐term).
Figuras y tablas -
Analysis 4.1

Comparison 4 Illicit drug use, Outcome 1 Illicit drug use (short‐term).

Comparison 4 Illicit drug use, Outcome 2 Illicit drug use (long‐term).
Figuras y tablas -
Analysis 4.2

Comparison 4 Illicit drug use, Outcome 2 Illicit drug use (long‐term).

Comparison 5 Cannabis use, Outcome 1 Cannabis Use (short‐term).
Figuras y tablas -
Analysis 5.1

Comparison 5 Cannabis use, Outcome 1 Cannabis Use (short‐term).

Comparison 5 Cannabis use, Outcome 2 Cannabis Use (long‐term).
Figuras y tablas -
Analysis 5.2

Comparison 5 Cannabis use, Outcome 2 Cannabis Use (long‐term).

Comparison 6 Alcohol, tobacco, and/or drug use, Outcome 1 Composite Substance Use (short‐term).
Figuras y tablas -
Analysis 6.1

Comparison 6 Alcohol, tobacco, and/or drug use, Outcome 1 Composite Substance Use (short‐term).

Comparison 6 Alcohol, tobacco, and/or drug use, Outcome 2 Composite Substance Use (long‐term).
Figuras y tablas -
Analysis 6.2

Comparison 6 Alcohol, tobacco, and/or drug use, Outcome 2 Composite Substance Use (long‐term).

Comparison 7 Antisocial behaviour and offending, Outcome 1 Antisocial Behaviour and Offending ‐ Any (short‐term).
Figuras y tablas -
Analysis 7.1

Comparison 7 Antisocial behaviour and offending, Outcome 1 Antisocial Behaviour and Offending ‐ Any (short‐term).

Comparison 7 Antisocial behaviour and offending, Outcome 2 Violent Offences.
Figuras y tablas -
Analysis 7.2

Comparison 7 Antisocial behaviour and offending, Outcome 2 Violent Offences.

Comparison 7 Antisocial behaviour and offending, Outcome 3 School or General Delinquency.
Figuras y tablas -
Analysis 7.3

Comparison 7 Antisocial behaviour and offending, Outcome 3 School or General Delinquency.

Comparison 7 Antisocial behaviour and offending, Outcome 4 Antisocial Behaviour and Offending ‐ Any (long term).
Figuras y tablas -
Analysis 7.4

Comparison 7 Antisocial behaviour and offending, Outcome 4 Antisocial Behaviour and Offending ‐ Any (long term).

Comparison 8 Vehicle‐related risk behaviours, Outcome 1 Driving Under the Influence of Alcohol and/or Drugs.
Figuras y tablas -
Analysis 8.1

Comparison 8 Vehicle‐related risk behaviours, Outcome 1 Driving Under the Influence of Alcohol and/or Drugs.

Comparison 9 Sexual risk behaviours, Outcome 1 Sexual Risk Behaviour (short‐term).
Figuras y tablas -
Analysis 9.1

Comparison 9 Sexual risk behaviours, Outcome 1 Sexual Risk Behaviour (short‐term).

Comparison 9 Sexual risk behaviours, Outcome 2 Sexual Risk Behaviour (long‐term).
Figuras y tablas -
Analysis 9.2

Comparison 9 Sexual risk behaviours, Outcome 2 Sexual Risk Behaviour (long‐term).

Comparison 10 Physical activity, Outcome 1 Physical Activity.
Figuras y tablas -
Analysis 10.1

Comparison 10 Physical activity, Outcome 1 Physical Activity.

Comparison 11 Mental health, Outcome 1 Depressive Symptoms (short‐term).
Figuras y tablas -
Analysis 11.1

Comparison 11 Mental health, Outcome 1 Depressive Symptoms (short‐term).

Comparison 11 Mental health, Outcome 2 Depressive Symptoms (long‐term).
Figuras y tablas -
Analysis 11.2

Comparison 11 Mental health, Outcome 2 Depressive Symptoms (long‐term).

Comparison 12 Unhealthy diet, Outcome 1 BMI.
Figuras y tablas -
Analysis 12.1

Comparison 12 Unhealthy diet, Outcome 1 BMI.

Comparison 12 Unhealthy diet, Outcome 2 Unhealthy Diet.
Figuras y tablas -
Analysis 12.2

Comparison 12 Unhealthy diet, Outcome 2 Unhealthy Diet.

Comparison 13 School‐related outcomes, Outcome 1 Academic Performance (short‐term).
Figuras y tablas -
Analysis 13.1

Comparison 13 School‐related outcomes, Outcome 1 Academic Performance (short‐term).

Summary of findings table for the effectiveness of targeted individual‐level multiple risk behaviour interventions compared to usual practice for outcomes up to 12 months post intervention

Patient or population: children and young people aged 0 to 18 years

Settings: varied settings (home, kindergarten, primary school, secondary school, clinic, community)

Intervention: multiple risk behaviour interventions

Comparison: no intervention/usual practice

Outcomes

Risk with usual practice

Risk with intervention

(95% CI)

Relative effect
(95% CI)

No. of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Tobacco use

156 per 1000

191 per 1000

(122 to 288)

OR 1.28
(0.75 to 2.19)

521
(2 RCTs)

⊕⊕⊕⊝
Moderatea

Alcohol use

613 per 1000

618 per 1000

(559 to 675)

OR 1.02
(0.80 to 1.31)

1204
(4 RCTs)

⊕⊕⊕⊝
Moderatea

Cannabis use

110 per 1000

120 per 1000

(79 to 179)

OR 1.10
(0.69 to 1.76)

126
(2 RCTs)

⊕⊕⊕⊝
Moderatea

Illicit drug use

32 per 1000

30 per 1000

(23 to 400)

OR 0.94
(0.71 to 1.25)

638
(3 RCTs)

⊕⊕⊕⊝
Moderatea

Antisocial behaviour

145 per 1000

170 per 1000

(135 to 213)

OR 1.21
(0.92 to 1.60)

764
(4 RCTs)

⊕⊕⊕⊝
Moderatea

Vehicle‐related risk behaviour

81 per 1000

49 per 1000

(12 to 179)

OR 0.59

(0.14 to 2.48)

94
(2 RCTs)

⊕⊝⊝⊝
Very lowb

Sexual risk behaviour

610 per 1000

533 per 1000

(434 to 628)

OR 0.73
(0.49 to 1.08)

494
(2 RCTs)

⊕⊕⊕⊝
Moderatea

Physical activity

134 per 1000

N/A

No studies in meta‐analysis

aDowngraded owing to high risk of bias due to lack of blinding and/or unclear risk of bias across additional domains.

bDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains, as well as imprecision related to width of the 95% confidence interval of the summary estimate and inconsistency between effect estimates (I² = 81%).
Note that variation was evident in measures of risk with usual practice. Baseline risk measures were calculated at follow‐up. When no data were reported for any study in that meta‐analysis, baseline measures were used.

CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Figuras y tablas -

Summary of findings table for the effectiveness of universal individual‐level multiple risk behaviour interventions compared to usual practice for outcomes up to 12 months post intervention

Patient or population: children and young people aged 0 to 18 years

Setting: varied settings (home, clinic, community)

Intervention: multiple risk behaviour interventions

Comparison: no intervention/usual practice

Outcomes

Risk with usual practice

Risk with intervention

(95% CI)

Relative effect
(95% CI)

No. of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Tobacco use

32 per 1000

33 per 1000

(10 to 98)

OR 1.03
(0.32 to 3.27)

1549
(2 RCTs)

⊕⊕⊝⊝
Lowa

Alcohol use

41 per 1000

33 per 1000

(24 to 45)

OR 0.80
(0.58 to 1.11)

1911
(4 RCTs)

⊕⊕⊕⊝
Moderateb

Cannabis use

264 per 1000

198 per 1000

(142 to 272)

OR 0.69
(0.46 to 1.04)

362
(2 RCTs)

⊕⊕⊕⊝
Moderateb

Illicit drug use

‐‐

N/A

No studies in meta‐analysis

Antisocial behaviour

131 per 1000

133 per 1000

(85 to 203)

OR 1.02

(0.62 to 1.69)

200

(1 RCT)

⊕⊕⊕⊝
Moderateb

Sexual risk behaviour

396 per 1000

216 per 1000

(84 to 450)

OR 0.42
(0.14 to 1.25)

162
(1 RCT)

⊕⊕⊕⊝
Moderateb

Physical activity

No data available to estimate risk

N/A

OR 1.11
(0.74 to 1.67)

1,530
(2 RCTs)

⊕⊕⊕⊝
Moderateb

aDowngraded owing to high risk of bias in relation to blinding and incomplete outcome data. We also downgraded the certainty of evidence owing to inconsistency because between‐study variance was high and variability was evident in the effect estimates of each study. The 95% CIs of one of the studies were wide, but researchers reported very few events, so certainty of evidence was not downgraded on this basis.

bDowngraded owing to high risk of bias due to lack of blinding and/or unclear risk of bias across additional domains.

Note that variation was evident in measures of risk with usual practice. Baseline risk measures were calculated at follow‐up. When no data were reported for any study in that meta‐analysis, baseline measures were used.

CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Figuras y tablas -

Summary of findings table for the effectiveness of targeted family‐level multiple risk behaviour interventions compared to usual practice for outcomes up to 12 months post intervention

Patient or population: children and young people aged 0 to 18 years

Setting: varied settings (home, community)

Intervention: multiple risk behaviour interventions

Comparison: no intervention/usual practice

Outcomes

Risk with usual practice

Risk with intervention

(95% CI)

Relative effect
(95% CI)

No. of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Tobacco use

176 per 1000

143 per 1000

(79 to 246)

OR 0.78
(0.40 to 1.53)

313
(2 RCTs)

⊕⊕⊕⊝
Moderatea

Alcohol use

269 per 1000

234 per 1000

(147 to 349)

OR 0.83
(0.47 to 1.46)

417
(3 RCTs)

⊕⊕⊕⊝
Moderatea

Cannabis use

180 per 1000

183 per 1000

(102 to 307)

OR 1.02
(0.52 to 2.02)

380
(3 RCTs)

⊕⊕⊝⊝
Lowb

Illicit drug use

265 per 1000

211 per 1000

(132 to 321)

OR 0.74
(0.42 to 1.31)

69
(1 RCT)

⊕⊕⊕⊝
Moderatea

Antisocial behaviour

291 per 1000

256 per 1000

(190 to 337)

OR 0.84
(0.57 to 1.24)

772
(5 RCTs)

⊕⊕⊕⊝
Moderatea

Sexual risk behaviour

750 per 1000

728 per 1000

(623 to 812)

OR 0.89
(0.55 to 1.44)

371
(3 RCTs)

⊕⊕⊕⊝
Moderatea

Physical activity

No data available to estimate risk

N/A

OR 0.72
(0.29 to 1.79)

61
(1 RCT)

⊕⊕⊕⊝
Moderatea

aDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains.

bDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains. The quality of the evidence was also downgraded on the basis of inconsistency because between‐study variance was high, and although I² was moderate, inconsistency was evident in effect estimates of individual studies, two of which had small sample sizes.

Note that variation was evident in measures of risk with usual practice. Baseline risk measures were calculated at follow‐up. When no data were reported for any study in that meta‐analysis, baseline measures were used.

CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Figuras y tablas -

Summary of findings table for the effectiveness of targeted school‐level multiple risk behaviour interventions compared to usual practice for outcomes up to 12 months post intervention

Patient or population: children and young people aged 0 to 18 years

Setting: school

Intervention: multiple risk behaviour interventions

Comparison: no intervention/usual practice

Outcomes

Risk with usual practice

Risk with intervention

(95% CI)

Relative effect
(95% CI)

No. of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Tobacco use

‐‐

‐‐

No data in meta‐analysis

Alcohol use

‐‐

‐‐

No data in meta‐analysis

Cannabis use

‐‐

‐‐

No data in meta‐analysis

Illicit drug use

50 per 1000

38 per 1000 (27 to 53)

OR 0.75
(0.53 to 1.06)

2454
(3 RCTs)

⊕⊕⊕⊝
Moderatea

Antisocial behaviour

No data available to estimate risk

N/A

OR 0.78
(0.59 to 1.05)

1,531
(3 RCTs)

⊕⊕⊕⊝
Moderatea

Sexual risk behaviour

‐‐

‐‐

No data in meta‐analysis

Physical activity

‐‐

‐‐

No data in meta‐analysis

aDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains.

Note that variation was evident in measures of risk with usual practice. Baseline risk measures were calculated at follow‐up. When no data were reported for any study in that meta‐analysis, baseline measures were used.

CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Figuras y tablas -

Summary of findings table for the effectiveness of universal school‐level multiple risk behaviour interventions compared to usual practice for outcomes up to 12 months post intervention

Patient or population: children and young people aged 0 to 18 years

Setting: school

Intervention: multiple risk behaviour interventions

Comparison: no intervention/usual practice

Outcomes

Risk with usual practice

Risk with intervention

(95% CI)

Relative effect
(95% CI)

No. of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Tobacco use

54 per 1000

42 per 1000

(33 to 52)

OR 0.77
(0.60 to 0.97)

15,354
(9 RCTs)

⊕⊕⊕⊝
Moderatea

Alcohol use

163 per 1000

123 per 1000

(98 to 152)

OR 0.72
(0.56 to 0.92)

8751
(8 RCTs)

⊕⊕⊕⊝
Moderatea

Cannabis use

110 per 1000

89 per 1000

(71 to 111)

OR 0.79
(0.62 to 1.01)

4140
(5 RCTs)

⊕⊕⊕⊝
Moderatea

Illicit drug use

41 per 1000

30 per 1000

(21 to 44)

OR 0.73
(0.50 to 1.07)

10,266
(5 RCTs)

⊕⊕⊝⊝
Lowb

Antisocial behaviour

172 per 1000

141 per 1000

(117 to 168)

OR 0.79
(0.64 to 0.97)

17,722
(11 RCTs)

⊕⊝⊝⊝
Very lowc

Sexual risk behaviour

131 per 1000

112 per 1000

(87 to 146)

OR 0.84
(0.63 to 1.13)

12,633
(6 RCTs)

⊕⊕⊝⊝
Lowd

Physical activity

276 per 1000

335 per 1000

(307 to 364)

OR 1.32
(1.16 to 1.50)

6,441
(4 RCTs)

⊕⊕⊕⊝
Moderatea

aDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains.

bDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains. Downgraded an additional level on the basis of inconsistency because substantial heterogeneity was evident (I² = 69%, Chi² = 15.88, P = 0.007), between‐study variance was moderate, and inconsistency between effect estimates of individual studies was apparent, with absence of overlap between 95% CIs of certain studies in the subgroup.

cDowngraded owing to high risk of bias on the basis of blinding and/or high or unclear risk of bias across additional domains. The quality of evidence was also downgraded on the basis of inconsistency because heterogeneity was substantial (I² = 68%, Chi² = 36.95, P = 0.0002), between‐study variance was moderate, and lack of overlap was apparent between 95% CIs for certain studies with large sample sizes. Last, evidence was downgraded on the basis of possible publication or small‐study bias.

dDowngraded owing to high risk of bias in relation to blinding and/or other domains. Certainty of the evidence was also downgraded owing to substantial heterogeneity (I² = 84%, Chi² = 25.07, P < 0.0001) and high between‐study variance, with lack of overlap between the 95% CIs of certain studies in the subgroup. Although there may be plausible explanations for such heterogeneity, these reasons could not be further investigated in this review.

Note that variation was evident in measures of risk with usual practice. Baseline risk measures were calculated at follow‐up. When no data were reported for any study in that meta‐analysis, baseline measures were used.

CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial.

GRADE Working Group grades of evidence
High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.

Figuras y tablas -
Table 1. Intracluster correlation coefficients

Study

Country

Age

Outcome variable

Reported intracluster correlation coefficient

Published or correspondence (comment)

ICCs used in primary analyses

Gatehouse Study (Bond 2004)

Australia

13‐14

Substance use

0.06

Published

All Stars 2 (Gottfredson 2010)

USA

11‐14

Aggression

0.025

Published

Fourth R (Wolfe 2012)

USA

14‐15

Violence

0.01

Published

All Stars 2 (Gottfredson 2010)

USA

11‐14

Delinquency

0.025

Published

Fourth R (Wolfe 2012)

USA

14‐15

Sexual risk behaviour

0.01

Published

Gatehouse Study (Bond 2004)

Australia

13‐14

Diet/physical activity

0.06

Publisheda

Positive Action (Chicago) (Li 2011)

USA

8‐13

Education

0.1

Published

Gatehouse Study (Bond 2004)

Australia

13‐14

Mental illness

0.01

Published

ICCs used in sensitivity analyses

LIFT/All Stars 2 (DeGarmo 2009; Gottfredson 2010)

USA

10/11‐14

Substance use

0.0

Published

All Stars 2 (Gottfredson 2010)

USA

11‐14

Aggression

0.0

Published

Fourth R (Wolfe 2012)

USA

14‐15

Violence

0.01

Published

All Stars 2 (Gottfredson 2010)

USA

11‐14

Delinquency

0.0

Published

Fourth R (Wolfe 2012)

USA

14‐15

Sexual risk behaviour

0.01

Published

All Stars 2, Gatehouse Study, Fourth R,

LIFT, Positive Action (Chicago) (Bond 2004; Gottfredson 2010; Wolfe 2012; DeGarmo 2009; Li 2011)

USA,

Australia

10‐15

Diet/physical activity

0.0263

Publishedb

Gatehouse Study (Bond 2004)

Australia

13‐14

Education

0.01

Publishedc

Gatehouse Study (Bond 2004)

Australia

13‐14

Mental illness

0.01

Published

ICC: intracluster correlation coefficient.

aThe highest ICC value was used to be conservative.

bAverage ICC value used from across these studies.

cICC related to school engagement.

Figuras y tablas -
Table 1. Intracluster correlation coefficients
Table 2. Outcomes not included in meta‐analysis

Author and year

Study name

Categorisation

Outcome

Authors' conclusions

1. Tobacco use

Bonds 2010

New Beginnings

Family‐Targeted

Tobacco use disorder (including nicotine withdrawal and dependence)

No difference between study arms in the proportion of participants meeting criteria for nicotine use disorder (6.7% in each arm)

Bush 1989

Know Your Body

School‐Universal

Serum thiocyanate (micromoles/L)

Mean difference from baseline to 1 year follow‐up was ‐9.87 (SE 2.5) in the intervention group, and 20.03 (SE 2.68) in the control group (P < 0.001). These data were based on a 50% subsample stratified at baseline, based on measurement after 1 year of intervention.

Connell 2007

Family Check‐Up

Family‐Universal

Nicotine abuse/dependence

Across treatment and control groups, no significant differences were found for nicotine abuse/dependence (Chi² (1, 998) = 3.09, P > 0.05). No significant correlation between assignment to experimental condition(s) and tobacco use over time

DeGarmo 2009

LIFT

School‐Universal

Initiation of tobacco use

With controls for parental drinking and deviant peer association, the intervention was associated with reduced risk of initiation of tobacco use (beta = ‐0.10, P < 0.01). The effect translated to odds ratios of a 10% reduction in risk for tobacco use.

Estrada 2015

Brief Familias Unidas

Family‐Targeted

Tobacco use in past 90 days

Brief Familias Unidas was not significantly efficacious in reducing tobacco use (beta = ‐0.09, P = 0.85) in the past 90 days.

Gonzales 2012

Bridges to High School

Family‐Targeted

Substance use

Study authors report that substance use at follow‐up was less in the intervention group than in the control group for adolescents who engaged in high levels (85th percentile) of baseline substance use (d = 3.65).

LoSciuto 1999

Woodrock Youth Development Project

School‐Universal

Substance use in past month (tobacco, alcohol, drugs)

Mean substance use in the past month was 1.1 for the intervention group and 1.15 for the control group (SMD 0.18)

McNeal 2004

All Stars

School‐Universal

Tobacco use in past 30 days

The teacher‐delivered All Stars programme was associated with reduced rate of growth in 30‐day usage of cigarettes (7.4% to 7.8%) compared to the specialist condition (11.0% to 13.8%) and the control group (15.1% to 17.9%).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Mean cigarettes per day

15‐year follow‐up: incidence of cigarettes smoked per day in past 6 months among those who received nurse visitation through pregnancy (group 3) was 0.91 compared to 1.30 among control participants (P = 0.49). Among a subgroup of women from low socioeconomic status (SES) households who were unmarried, the comparison was 1.32 vs 2.50 among control participants (P = 0.07). Incidence of cigarettes smoked per day in the past 6 months among those who received nurse visitation until the child's second birthday was 1.28 compared to 1.30 among control participants (P = 0.76). Subgroup analysis of women from low SES households who were unmarried showed that incidence was 1.50 among the intervention group compared to 2.50 among controls (P = 0.1).

Perry 2003

DARE and DARE‐Plus

School‐Universal

Current smoker (growth rate)

Growth curve analysis showed that for boys: the growth rate of tobacco use was 0.31 (0.05) in the control group, 0.28 (0.05) in the DARE group, and 0.18 (0.05) in the DARE Plus group (DARE vs control P = 0.28; DARE Plus vs control P = 0.02; DARE Plus vs DARE P = 0.08). Among girls: the growth rate was 0.28 (0.07) in the control group, 0.25 (0.07) in the DARE group, and 0.22 (0.07) in the DARE Plus group (DARE vs control P = 0.38; DARE Plus vs control P = 0.25; DARE plus vs DARE P = 0.35).

Piper 2000

Healthy for Life

School‐Universal

Tobacco use in past 30 days

The age‐appropriate condition showed no benefit over the control condition at 12‐month follow‐up (prevalence 24% in both arms; HLM coefficient 0.18, SE 0.12, P > 0.1) or at 24‐month follow‐up, where prevalence was higher in the intervention group (prevalence 36% vs 30% in the control group, coefficient 0.41, SE 0.2, P < 0.1). Among those receiving the intensive condition, prevalence was similar in both study arms (12 months: 22% vs 24% in the control group; coefficient ‐0.3, SE 0.17, P > 0.1; 24 months: 28% vs 30% in the control arm; coefficient ‐0.38, SE 0.15, P < 0.05).

Saraf 2015

(none given)

School‐Universal

Tobacco use

Current smoking (in the past month) changed from 13.1% (95% CI 10.2% to 15.9%) to 3.1% (95% CI 0.2% to 5.9%) in the intervention group; and from 7.7% (95% CI 5.0% to 10.4%) to 5.4% (95% CI 2.6% to 8.2%) in the control group (overall difference between groups in pre‐ to post‐change ‐7.7 (‐10.7 to ‐4.7); P < 0.01.

Schweinhart 1980

High/Scope Perry Preschool Study

School‐Targeted

Tobacco use

No impact of the intervention on smoking cigarettes 22 years after the end of the programme: 45% of those in the intervention group smoked compared to 56% of those in the control group (P = 0.231). Effect size 0.22

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Likelihood of smoking

Those receiving the intervention were reported to be 19.7% less likely to start smoking compared to controls (males receiving Big Brothers Big Sisters were 24.5% less likely to start smoking, and females 9.9%). Males from an ethnic minority receiving Big Brothers Big Sisters had a 29.9% increased likelihood of smoking compared to controls, but among females there was a 1.9% reduction. White males and females receiving the intervention had a 47.9% and 14.7% reduced likelihood of smoking, respectively.

Walter 1989

Know Your Body

School‐Universal

Smoking

Among the schools in Westchester, results showed a beneficial impact of the intervention: the school mean at the end of the intervention was 3.5% (SD 4.3%) compared to 13.1% (SD 5.2) among control schools; P < 0.005. This is equivalent to a 73% reduction in the rate of initiation of smoking.

2. Alcohol use

Bonds 2010

New Beginnings

Family‐Targeted

Alcohol use, binge drinking, age commencing drinking

15‐year follow‐up: alcohol use in the past month higher in the intervention arm than in the control arm (d = 0.23, 95% CI ‐0.26 to 0.72). Intervention arm commenced drinking at a mean age 0.47 years younger than the control group (95% CI ‐1.31 to 0.23 years). Binge drinking in the past year higher in the intervention group than in the control arm (d = 0.16, 95% CI ‐0.14 to 0.46).

Conduct Problems Prevention Research Group 2010

Fast Track

School‐Targeted

Binge drinking problem

The intervention marginally decreased binge drinking at 10‐year follow‐up (adjusted OR 0.75, 95% CI 0.55 to 1.01, P = 0.057).

Connell 2007

Family Check‐Up

Family‐Universal

Alcohol use

No significant association was noted between assignment to experimental condition(s) and alcohol abuse/dependence over time (Chi² (1, 998) = 0.98, P > 0.05), with the exception of Time 2, when a correlation between treatment assignment and alcohol use was observed (r = 0.09, P ≤ 0.05).

Cunningham 2012

SafERteens

Individual‐Targeted

Alcohol use

Reduction in the proportion of participants scoring ≥ 3 on AUDIT‐C from 50% at baseline to 34.4% at 3 months and 37.3% at 12 months (‐12.7% change at 12 months; OR 1.09, 95% CI 0.77 to 1.56) for those in the therapist intervention arm; and a reduction from 45.6% at baseline to 32.7% at 3 months and 28.9% at 12 months (‐16.7% change at 12 months; OR 0.95, 95% CI 0.66 to 1.37) for those in the computer arm . For controls, a reduction from 47.7% to 38.1% at 3 months and 34.7% at 12 months was evident (‐13% change at 12 months).

Cunningham 2012

SafERteens

Individual‐Targeted

Binge drinking

Reduction in the proportion of participants reporting any binge drinking from 52.8% at baseline to 34.4% at 3 months and 38.7% at 12 months (‐14.1% reduction at 12 months; OR 0.95, 95% CI 0.66 to 1.36) among those in the therapist group; and a reduction from 48.5% to 28.8% at 3 months and 30.3% at 12 months (‐18.2% reduction; OR 0.83, 95% CI 0.58 to 1.19) among those in the computer group. Similar reductions were seen in the control group: a reduction from 54% at baseline to 34.6% at 3 months and 36.1% at 12 months (‐17.9% reduction at 12 months).

Estrada 2015

Familias Unidas – Brief

Family‐Targeted

Alcohol use

Brief Familias Unidas was not significantly efficacious in reducing alcohol use (beta = 0.17; P = 0.51) in the past 90 days.

Friedman 2002

Botvin Life Skills Training and Anti‐violence

Individual‐Targeted

Degree of alcohol use

Alcohol use was decreased among intervention participants compared to controls (t = ‐1.24, P > 0.05).

Gonzales 2012

Bridges to High School

Family‐Targeted

Substance use

Study authors report that substance use was less at follow‐up in the intervention group compared to the control group for adolescents who engaged in high levels (85th percentile) of baseline substance use (d = 3.65).

Jalling 2016

Comet 12‐18

Family‐Targeted

Alcohol use (AUDIT score)

No significant difference was found between groups: at T2, mean AUDIT score was 7.59 (SD 7.60) in the intervention group vs 6.26 (SD 6.79) in the control group.

Jalling 2016b

ParentSteps

Family‐Targeted

Alcohol use (AUDIT score)

No significant difference was found between groups: at T2, mean AUDIT score was 5.10 (SD 6.38) in the intervention group vs 6.26 (SD 6.79) in the control group.

Kellam 2008

Good Behaviour Game

School‐Universal

Lifetime alcohol abuse/ dependence

The Good Behaviour Game (GBG) was associated with a reduction in lifetime alcohol abuse/dependence disorders compared to control: 13% for GBG vs 20% for controls (P = 0.08). The effect was similar for males and females.

Murry 2014

SAAF

Family‐Targeted

Escalation of alcohol use

Study authors report through structural equation modelling analysis that youth avoidance of risk opportunity situations served a role in delaying initiation and escalation of use of alcohol and other substances as they transitioned from early to late adolescence.

Monti 1999

Alcohol Screening and Brief Intervention

Individual‐Targeted

Alcohol use score

With a 2 × 2 (group × time) repeated measures analysis of variance, time effect showed reductions in alcohol scores (F(1,79) = 24.55, P < 0.001) with no group differences or interactions.

Olds 1998

Nurse Family Partnership

Family‐Targeted

Alcohol use

15‐year follow‐up: incidence of days drunk alcohol in past 6 months among those who received nurse visitation through pregnancy (group 3) was 1.81 compared to 1.57 among control participants (P = 0.97). Among a subgroup of women from low socioeconomic status (SES) households who were unmarried, the comparison was 1.84 vs 2.49 among control participants (P = 0.41). Incidence of days drunk alcohol in past 6 months among those who received nurse visitation until the child's second birthday was 1.87 compared to 1.57 among control participants (P = 0.96). Subgroup analysis of women from low SES households who were unmarried show the incidence was 1.09 among the intervention group compared to 2.49 among controls (P = 0.03).

Perry 2003

DARE vs DARE Plus

School‐Universal

Alcohol consumption in past month

Growth curve analysis showed that for boys: the growth rate in alcohol use in the past month (mean, SE) was 0.14 (0.02) for those in the control group, 0.11 (0.02) for the DARE group (P = 0.12), and 0.08 (0.02) for the DARE Plus group (P = 0.01) (DARE Plus vs DARE, P = 0.12). Among girls: values were 0.12 (0.03) for controls, 0.13 (0.02) for those in the DARE group (P = 0.40), and 0.08 (0.03) for those in the DARE Plus group (P = 0.15) (DARE Plus vs DARE, P = 0.10).

Piper 2000

Healthy for Life

School‐Universal

Alcohol use in past 30 days

Results showed a negative treatment effect at 12 months and 24 months of follow‐up: in the age‐appropriate intervention, prevalence of alcohol use in the past month was 33% in the intervention group and 28% in the control group at 12 months (hierarchical linear modelling (HLM) coefficient 0.34, SE 0.19, P < 0.1). At 24 months, the prevalence of alcohol use in the past month was 48% in the intervention group and 41% in the control group at 24 months (HLM coefficient 0.3, SE 0.14, P < 0.05). In the intensive version of the intervention, the prevalence of alcohol use at 12 months was 33% vs 28% in the control arm (HLM coefficient 0.2, SE 0.09, P < 0.05), and at 24 months, prevalence was 45% vs 41% in the control arm (HLM coefficient 0.27, SE 0.1, P < 0.05).

Schweinhart 1980

High/Scope Perry Preschool Study

School‐Targeted

Alcohol use

No impact of the intervention on drinking alcoholic beverages several or more times a week 22 years after the end of the programme: 16% of those in the intervention group drank alcohol several or more times a week compared to 26% of those in the control group. Effect size for drinking alcoholic beverages was 0.27 (P = 0.141).

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Likelihood of initiating alcohol use

Those receiving the intervention were 27.4% less likely to start using alcohol than those in the control group (19.2% reduction in likelihood among males and 38.8% among females). The reduction in likelihood was 11.4% among males from an ethnic minority, 53.7% among females from an ethnic minority; 34.5% among white males, and 8.4% among white females.

3. Illicit drug use

Connell 2007

Family Check‐Up

Family‐Universal

Marijuana use

Across treatment and control groups, no significant differences were found for marijuana abuse/dependence (Chi² (1, 998) = 0.74, P > 0.05). No significant correlation was noted between assignment to experimental condition(s) and marijuana use over time, with the exception of Time 2 (r = 0.10, P ≤ 0.05).

Bonds 2010

New Beginnings

Family‐Targeted

Marijuana use, polydrug use, other drug use

6‐year follow‐up: results showed no significant group effects for drug dependence, drug symptom count, or polydrug use (all P > 0.05).

15‐year follow‐up: intervention group displayed lower past year polydrug use (d = ‐.44, 95% CI ‐.88 to .00) and past year other drug use (d = ‐.06, 95% CI ‐.11 to ‐.00) compared to control group. No difference was observed for marijuana use between intervention and control groups (d = .00, 95% CI ‐.47 to .47).

DeGarmo 2009

LIFT

School‐Universal

Percentage of participants who have not used marijuana

One year post intervention, 2.2% had not used marijuana in the past year compared to 2.3% in the control group.

Estrada 2015

Brief Familias Unidas

Family‐Targeted

Illicit drug use (past 90 days)

Brief Familias Unidas was not significantly efficacious in reducing illicit drug use (beta = 0.03; P = 0.93) in the past 90 days.

Friedman 2002

Botvin Life Skills Training and Anti‐violence

Individual‐Targeted

Degree of drug use and involvement in selling of drugs

Among intervention participants compared to controls, data showed a greater reduction in drug use (t = ‐2.58, P < 0.01) and a greater reduction in the frequency of involvement in the selling of drugs (t = ‐1.99).

Freudenberg 2010

REAL MEN

Individual‐Targeted

Daily marijuana use in past 90 days

Intervention associated with reduced odds of daily marijuana use (OR 0.751). No 95% confidence interval or standard error was provided.

Freudenberg 2010

REAL MEN

Individual‐Targeted

Hard drug use tried in past 90 days

Intervention was associated with reduced odds of trying hard drugs (OR 0.166, P < 0.05). No 95% confidence interval or standard error was provided.

Griffin 2006

Life Skills Training

School‐Universal

High‐risk substance use

32.4% of participants in the intervention group engaged in high‐risk substance use at the young adult follow‐up compared to 37.1% of those in the control group 11 years following completion of the intervention.

Jalling 2016

Comet 12‐18 and Parent Steps

Individual‐Targeted

Any illicit drug use (%)

Higher odds of illicit drug use were evident among those whose parents took part in the study, although 95% CIs were wide. Comet 12‐18: OR 3.52, 95% CI 1.23 to 10.10. ParentSteps OR 3.23, 95% CI 1.06 to 9.08

McNeal 2004

All Stars

School‐Universal

Marijuana use in past 30 days

Marijuana use in the past 30 days for those in the specialist arm increased from 3.2% to 4.1% in the intervention group and from 5.0% to 8.7% in the control group (standardised B coefficient = 0.02, P > 0.05). For those in the teacher‐delivered arm, the increase was 3.2% at baseline and follow‐up compared to a change from 5.0% to 8.7% in the control group (standardised B coefficient ‐0.01, P > 0.05).

Piper 2000

Healthy for Life

School‐Universal

Marijuana use in past 30 days

In the age‐appropriate condition, prevalence of marijuana use was similar in the intervention and control groups at 12‐month and 24‐month follow‐up (prevalence 4% vs 5% in the control group; OR 0.77, P > 0.1; and 12% vs 10% in the control group; OR 1.28, P > 0.1, respectively). Among those receiving the intensive version of the programme, findings suggested benefit of the intervention: prevalence 5% in both arms at 12 months (OR 0.56, P < 0.05) and prevalence 8% vs 10% in the control condition (OR 0.56, P < 0.05).

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Likelihood of initiating drug abuse

Overall, receiving the intervention was associated with a 45.8% reduction in the likelihood of initiating drug abuse (55% among males and 26.6% among females). The impact was greatest among males and females from an ethnic minority, among whom results showed a 67.8% and 72.6% reduced likelihood of initiating drug use, respectively. White males in the intervention group were 32.7% less likely to start using drugs compared to white males in the control group, but white females were 49.5% more likely to start using drugs compared to white females in the control group.

4. Substance misuse (composite)

Beach 2016

ProSAAF

Family‐Targeted

Substance use in lifetime (self‐reported use of cigarettes, alcohol, and/or marijuana)

At 9 months, young people in the intervention group reported lower levels of substance use initiation compared to those in the control group (coefficient ‐2.25, SE 0.64, t = 3.54, P < 0.01).

Berry 2009

Coaching for Communities

Individual‐Community

Alcohol and drug use

At the end of the intervention, the mean use of alcohol and drugs in the past 30 days was 0.83 in the intervention group and 2.55 in the control group.

Estrada 2015

Brief Familias Unidas

Family‐Targeted

Substance use (alcohol, tobacco, and/or drugs)

Growth curve analyses showed a non‐significant difference in past 90‐day substance use between brief Familias Unidas and CPC (beta = 0.24; P = 0.37).

Freudenberg 2010

REAL MEN

Individual‐Targeted

Alcohol or drug dependence in the past year

Reduced odds of alcohol or drug dependence in the past year following receipt of intervention (OR 0.519, P < 0.05). No 95% confidence interval or standard error was provided.

Gonzales 2014

Bridges to High School (Bridges/ Puentes)

Family‐Targeted

Substance use

Intervention status was associated with a reduction in substance use at 2 years and 5 years post‐test (unstandardised regression coefficients ‐0.3 and ‐0.13, respectively).

Griffin 2006

Life Skills Training

School‐Universal

High‐risk substance use

32.4% of participants in the intervention group engaged in high‐risk substance use at the young adult follow‐up compared to 37.1% of those in the control group.

LoSciuto 1999

Woodrock Youth Development Project

School‐Universal

Substance use in past month

Participation in the programme was associated with higher average scores for lifetime substance use (F(1,711) = 6.10, P = 0.01, Cohen's d = 0.19) and past month substance use (F(1,712) = 5.93, P = 0.02, Cohen's d = 0.18). The data could not be adjusted for clustering owing to insufficient information reported.

Olds 1998

Nurse Family Partnership

Family‐Targeted

Drug use

At 15‐year follow‐up, data showed no significant difference in the incidence of days of drug use in the past 6 months between intervention and control groups. Among those who received nurse visitation during pregnancy, incidence was 3.55 vs 2.28 among controls (P = 0.49) (low SES, unmarried subgroup: 9.38 vs 4.04, P = 0.01). Among those who received nurse visitation until the child's second birthday, incidence was 2.04 vs 2.28 in the control group (P = 0.54) (low SES, unmarried subgroup: 2.5 vs 4.04 among controls, P = 0.24).

5. Antisocial behaviour and offending

Averdijk 2016

Triple P

Family‐Targeted

Delinquency

No substantial effect of the intervention was found at long‐term follow‐up (age 15 years, beta = 0.004, 95% CI ‐0.15 to 0.15; ES = 0.002).

Averdijk 2016

PATHS

School‐Universal

Delinquency

No substantial effect of the intervention was found at long‐term follow‐up (age 15 years, beta = ‐0.04, 95% CI ‐0.19 to 0.11; ES = ‐0.022).

Beach 2016

ProSAAF

Family‐Targeted

Conduct problems

Follow‐up revealed a beneficial effect of the intervention on conduct problems: coefficient for conduct problems ‐0.54, SE 0.22, t = 2.42, P = 0.05.

Berry 2009

Coaching for Communities

Individual‐Targeted

Variety and volume of offending

For variety of offending, the mean in the intervention group was 3.5 vs 5.95 in the control group at the end of intervention; and for volume of offending, the mean in the intervention group was 18.1 vs 23.9 in the control group.

Conduct Disorders Prevention Research Group 2010

Fast Track

School‐Targeted

Antisocial personality disorder (ASPD)

10 years post intervention, the prevalence of being in the DSM‐IV clinical range for ASPD was lower in the intervention group than in the control group (OR 0.60, 95% CI 0.39 to 0.93, P = 0.022).

Connell 2007

Family Check‐Up

Family‐Universal

Antisocial Behaviour

Across treatment and control groups, no significant differences were found for marijuana abuse/dependence (Chi² (1, 781) = 0.69, P > 0.05). No significant correlation between assignment to experimental condition(s) and antisocial behaviour over time

Cunningham 2012

SafERteens

Individual‐Targeted

Any peer aggression

A reduction of 34.3% in the proportion reporting any severe peer aggression at 3 months (from 82.7%), increasing to a 43.3% reduction at 12 months (OR 1.36, 95% CI 0.87 to 2.12) for the therapist group. For the computer group, a reduction of 21.3% was evident at 3 months, and 26.2% at 12 months (OR 0.88, 95% CI 0.57 to 1.34). For controls, a 16.4% reduction was evident at 3 months, increasing to 25.9% at 12 months.

Cunningham 2012

SafERteens

Individual‐Targeted

Any peer victimisation or peer violence

Reduction of 10.4% at 3 months and 22.7% at 12 months for those in the therapist group (baseline 47.6%) (OR 1.25, 95% CI 0.87 to 1.79); and reduction of 2.5% at 3 months and 17.4% at 12 months for the computer group (OR 1.06, 95% CI 0.73 to 1.52). Among those in the control group, results showed a 4.7% increase at 3 months but a 12.3% reduction in reported experience of peer violence at 12 months.

DeGarmo 2009

LIFT

School‐Universal

Percentage arrested or detained

At initial follow‐up, 300 days post intervention, 0.6% of those in the intervention group had been detained or arrested vs 4.1% in the control group. 2.5 years post intervention (900 days), 5.1% had been arrested/detained in the intervention group vs 10.3% in the control group.

Friedman 2002

Botvin Life Skills Training and Anti‐violence

Individual‐Targeted

Degree of illegal offences

Among intervention participants vs controls, there was a slight reduction in the degree of illegal offences (t = ‐1.53).

Freudenberg 2010

REAL MEN

Individual‐Targeted

Criminal justice outcomes (re‐arrest, re‐incarceration, problematic behaviour)

Intervention associated with reduced odds of re‐arrest (OR 0.871). No 95% confidence interval or standard error was provided. Odds of re‐incarceration 1.019; the intervention was associated with reduced odds of often engaging in problematic behaviour (OR 0.789)

Gonzales 2014

Bridges to High School (Bridges/ Puentes)

Family‐Targeted

Externalising symptoms

Intervention associated with small reduction in externalising symptoms at 2 and 5 years post‐test (unstandardised regression coefficients ‐0.02 and ‐0.01, respectively)

Kellam 2008

Good Behaviour Game (GBG)

School‐Universal

Lifetime antisocial personality disorder (ASPD)

At ˜ 12 years' follow‐up (participants were aged 19 to 21 years), overall rates of ASPD were lower for those in the GBG groups (17%) vs internal controls (25%) (P = 0.07).

LoSciuto 1999

Woodrock Youth Development Project

School‐Universal

Aggression

No strong evidence showed a greater reduction in aggression in experimental vs control groups at post‐test (F(1, 342) = 2.95, P = 0.09, Cohen's d = 0.19). Insufficient data were available to adjust these findings for clustering of participants by classroom.

Olds 1998

Nurse Family Partnership

Family‐Targeted

Major delinquent acts

At 15‐year follow‐up, results showed no difference between intervention and control groups in the mean number of major delinquent acts committed: mean 2.79 among those who received nurse visitation through pregnancy vs 3.02 in the control group (P = 0.93). Among a subgroup of women from low socioeconomic status (SES) households who were unmarried, the comparison was 3.45 vs 4.09 (P = 0.60).

Among those receiving nurse visitation through to the child's second birthday, the comparison was 3.57 vs 3.02 (P = 0.48). Among a subgroup of women from low SES households who were unmarried, the comparison was 3.99 vs 4.09 (P = 0.77).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Mean number of arrests

Differences between groups were evident regarding the incidence of arrests at 15‐year follow‐up. For those visited during pregnancy, the incidence of arrests among children was 0.16 vs 0.36 (P = 0.005); among a subgroup of women from low SES households who were unmarried, the comparison was 0.15 vs 0.45 (P = 0.02).

Among those visited through pregnancy and infancy, their children were arrested a mean of 0.17 times vs 0.36 times among controls (P = 0.005); and among a subgroup of women from low SES households who were unmarried, the comparison was 0.20 vs 0.45 (P = 0.03).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Externalising problems

At 15‐year follow‐up, results showed no difference between intervention and control groups in the mean number of externalising problems: mean 13.65 among those who received nurse visitation through pregnancy vs 13.73 in the control group (P = 0.95). Among a subgroup of women from low socioeconomic status (SES) households who were unmarried, the comparison was 15.63 vs 14.18 (P = 0.42).

Among those receiving nurse visitation through to the child's second birthday, the comparison was 13.88 vs 13.73 (P = 0.89) Among a subgroup of women from low SES households who were unmarried, the comparison was 11.85 vs 14.18 (P = 0.17).

Perry 2003

DARE vs DARE Plus

School‐Universal

Physical victimisation

Among boys, those in DARE‐Plus schools were less likely than those in control schools to show increases in victimisation (growth rate ‐0.1, SE 0.04, P = 0.02); there was no difference between DARE and control (growth curve rate, mean ‐0.03, SE 0.04, P = 0.18). No differences were evident between groups among girls.

Schweinhart 1980

High/Scope Perry Preschool Program

School‐Targeted

Carried a gun or knife once or more often

At 10‐year follow‐up (when participants were ˜ age 15), 13 of 44 (29.5%) in the intervention group had carried a gun or knife once or more compared to 15 of 55 (27%) in the control group.

Shetgiri 2011

[No study name]

School‐Targeted

Been in trouble with the police in the past 12 months

Eighteen per cent of those in the intervention group had been in trouble with the police in the past 12 months at follow‐up post intervention (21% at baseline) compared to 26% of those in the control group at follow‐up (32% at baseline) (P = 0.41).

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Hitting, stealing, and damaging property

Participation in the intervention was associated with a 32% reduction in the number of times participants hit someone compared to control. The percentage reduction was greater in females than in males (43% vs 25%). Results showed a greater reduction among females from an ethnic minority than among white females (48% reduction vs 2% reduction), and a greater reduction was observed in white males (45%) than in males from an ethnic minority (4%). Data show a 19% reduction in the number of times participants in the intervention group vs the control group stole something and little change (0.15% reduction) in the number of times participants damaged property. Findings show a 16% reduction in the number of times participants in the intervention group took something from a store compared to controls, and a 17% reduction in the number of times participants did risky things. Little change was evident in relation to behavioural conduct (1% reduction in intervention vs control) and the number of times participants were involved in a fight (1% reduction in intervention vs control).

6. Vehicle‐related risk behaviour

Schweinhart 1980

High/Scope Perry Preschool Study

School‐Targeted

Wearing seatbelt

Among those in the intervention group, 24 of 56 (43%) wore a seatbelt sometimes or never 22 years after the end of the programme compared to 40/61 (66%) of those in the control group. Effect size for wearing a seatbelt was 0.37 (P = 0.052).

D'Amico 2002

Risk Skills Training vs DARE

School‐Universal

Driving under the influence/riding with a drinking driver

No differences were observed at 6‐month follow‐up in relation to driving after drinking and riding with a drinking driver (mean values for baseline and 6‐month follow‐up: risk skills training programme group: mean 1.25 (SD 3.30) to 0.95 (SD 2.20); DARE‐A group: mean 0.75 (SD 1.42) to 0.67 (SD 1.26); control group: mean 1.58 (SD 5.32) to 1.32 (4.42).

Nirenberg 2013

ROAD

Individual‐Targeted

Speeding and distracted driving

Scores for speeding and distracted driving were lower in the control group (community service) than in the combined motivational interviewing study arms 6 months post intervention (t(607) = ‐2.32; P = 0.02) (i.e. the control group reported less of the behaviour) (Log+1 transformed mean values: control 2.49 (SD 1.57); combined MI 2.81 (SD 1.53)). No difference between groups was evident in relation to dangerous driving factor scores (t(607) = ‐0.21, P = 0.84) (Log+1 transformed means: control 1.39 (SD 1.46); combined MI 1.34 (SD 1.39)) or scores for alcohol, drugs, and driving (Log+1 mean values: control 0.58 (SD 1.14); combined MI 0.60 (SD 1.14)).

7. Sexual risk behaviour

Bonds 2010

New Beginnings

Family‐Targeted

Number of sexual partners

Significant group effect for number of sexual partners (control mean = 1.65, intervention mean = 0.68, P = 0.01, d = 0.49)

Estrada 2015

Brief Familias Unidas

Family‐Targeted

Inconsistent condom use in past 90 days

Growth curve analyses showed no significant differences in unsafe sexual intercourse, defined as inconsistent condom use, during the past 90 days between brief Familias Unidas and CPC (beta = 0 .26, P = 0 .25).

Freudenberg 2010

REAL MEN

Individual‐Targeted

Engaged in risky sexual behaviour in past 90 days

No difference was observed between the intervention arm and the control arm in relation to the proportion of participants engaging in risky sexual behaviour in the past 90 days (OR 0.856, no 95% CI given, but P > 0.05).

Griffin 2006

Life Skills Training

School‐Universal

Multiple sexual partners

21.3% of those in the intervention group had multiple sex partners at age 24 years (˜ 11 years following the end of the intervention) vs 24.5% of those in the control group.

Griffin 2006

Life Skills Training

School‐Universal

Condom use

Results showed no difference across experimental conditions in relation to condom use at age 24 years (˜ 11 years following the end of the intervention): 78.7% of the intervention group reported that they did not always use condoms vs 78.2% of controls (Chi² (1) = 0.05, P < 0.82).

McBride Murry 2014

SAAF (Stronger African American Families)

Family‐Targeted

Sexual behaviour

The effect size of the intervention on post‐test sexual behaviour was 0.01, although study authors state that detecting a substantial effect size was unlikely with a sample of < 1000 and owing to the length of time between the programme and longer‐term (65‐month) follow‐up. Using structural equation modelling, study authors also report that participation in SAAF led to protection in engagement in sexual risk behaviour through an indirect mechanism involving increased intervention‐targeted parenting practices (beta = 0.35, P < 0.01), which were associated in turn with increased youth self‐pride (beta = 0.25, P < 0.05), in turn associated with increased protective sexual norms (beta = 0.7, P < 0.01), in turn associated with reduced sexual risk behaviour (beta = ‐0.22, P < 0.01).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Lifetime incidence of sex partners in past 6 months

At 15‐year follow‐up: among those visited during pregnancy, the mean number of sex partners was 1.10 vs 1.56 (P = 0.48); and among a subgroup of women from low SES households who were unmarried, the mean number of sex partners in the past 6 months was 2.23 vs 2.48 (P = 0.73). Among those visited during pregnancy and infancy, mean incidence of sex partners was 1.16 vs 1.56 (P = 0.90); and for the subgroup of women from low SES households who were unmarried, mean was 0.92 vs 2.48 (P = 0.003).

Piper 2000

Healthy for Life

School‐Universal

Sexual intercourse in past 30 days

Students were followed up in the ninth and 10th grades, at 12 and 24 months. Students in the age‐appropriate condition reported higher rates of intercourse than those in the control group (13% vs 11%; HLM coefficient 0.4, SE 0.16, P < 0.05) at 12 months; at 24 months, prevalence was 23% vs 19% (HLM coefficient 0.32, SE 0.2, P > 0.1). The intensive condition had no effect on rates of intercourse in the 2 groups at 12 months (prevalence 15% vs 11% in the control arm, HLM coefficient 0.25, SE 0.21, P > 0.1) nor at 24 months (prevalence 21% vs 19% in the control arm; HLM coefficient ‐0.07, SE 0.15, P > 0.1).

8. Physical inactivity

Bush 1989

Know Your Body

School‐Universal

Fitness score

Study authors highlight that significant changes were observed in a favourable direction in relation to fitness. The observed difference between intervention and control group mean change after 2 years of intervention was ‐0.28 (SE 0.19); and ‐0.38 (SE 0.15) after adjustment for baseline value, age, sex, and socioeconomic status.

O’Neill 2016

Michigan Model for Health

School‐Universal

Physical activity skills

Six weeks following the intervention, results showed a significant intervention effect for physical activity skills: F[53,590.79] = 4.42, P = 0.001.

Saraf 2015

(none given)

School‐Universal

Total time spent watching TV (minutes)

Weak evidence for a reduction in time spent watching television in the intervention group: reported reduction from 70.4% (95% CI 67.0% to 73.8%) at pre‐test to 56.1% (95% CI 53.9% to 58.4%) at post‐test (P < 0.05). In comparison, a slight increase in time spent watching TV was observed in the control group: 56.4% (95% CI 53.9% to 58.9%) at pre‐test increasing to 57.9% (95% CI 55.2% to 60.8%) at post‐test; overall difference 15.8 (95% CI 15.7 to 16.9) (P < 0.01).

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Number of times participated in an outdoor activity

Overall, researchers reported a 23% reduction in the number of times participants participated in an outdoor activity. The effect was greater for males than for females (25% vs 18% reduction). Data show a greater reduction among females from a minority ethnic group (43%) than among males from an ethnic minority group (14%); and a greater reduction among white males (29%) than among males from an ethnic minority (14%). Data for white females were not available.

Walter 1989

Know Your Body

School‐Universal

Recovery index score

In Westchester, the recovery index in the intervention group changed by ‐0.7 per year vs ‐1.4 in the control group (overall difference in school means 0.7 (95% CI ‐0.1 to 1.5)). Among schools in the Bronx, the rate of change per year in the intervention group was ‐2.5 vs ‐2.5 in the control group (difference in school means 0.0, 95% CI ‐1.3 to 1.3).

9. Nutrition

O’Neill 2016

Michigan Model for Health

School‐Universal

Nutritional behaviours

Six weeks following the intervention, results show a significant effect on nutritional behaviours: F[53,213.47] = 2.32, P = 0.04.

Walter 1989

Know Your Body

School‐Universal

BMI

In Westchester, data showed no change per year among intervention schools (mean 0.0 (SD 0.1)) vs a change of 0.1 per year (SD 0.1) among control schools (difference ‐0.1, 95% CI ‐0.3 to 0.1). In the Bronx, the rate of change per year among intervention schools was 0.1 (SD 0.1) vs 0.2 (SD 0.1) among control schools (difference ‐0.1, 95% CI ‐0.3 to 0.1).

Walter 1989

Know Your Body

School‐Universal

Plasma total cholesterol (mg/dL)

In Westchester, the rate of change in total cholesterol was ‐2.1 mg/dL/y (SD 1.0) among intervention schools but ‐0.4 mg/dL/y (SD 0.7) among control schools ‐ equivalent to a net mean change in total cholesterol of ‐1.7 mg/dL/y (‐2.7 to ‐0.7 mg/dL). Among intervention schools in the Bronx, the rate of change was ‐2.6 mg/dL/y (SD 1.5) vs ‐1.6 (SD 1.8) among control schools ‐ equivalent to a difference of ‐1.0 mg/dL/y (95% CI ‐2.3 to 0.3 mg/dL).

Walter 1989

Know Your Body

School‐Universal

Total fat (% of total kcal)

In Westchester, the net mean reduction in total fat intake between intervention and control schools was ‐3.6% (95% CI ‐7.1 to ‐0.1%); in the Bronx, the net mean reduction in total fat intake was ‐1.9% (95% CI ‐7.1 to 3.3%). Data are presented from a random subsample of the total study population.

Walter 1989

Know Your Body

School‐Universal

Systolic blood pressure (mmHg)

Among schools in Westchester, systolic blood pressure changed by 0.6 mmHg (SD 0.8) vs 0.8 mmHg (SD 0.6) in the control group, for an overall difference of ‐0.2 mmHg (‐1.0 to 0.6 mmHg).

10. Mental health

Bonds 2010

New Beginnings

Family‐Targeted

Internalising disorder, externalising disorder

6‐year follow‐up: the MPCP intervention arm had significantly fewer externalising problems (‐0.11, SE 0.11) compared to the control group (0.08, SE 0.14) (P = 0.02). There was no difference between intervention and control in the number of internalising problems nor in the mental disorder symptom count (P ≥ 0.05).

15‐year follow‐up: lower proportion of intervention group participants with (1) internalising disorder diagnosed in past 9 years; intervention: 4.55% (SD 2.69), control: 16.7% (SD 3.25, OR 0.26), and (2) externalising disorder diagnosed in past 9 years; intervention: 0% (SD 0), control: 3.64% (SD 0.04).

Gonzales 2014

Bridges to High School (Bridges/ Puentes)

Family‐Targeted

Internalising symptoms

Intervention was associated with slight increase in internalising symptoms at 2 years post‐test (unstandardised regression coefficient 0.42) but a small reduction in internalising symptoms at 5 years post‐test (unstandardised regression coefficient ‐0.02).

Kellam 2008

Good Behavior Game

School‐Universal

Lifetime major depressive disorder and generalised anxiety disorder

At ˜ 12 years following intervention, when participants were aged 19 to 21 years, unadjusted rates of lifetime major depressive disorder were lower for the GBG group (10%) than for the control group (15%) (P = 0.27). The difference was slightly larger for males than for females (males: 9% for GBG, 14% for controls; females: 12% for GBG, 15% for controls). Overall rates of generalised anxiety disorder were small and did not differ by intervention condition (2% for GBG, 3% for control; P = 0.37).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Internalising problems

Results showed no difference between study arms in the mean number of internalising problems at 15‐year follow‐up: for those visited during pregnancy, mean 11.19 vs 10.58, P = 0.46; and among a subgroup of women from low SES households who were unmarried, mean 11.15 vs 10.82, P = 0.8.

For those visited through pregnancy and infancy, no difference between groups was evident: mean 11.66 vs 10.58, P = 0.19; among the subgroup of women from low SES households who were unmarried, mean 9.85 vs 10.82, P = 0.44.

Cho 2005 (Sanchez 2007, Hallfors 2006)

Reconnecting Youth

School‐Targeted

Anger

Findings regarding depression and anxiety were not reported. However, study authors report that at 6‐month follow‐up, a negative outcome was observed for those in the experimental arm compared to those in the control arm: main programme effect: F= ‐3.62, P = 0.058 (i.e. those in the intervention arm showed greater frequency of anger compared to those in the control arm).

Walker 2002

(none given)

Individual‐Universal

Mental health score

Data show no difference in change in mental health score between intervention and control participants at 3 or 12 months. However, among young people who scored 16 or more on the depression scale (indicating probable depression), there was a greater reduction in mental health score than among those in the control group (‐8.1 intervention, ‐1.4 control, 95% confidence interval (CI) for mean difference ‐0.3 to ‐13.3, P = 0.04 at 3 months; ‐1.6 intervention, 4.4 control, 95% CI ‐0.5 to ‐11.5, P = 0.03).

11. Educational attainment

Berry 2009

Coaching for Communities

Individual‐Targeted

In education/employment

At follow‐up (post intervention), 85% of those in the intervention group were in education or employment vs 59% of those in the control group (P < 0.05).

Bond 2004

Gatehouse Project

School‐Universal

Low school attachment

Two years following the intervention, the OR for low school attachment was 1.21 (95% CI 0.93 to 1.57).

Conduct Problems Prevention Research Group 2010

Fast Track

School‐Targeted

Graduated from high school or graduate equivalency diploma

At long‐term follow‐up, the adjusted OR for graduation from high school or a graduate equivalency diploma was 0.93 (95% CI 0.68 to 1.27, P = 0.654).

Freudenberg 2010

REAL MEN

Individual‐Targeted

Enrolled in educational or vocational programmes in the past year

Receipt of the intervention was associated with increased odds of being enrolled in educational/vocational programmes (OR 1.330). No 95% CI or standard error was provided.

Friedman 2002

Life Skills Training and Anti‐violence Program

Individual‐Targeted

School problems

No difference was evident between groups in relation to school problems (t = 0.91, P > 0.05).

Gonzales 2014

Bridges to HIgh School (Bridges/Puentes)

Family‐Targeted

High school dropout (no high school degree or equivalent and no attendance at high school at 12th grade assessment)

The path coefficient estimate for high school dropout at 5‐year follow‐up was not statistically significant (unstandardised regression coefficient ‐0.16), but an indirect effect of the intervention was identified through school engagement (unstandardised regression coefficient ‐0.062, 95% CI ‐0.517 to ‐0.001).

Kellam 2014

Good Behavior Game

School‐Universal

High school graduation

High school graduation rates were slightly higher for those in the GBG arm (72%) than for those in the control arm (64%), and this effect was larger for males than for females (68% vs 54%, respectively). However, these data were not adjusted for clustering.

Kitzman 2010

Nurse Family Partnership

Family‐Targeted

Academic achievement (grade point average ‐ GPA)

The GPA for grades 1 to 6 for those in the intervention group was 2.39 (0.04) vs 2.48 (0.05) for those in the control group (P = 0.19, mean difference 0.09 (‐0.04 to 0.22)). For PIAT scores (reading and maths) at 12 years, the mean difference was 1.27 (‐0.44 to 2.98) (P = 0.14). Among families of lower socioeconomic status, those in the intervention group had higher PIAT scores in reading and math at age 12 (ES 0.25, P = 0.009), higher GPAs and group‐based achievement test scores in reading and math in grades 1 through 6 (ES 0.18, P = 0.03; ES 0.22, P = 0.02, respectively), and higher GPAs in reading and math in grades 4 through 6 (ES 0.18, P = 0.047).

Li 2011

Positive Action

School‐Universal

Suspension from school

No difference between study arms was observed at follow‐up in relation to suspensions from school (IRR 0.58, 95% CI 0.15 to 2.26).

LoSciuto 1999

Woodrock Youth Development Project

School‐Universal

School attendance

Participants in the intervention group reported better average scores for self‐reported school attendance (F(1,705) = 12.18, P < 0.01, Cohen's d = 0.26). Insufficient data were available to adjust these findings for clustering of participants by classroom.

Melnyk 2003

COPE

School‐Universal

Academic competence

Academic competence was slightly higher in the intervention group than in the control group (adjusted mean 97.97, 95% CI 96.35 to 99.59; vs 95.69, 95% CI 94.21 to 97.18), respectively. F = 4.03, P = 0.05.

Morris 2003

Self‐Sufficiency Project

Family‐Targeted

Dropped out of school (aged 15 to 18)

At 36‐month follow‐up, math score at age 12 to 14 was 0.45 in the intervention group compared to 0.46 in the control group (ES ‐0.03); and average achievement was 3.43 compared to 3.54 in the control group (ES ‐0.11). Child‐reported average achievement was 3.50 in the intervention group vs 3.57 in the control group (ES ‐0.09).

Morris 2003

Self‐Sufficiency Project

Family‐Targeted

Average achievement (self‐reported)

Self‐reported average achievement was similar between intervention and control groups at 36 months of follow‐up (effect size ‐0.09).

Olds 1998

Nurse Family Partnership

Family‐Targeted

Mean long‐term school suspensions

Data show no difference between study arms in the mean number of long‐term school suspensions at 15‐year follow‐up. For those visited during pregnancy: mean 0.0 vs 0.04, P = 1.0; among a subgroup of women from low SES households who were unmarried: mean 0.01 vs 0.15, P = 0.97.

For those visited through pregnancy and infancy, no difference between groups was evident: mean 0.01 vs 0.04, P = 1.0; among the subgroup of women from low SES households who were unmarried, mean 0.04 vs 0.15, P = 0.25.

Schweinhart 1993

High/Scope Perry Preschool Program

School‐Targeted

Total school achievement

Those in the intervention group had higher total achievement (mean 122.2, SD 41.6) than those in the control group (mean 94.5, SD 35.5) at 9‐year follow‐up.

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Grade point average

Educational impacts of the intervention were more pronounced among girls than among boys. Overall, those receiving the intervention overall had a higher grade point average (GPA) than those given control (average 2.71 vs 2.63). The difference was particularly marked among girls from an ethnic minority (average GPA 2.83 vs 2.62 for those in the control group).

Tierney 1995

Big Brothers Big Sisters

Individual‐Targeted

Truancy/skipping school

Participants of Big Brothers Big Sisters showed a 52% reduction in the number of times they skipped a day of school and a 37% reduction in the number of times they skipped class. The impact was greater among girls than boys, for instance those in the intervention group showed 84% reduction compared to 4% reduction for skipping a day of school. The reduction was greater for white females than for females from an ethnic minority (92% reduction vs 78% reduction, respectively, for skipping a day of school; 72% vs 46% reduction for skipping class). Among white males compared to males from an ethnic minority, the reduction was similar for skipping a day of school, but a greater impact was evident among ethnic minority males than among white males for skipping class (22% vs 12% reduction).

12. Teenage pregnancy

Olds 1998

Nurse Family Partnership

Family‐Targeted

Ever pregnant or made someone pregnant in the previous 6 months

At age 15 (˜ 13 years following the intervention), 5 of 176 (2.8%) in the intervention group had ever been pregnant or made someone pregnant compared to 4 of 148 (2.7%) in the control group (OR 1.04, SE 0.65).

Conduct Problems Prevention Research Group 2010

Fast Track

School‐Targeted

Pregnancy by age 18

At age 18, the proportion of participants reporting pregnancy were as follows: girls: urban African American 40%, urban European American 21%, rural European American 17%; boys: urban African American 27%, urban European American 12%, rural European American 11%.

Schweinhart 1993

High/Scope Perry Preschool Program

School‐Targeted

At least 1 pregnancy by age 19 (females only)

At 14‐year follow‐up, 12 of 25 (48%) in the intervention group had had at least 1 pregnancy by age 19 compared to 16 of 24 (67%) in the control group (effect size 0.5).

13. Health problems

Schweinhart 1993

High/Scope Perry Preschool Study

School‐Targeted

Long‐term health problems

22 years following completion of the intervention, 36% of those in the intervention group had had health problems treated in the previous 5 years compared to 38% in the control group (effect size 0.04, P = 0.823). 30% of those in the intervention group had been hospitalised in the previous 12 months compared to 15% of those in the control group (effect size 0.38; P = 0.043).

ASPD: antisocial personality disorder.

CI: confidence interval.

COPE: Creating Opportunities for Personal Empowerment.

CPC: Community Practice Condition.

DARE: Drug Abuse and Resistance Education.

DSM: Diagnostic and Statistical Manual of Mental Disorders.

ES: Effect size.

GBG: Good Behaviour Game.

GPA: grade point average.

HLM: hierarchical linear modelling.

LIFT: Linking the Interests of Families and Teachers.

MPCP: Mother program plus child program.

OR: odds ratio.

PATHS: Promoting Alternative Thinking Strategies.

ProSAAF: Protecting Stronger African American Families program.

ROAD: Reducing Offenses of Adolescent Drivers.

SafERteen: brief intervention aimed at reducing and preventing violence and alcohol use.

SD: standard deviation.

SE: standard error.

SES: socioeconomic status.

SMD: standardised mean difference.

Figuras y tablas -
Table 2. Outcomes not included in meta‐analysis
Table 3. Summary of studies by intervention type

Primary authors

Trial

Study type

Country

Duration

Theory

Follow‐up (post intervention)

Components

Age targeted

Behaviour targeted

N behaviours targeted

Process evaluation

Targeted individual‐level interventions

Bernstein 2010

Reaching Adolescents for Prevention

IT

USA

< 3 months (motivational intervention, referral to services, telephone conversation)

N/S

12 months

Motivational interview, referral to community resources and drug treatment services

14‐21

Alcohol use, vehicle‐related risk, antisocial behaviour, sexual risk behaviour

4

Assessed adherence to intervention

Berry 2009

Coaching for Communities

IT

UK

10 months

Distinction‐based learning

Post intervention

1‐week residential programme and 9 months of mentoring

15‐18

Alcohol use, drug use, antisocial behaviour, educational attainment

4

Assessed impact of ‘quality’ of the mentor and examined impact of dose of the intervention

Bodin 2011

A Mentoring Intervention

IT

Sweden

12 months

Rhodes model regarding role modelling on healthy relationships with adults

12 months

Mentoring

13‐18

Alcohol use, drug use, academic development, mental health

5

Yes (partial) ‐ assessed dropout, adherence, meetings, positive views of programme, intensity

Cunningham 2012

SafERteens

IT

USA

< 1 month (MI)

Traditional motivational interviewing model

12 months

Traditional MI using computer or therapist

14‐18

Antisocial behaviour, alcohol use

2

No

Dolan 2010

BBBS Ireland

IT

Ireland

12 months

Rhodes model of youth mentoring

2 years

Individual (mentoring)

10‐16 (93% < 14 years)

Alcohol use, tobacco use, cannabis use, antisocial behaviour, educational attainment, mental health

6

Yes ‐ full

Freudenberg 2010

REAL MEN

IT

USA

1‐2 months

‐‐‐

12 months

Individual components (jail‐based intervention and intervention within community setting)

17‐18

Drug use, sexual risk behaviour, antisocial behaviour, education, and employment

4

No

Friedman 2002

Botvin LST and Anti‐violence

IT

USA

6‐12 months (55 sessions for classroom programme, 20 sessions for violence programme)

Life Skills Training model; social cognitive procedures

6 months

Individual (triple‐modality classroom programme)

13‐18

Alcohol use, drug use, antisocial behaviour, educational attainment

4

Yes ‐ partial (feedback from participants, adherence, intensity)

Monti 1999

Alcohol Screening and brief intervention

IT

USA

1 day

MI

6 months

1 individual component: brief motivational interviewing

18

Alcohol use, vehicle risk behaviour

2

Yes (partial) ‐ rating of adherence, videotaping of interventionist to rate fidelity

Nirenberg 2013

ROAD

IT

USA

1 day (BI)

No

6 months

2 individual components: motivational interviewing and placement in hospital emergency department or in the community

18

Alcohol use, vehicle risk behaviour

2

No

Redding 2015

Step‐by‐Step

IT

USA

9 months

Transtheoretical model (TTM)

18 months

2 individual components: TTM‐tailored feedback via a computer‐based system and personalised stage‐targeted counselling

16

Tobacco use, sexual risk behaviour

2

Yes (full) ‐ counsellors reported on what was covered in sessions and activities they used. Teen feedback was obtained.

Tierney 1995

Big Brothers Big Sisters

IT

USA

12 months

N/S

6 months

Individual component: mentoring (3 meetings per month over a period of nearly 12 months on average)

10‐16

Alcohol use, drug use, antisocial behaviour, educational attainment

5

Yes (partial). Participants’ feedback regarding volunteer impact. Adherence.

Wagner 2014

Guided Self‐Change

IT

USA

2 months

Guided Self‐Change

6 months post intervention

Individual component: brief motivational interview via cognitive‐behavioural approach

14‐18

(mean 16.2, SD 1.2)

Alcohol use, drug use, antisocial behaviour

3

Yes (partial) ‐ recorded sessions and reviewed to assess adherence

Universal individual‐level interventions

Johnson 2015

Healthy Futures

IU

USA

3 months

Social learning theory

6 months

Individual component: 3 sessions of motivational interviewing (once per month) with follow‐up in‐between

14‐21

(mean 16)

Alcohol use, illicit drug use, antisocial behaviour

3

No

Lana 2014

Prevencanadol

IU

Spain, Mexico

9 months (1 school year)

Transtheoretical model of behaviour change; ASE model

Post intervention

Individual component – website regarding prevention and treatment of cancer

12‐15

Alcohol use, tobacco use, diet, physical activity

4

No

Minnis 2014

Yo Puedo

IU

USA

6 months

Social learning theory; behavioural economics

Post intervention

Individual components: cash payments and life skills sessions

16

Alcohol use, drug use, antisocial behaviour, sexual risk behaviour

5

Yes (partial) ‐ adherence, attendance, cash earned, meeting goals

Walker 2002

[No study name]

IU

UK

1 day (BI)

Self‐efficacy theory

12 months post intervention

Individual component: 20‐minute consultation with practice nurse to discuss health and health‐related behaviour

14‐15

Alcohol use, tobacco use, physical activity, nutrition, mental health

5

Yes (full) ‐ attendees: acceptability; observation of nurses to assess adherence

Targeted family‐level interventions

Beach 2016

ProSAAF

FT

USA

2‐3 months

Based on previous interventions and stress‐spillover theory

˜ 9 months

Family component: six 2‐hour home‐delivered sessions that focused on couple‐based issues and development of protective couple and parenting processes

10‐13

Tobacco use, alcohol use, illicit drug use, antisocial behaviour

4

No

Bonds 2010

New Beginnings

FT

USA

3 months

Cascading pathway model

3 months, 6 months, 6 years, 15 years

Family components only (mother‐only, mother‐plus‐child, and child‐only programmes)

9‐12 (average 10.4, SD 1.1)

Substance use (marijuana and alcohol), antisocial behaviour, sexual risk behaviour, mental health

5

Yes: adherence, feedback, manuals, training, supervision, scores, rating

Brody 2012

SAAF‐T

FT

USA

2‐3 months

N/S

22 months

Family components only (5 meetings for caregivers and adolescents separately, followed by a 1‐hour session for families together)

16

Drug use, antisocial behaviour, mental health

3

Adherence to curriculum, video of sessions and feedback

Catalano 1999

Focus on Families

FT

USA

1 year (approximately)

Social development model

8 months

Family components: parent skills training sessions and case management

3‐14

Alcohol use, tobacco use, cannabis use, antisocial behaviour

4

No ‐ adherence only

Estrada 2016

Brief Familias Unidas

FT

USA

2‐3 months

NS

24 months

5 weekly 2‐hour parent group sessions, 3 homework assignments for parents, and one 1‐hour family visit

15

(mean 15.3 years, SD 0.89)

Tobacco use, alcohol use, illicit drug use, sexual risk behaviour

4

Yes (partial) ‐ attendance data only

Gonzales

Bridges to High School

FT

USA

3 months

Social development model

5 years

Family components (weekly group sessions with separate and joint sessions and home visits)

12‐13

Alcohol use, tobacco use, drug use, antisocial behaviour, educational attainment, mental health

6

Yes (full) ‐ post‐test interview ‐ rating by parents; attendance; video recording of intervention sessions ‐ adherence

Jalling 2016

Comet 12‐18

FT

Sweden

3‐4 months

Operant learning and social learning principles

6 months

Family component: 9 weekly group sessions of 2 to 2.5 hours and 1 optional booster session. Sessions involved role‐play, home assignments, and use of video as a basis for discussion.

12‐18

(mean 14)

Alcohol use, illicit drug use, antisocial behaviour

3

Yes ‐ attendance and self‐assessment by group leaders of extent to which programme manual was fulfilled in sessions

Jalling 2016b

ParentSteps

FT

Sweden

3‐4 months

Resilience model

6 months

Family component: 6 weekly parent sessions of 1.5 to 2 hours

12‐18

(mean 14)

Alcohol use, illicit drug use, antisocial behaviour

3

Yes (partial) ‐ attendance and leader self‐assessment of extent to which programme manual was fulfilled in sessions

Kim 2011

Middle School Success

FT

USA

3 months with ongoing support for 1 school year

‐‐‐

2 years

Family components: curriculum to parents through 6 group sessions for parents plus follow‐up sessions; 6 skills‐based sessions for girls; ongoing training and support for parents and adolescents

11‐12

Antisocial behaviour, tobacco use, alcohol use, marijuana use, mental health

5

No

Kitzman 2010

Nurse Family Partnership 2

FT

USA

2.5 years

Theories of child development, behaviour change, human ecology, self‐efficacy, and attachment

12 years

Family components: free transportation to prenatal visits, screening, referral; prenatal, infant, and child home visitation

0 ‐2

Alcohol use, tobacco use, drug use, antisocial behaviour, mental health, educational attainment

6a

No

Li 2002

imPACT

FT

USA

90 minutes (1 day)

None

12 months

1× individual (video)

14

Alcohol use, tobacco use, drug (cannabis) use, antisocial behaviour, sexual risk behaviour

5

No

Milburn 2012

STRIVE

FT

USA

2 months

Cognitive‐behavioural theories

12 months post intervention

Family component: 5 sessions delivered to young person and parent

14

Drug use, antisocial behaviour, sexual risk behaviour

3

Yes (partial) ‐ attendance, satisfaction of parents/ adolescents; manual ‐ assessed fidelity of session delivery

Morris 2003

Self‐Sufficiency Project

FT

Canada

Up to 3 years

Economics and psychology theories

Post intervention

Individual component: earnings supplement to single parents who left welfare for full‐time employment

0‐2, 3‐8, or 9‐15

Alcohol use, tobacco use, drug use, antisocial behaviour, mental health educational attainment

6

No

Murry 2014

SAAF (Stronger African American Families)

FT

USA

7 weeks

Social learning theory, problem behaviour theory, Gibbons and Gerrard’s cognitive model of adolescent behaviour

65 months

1 family component: separate 1‐hour caregiver and adolescent session followed by joint session to practice skills

Mothers and children aged 11 years

Alcohol use, sexual risk behaviour

2

Yes ‐ attendance measured, fidelity assessed using video

Olds 1998

Nurse Family Partnership

FT

USA

2 years

Human ecology, self‐efficacy, human attachment

15 years

1 family component

Mothers aged < 19, children aged 0‐2

Drug use, antisocial behaviour, sexual risk behaviour, educational attainment

6a

No

Pantin 2009

Familias Unidas

FT

USA

N/S

N/S

2.5 years post intervention

2 family components: nine 2‐hour group sessions, ten 1‐hour family visits, four 1‐hour booster sessions.

13‐14

Alcohol use, tobacco use, drug use, antisocial behaviour, sexual risk behaviour

5

Yes (partial) ‐ sessions video recorded and rated on adherence and quality

Schwinn 2014

[No study name]

FT

USA

1 month

N/S

5 months

1 family component: 3‐session online health promotion programme

11‐12

Drug use, physical activity, nutrition

3

Minimal adherence data only

Universal family‐level interventions

Averdijk 2016

Triple P

FU

Switzerland

3 months

N/S

9 years

1 family component: a group‐based course with 4 units of 2 to 2.5 hours and 4 follow‐up telephone contacts with each participant

7.5 years

Tobacco use, alcohol use, illicit drug use, antisocial behaviour

4

Yes ‐ attendance at sessions, satisfaction with programme, provider competency, and delivery of course material

Connell 2007

Family Check‐Up

FU

USA

2.5 years

Life skills training programme

3.5 years (6 years including intervention)

School programme including a universal classroom intervention; parenting practice component with assessment and feedback, family management treatment

11‐12

Alcohol use, tobacco use, drug use, anti‐social behaviour

4

Assessed adherence only

Haggerty 2007

Parents Who Care

FT

USA

2‐3 months

Social development model

2 years post intervention

7‐session group intervention for parent and adolescent or a 7‐session self‐administered intervention for adolescent and parent with telephone support

13‐14

Alcohol use, tobacco use, drug use, antisocial behaviour, sexual risk behaviour

5

Yes (full) ‐ parents; satisfaction; student satisfaction; adherence, quality

Targeted school‐level interventions

Conduct Problems Prevention Research Group 2014

Fast Track

ST

USA

10 years

Developmental model of conduct disorders

19 years

School and family components (family group programme, parent training groups, parent‐child interaction groups, tutoring; school curriculum; youth groups, youth forums)

Kindergarten to 12th grade

Antisocial behaviour, educational attainment, mental health

3

Yes ‐ training, supervision, fidelity ‐ rating of quality of implementation, observation, quality of teaching, quality of classroom management

Lochman 2003

Coping Power 1

ST

USA

12 months

Social learning theory

12 months post intervention

Parent and child components: parent group meetings; teacher meetings; group school‐based sessions for children

10‐11

Alcohol use, tobacco use, drug use, antisocial behaviour

4

Yes (partial) ‐ adherence to intervention (manuals, guidelines, training); attendance

Lochman 2004

Coping Power 2

ST

USA

15 months

Social learning theory

12 months post intervention

Parent and child components: parent group sessions and child school‐based group sessions

9‐10

Alcohol use, drug use, antisocial behaviour

3

Yes (full) ‐ meetings with target children; staff rated accomplishment of objectives, audio/video taping; observation

Sanchez 2007

Reconnecting Youth

ST

USA

2 years

Theoretical framework based on strain, social learning and control theories

1 year post intervention

1 school component: 55 school lessons and 24 booster lessons

15

Drug use, antisocial behaviour, mental health, educational attainment

4

Yes (full) ‐ teacher logs, attendance records, observations in classroom, student questionnaires, implementation

Schweinhart 1993

High/Scope Perry Preschool Project

ST

USA

2 school years

Piaget's constructivist theory of child development

36 years

Preschool and parent components: preschool for 2.5 hours each weekday morning, home visits by teachers for 1.5 hours per week, parent group meetings

3‐4

Antisocial behaviour, educational attainment

2

No

Shetgiri 2011

[No study name]

ST

USA

9 months (1 school year)

N/S

Post intervention

3 school components; 28 weekly peer groups facilitated by a school clinical social worker, field trips, community service activities

13‐15

Alcohol use, tobacco use, drug use, antisocial behaviour

4

No

Universal school‐level interventions

Averdijk 2016b

PATHS

SU

Switzerland

1 year

Not clear

8 years

School curriculum: 46 lessons addressing problem‐solving, social relationships, self‐regulation, emotional understanding, rules, and positive self‐esteem

8‐9

(year 2 primary school)

Tobacco use, alcohol use, illicit drug use, antisocial behaviour

4

Yes ‐ classroom observations, rating of lesson quality

Beets 2009

Positive Action (Hawaii)

SU

USA

4‐5 years

Theory of self‐concept, consistent with theories of triadic influence

Post intervention

School curriculum with school‐wide climate programme, family, community components

K‐12

Alcohol use, tobacco use, drug use, violent behaviours, sexual risk behaviour

5

No

Bond 2004

Gatehouse Project

SU

Australia

24 months

Health‐promoting schools framework, ecological approach

4 years

Whole‐school approach involving a curriculum, institutional and individual‐focused components

13‐14

Alcohol use, tobacco use, cannabis use, mental health (emotional well‐being)

6

Yes ‐ full

Bush 1989

Know Your Body 2

SU

USA

5 years

Social learning theory

Following 2 years of intervention

School curriculum, screening and feedback, parental involvement, and newsletter to families

10‐13

Tobacco use, physical activity, cardiovascular risk factors (nutrition)

3

No ‐ adherence only

D'Amico 2002

Risk Skills Training Programme vs DARE

SU

USA

< 1 month (1 hour)

Based on models ‐ not theory (DARE, alcohol skills training programme, BASICS)

4 months

School components only: school‐based group session with curriculum

14‐19

Alcohol use, drug use, vehicle‐related risk behaviour

3

Yes ‐ rating of audiotapes, adherence to protocol, student rating of acceptability and feedback

DeGarmo 2009

LIFT

SU

USA

3 months

Developmental model centred on moment‐to‐moment social interaction processes

7 years

School and family components (parent management training, child skills training, school recess component)

10‐11

Alcohol use, tobacco use, drug use, antisocial behaviour

4

Yes ‐ acceptability of intervention to parents and teachers, adherence using checklists, family participation recorded, completion of critical components

Fearnow‐Kenney 2003

All Stars Sr

SU

USA

9 months (1 school year)

No

Post intervention

School components only (All Stars activities implemented by teachers)

13‐19

Alcohol use, cannabis use, tobacco use, nutrition

4

Yes ‐ full (teacher and student focus groups re perceptions of programme, % implemented)

Flay 2004

Aban Aya

SU

USA

4 years

Theory of triadic influence, incorporation of Nguzo Saba principles to promote African American cultural values

Post intervention

School curriculum from grade 5 to 8 (SDC); or curriculum plus parental support, school climate and community components

10‐14

Antisocial behaviour, sexual risk behaviour, substance use

3

Yes ‐ full

Gottfredson 2010

All Stars 2

SU

USA

32 weeks (1 school year)

Social learning theory

Post intervention

School components only: classroom interactive sessions, homework assignments, parental attendance at graduation ceremonies

11‐14

Alcohol use, tobacco use, drug use, antisocial behaviour

4

Yes (full) ‐ adherence ‐ observed implementation with site visits, fidelity checklists; quality rating and how sessions met objectives; adherence and delivery

Griffin 2006

Life Skills Training

SU

USA

3 years

Life skills training

10 years

School components: skills‐based curriculum with interactive teaching methods

12‐13

Alcohol use, drug use, sexual risk behaviour

3

Yes (partial) ‐ monitoring of classes ‐ completeness with respect to the % of curriculum covered

Griffin 2009

BRAVE (Building Resiliency and Vocational Excellence)

SU

USA

3 school years

Social learning theory

1 year

School component and individual component: health education classroom sessions and training sessions on life skills, manhood development, or violence prevention; development of career plans, buddy system; plus mentoring

13‐14

Alcohol use, tobacco use, drug use, antisocial behaviour

4

No ‐ just reviewed and practised using material with trainers; trainers were required to have schedule for delivery of lesson

Ialongo 1999

Classroom‐Centred (CC) programme including Good Behavior Game (GBG)

SU

USA

1 school year (9 months)

Life course/social field theory

5 years

Curriculum, behaviour management using the GBG, and strategies for children who failed to respond to intervention

6‐7

Antisocial behaviour, mental health, academic achievement (antecedents of substance use)

6a

Yes (full) ‐ checked classroom setup, observed classroom sessions, visit records. Classroom record reviews completed by students were reviewed.

Ialongo 1999

FSP

SU

USA

1 school year (9 months)

Life course/social field theory

5 years

Training for teachers and staff, home‐school learning activities, 9 workshops for parents

6‐7

Antisocial behaviour, mental health, academic achievement (antecedents of substance use)

6a

Yes (full) ‐ documentation of contact with parents. Parents reported on implementation and usefulness. Recorded observations of workshops

Kellam

Good Behaviour Game (GBG)

SU

USA

2 school years

Life course/social field theory

Up to 12 years

School component: behaviour management

6‐8

Antisocial behaviour, educational achievement (antecedents of substance use)

5a

No

Lewis 2013

Positive Action (Chicago)

SU

USA

6 school years

Self‐esteem enhancement theory, social ecological theory

Post intervention

School components: classroom curriculum; school‐wide climate development; teacher, family, counsellor, and community training

8‐13 (grades 3‐8)

Alcohol use, tobacco use, drug use, antisocial behaviour

6

Yes (partial) ‐ adherence to programme; workshops for teachers; unit implementation report at the end of each unit

LoSciuto 1999

Woodrock Youth Development Project

SU

USA

2 school years

None

Post test

3 components: education through seminars, psychosocial support (mentoring, tutoring, extracurricular activities), and family and community supports (family involvement, counselling, and outreach)

6‐14

(mean 10, SD 1.7)

Alcohol use, tobacco use, drug use, antisocial behaviour, educational attainment

5

No

Mathews 2016

PREPARE

SU

South Africa

12 months

Social cognition models including the reasoned action framework and the I‐Change theoretical model

12 months

4 school components: 21 interactive and skills‐based sessions of 1‐1.5 hours once per week; a school health service delivered by a nurse from a local public clinic; safety training to school personnel and parent representatives; and a school safety programme delivered to a randomly selected group of 20 volunteers

Grade 8 (mean 13 years)

Antisocial behaviour, sexual risk behaviour

2

Yes ‐ rating by participants of the quality of sessions, attendance at lessons, visits to the school nurse, and attendance at the safety programme. Facilitator performance scores

McNeal 2004

All Stars 1

SU

USA

9 months (1 school year)

Social learning theory

12 months post intervention

School‐based components: curriculum including classroom, group, and 1‐to‐2 sessions; homework to increase parental interaction/involvement

11‐13

Alcohol use, tobacco use, drug use, antisocial behaviour, sexual risk behaviour

5

Yes ‐ rating of sessions, rating of objectives achieved

Melnyk 2013

COPE

SU

USA

4 months

Cognitive‐behavioural theory

6 months

School and parent components: educational and cognitive‐behavioural skills‐building programme, including physical activity in each of the 15 sessions, homework, and a parent newsletter

14‐16

Alcohol use, drug use, physical activity, nutrition, mental health, educational attainment

6

Yes (partial) ‐ participants completed questionnaires, parents provided feedback. Fidelity of teachers measured

Nader 1999

CATCH 3

SU

USA

3 years

Social cognitive theory, social learning theory, organisational change theory

3 years post intervention

School and family components: classroom curriculum, teacher training, consultation to teachers, physical activity sessions, school policy, home‐based tobacco use prevention programme, family activities to promote physical activity

8‐11

Tobacco use, physical activity, nutrition

3

Yes ‐ full

O'Neill 2011

Michigan Model for Health

SU

USA

2 years

Health Belief Model, social learning theory

2 months

School‐based curriculum and skills‐based learning (24 lessons in grade 4; 28 lessons in grade 5)

9‐11

Alcohol use, tobacco use, antisocial behaviour, physical activity, nutrition

5

Yes (partial) ‐ adherence to instructor, fidelity to protocol, teacher survey regarding lessons delivered, implementation fidelity assessment

Perry 2003

DARE vs DARE+

SU

USA

18 months

N/S

Post intervention

DARE: 1 school component involving 10‐session curriculum delivered by police officers. DARE‐Plus: 3 school components including 4‐session classroom‐based, peer‐led, parental involvement programme, extracurricular activities, and neighbourhood action teams organised by community leaders

12‐13

Alcohol use, tobacco use, drug use, antisocial behaviour

4

No

Piper 2000

Healthy for Life

SU

USA

Intensive: 3 months; age appropriate: 3 years

Social influences model

24 months post intervention

2 school components: 54‐lesson curriculum, use of peer leaders.

1 family component: parent orientation session, home mailings, homework involving parent/adult interviews.

1 community component: community organiser, sponsorship of health event, policy work

11

Alcohol use, tobacco use, drug (cannabis) use, sexual risk behaviour, nutrition

5

Yes (full) ‐ teacher logs re sessions; observation; interviews with staff, teachers, students; feedback surveys. Context, implementation

Saraf 2015

[No study name]

SU

India

9 months

N/S

Post intervention

2 school components: a school health committee, classroom activities, school policies.

2 community component: community outreach

12‐13

Tobacco use, physical activity, nutrition

3

No – number of schools that adopted policies is stated

Shek 2011

PATHS

SU

Hong Kong

20 hours for each of the 3 school years

N/S

Post intervention

1 school component: school curriculum

12‐14

Alcohol use, tobacco use, drug use, antisocial behaviour, sexual risk behaviour

5

No

Simons‐Morton 2005

Going Places

SU

USA

3 school years

Social development and social cognitive theory

3 months post intervention

2 school components: social skills curriculum (18 sessions in 6th grade, 12 in 7th grade, 8 in 8th grade) and school environment enhancement.

Parent component: parent education via video, booklet, periodic newsletters, and involvement in homework

11‐14

Alcohol use, tobacco use, antisocial behaviour

3

Yes (full) ‐ adherence, teacher rating of students’ participation, student feedback regarding attendance, helpfulness, knowledge. Parent interviews.

Walter 1989

Know Your Body 1

SU

USA

5‐6 years

PRECEDE health education planning model (which incorporates elements of social learning theory and the Health Belief Model)

Post intervention

School component: curriculum for 2 hours per week for each school year (grades 4‐9).

Parent component: self‐assessment, newsletters, participation in activities, seminars.

8‐15

(mean 9 years at baseline)

Tobacco use, physical activity, nutrition

3

Yes (partial) ‐ adherence to protocol, visits to classrooms

Wolfe 2009

Fourth R‐Skills for Youth Relationships

SU

Canada

8 months

N/S

2 years

School component: 21‐lesson curriculum.

14‐15

Alcohol use, tobacco use, drug use, antisocial behaviour, sexual risk behaviour

5

Yes (partial) ‐ teacher checklists re completion

ASE: Attitude ‐ social influence ‐ self‐efficacy model.

BASICS: Brief Alcohol Screening and Intervention for College Students.

BBBS: Big Brothers Big Sisters.

BI: Brief intervention.

BRAVE: Building Resiliency And Vocational Excellence.

CATCH: Coordinated approach to child health.

CC: classroom‐centred.

COPE: Creating Opportunities for Personal Empowerment.

DARE: Drug Abuse and Resistance Education.

FSP: Family Schools Partnership.

FT: family‐targeted.

FU: family‐universal.

GBG: Good Behaviour Game.

imPACT: Informed Parents and Children Together.

IT: individual‐targeted.

IU: individual‐universal.

LIFT: Linking the Interests of Families and Teachers.

LST: Life Skills Training.

MI: motivational intervention.

N/S: not stated.

PATHS: Promoting Alternative Thinking Strategies.

ProSAAF: Protecting Stronger African American Families.

ROAD: Reducing Offenses by Adolescent Drivers.

SAAF: Stronger African American Families.

SD: standard deviation.

SDC: Social Development Curriculum.

ST: school‐targeted.

STRIVE: Support to Reunite, Involve and Value Each Other.

SU: school‐universal.

TTM: transtheoretical model.

Figuras y tablas -
Table 3. Summary of studies by intervention type
Table 4. Number of studies targeting different behaviours by study type

N studies

N outcomes targeted (average)

Primary behaviours

Secondary behaviours

Tobacco

Alcohol

Drugs

ASB

Self‐harm

Gambling

Vehicle risk

Sexual risk

Physical activity

Nutrition

Education and attainment

Mental illness

Individual Targeted

12

4

4

10

7

9

3

3

5

2

Individual Universal

3

4

2

4

2

2

1

2

2

1

1

Family Targeted

17

4

8

13

16

14

7

1

1

4

7

Family Universal

3

4

3

3

3

3

1

School Targeted

6

3

2

3

4

6

3

2

School Universal

28

4

23

22

21

19

1

9

6

8

6

5

TOTAL

70

42

55

53

53

0

0

4

21

9

11

19

17

Figuras y tablas -
Table 4. Number of studies targeting different behaviours by study type
Table 5. Behaviours targeted by interventions by study type

Author, date

Study type

Tobacco use

Alcohol use

Illicit drug use

Antisocial behaviour

Vehicle risk

Sexual risk behaviour

Physical inactivity

Nutrition

Mental illness

Education & attainment

Bernstein 2010

IT

X

X

X

X

Berry 2009

IT

X

X

X

X

Bodin 2011

IT

X

X

X

X

X

Cunningham 2012

IT

X

X

Dolan 2010

IT

X

X

X

X

X

X

Freudenberg 2010

IT

X

X

X

X

Friedman 2002

IT

X

X

X

X

Monti 1999

IT

X

X

Nirenberg 2013

IT

X

X

Redding 2015

IT

X

X

Tierney 1995

IT

X

X

X

X

X

Wagner 2014

IT

X

X

X

Johnson 2015

IU

X

X

X

Lana 2014

IU

X

X

X

X

Minnis 2014

IU

X

X

X

X

X

Walker 2002

IU

X

X

X

X

X

Beach 2016

FT

X

X

X

X

Bonds 2010

FT

X

X

X

X

X

Brody 2012

FT

X

X

X

Catalano 1999

FT

X

X

X

X

Estrada 2016

FT

X

X

X

X

Gonzales

FT

X

X

X

X

X

X

Jalling 2016

FT

X

X

X

Jalling 2016b

FT

X

X

X

Kim 2011

FT

X

X

X

X

X

Kitzman 2010

FT

Indirect

X

X

X

X

X

Li 2002

FT

X

X

X

X

X

Milburn 2012

FT

X

X

X

Morris 2003

FT

X

X

X

X

X

X

Murry 2014

FT

X

X

Olds 1998

FT

Indirect

X

X

X

X

X

Pantin 2009

FT

X

X

X

X

X

Schwinn 2014

FT

X

X

X

Averdijk 2016

FU

X

X

X

X

Connell 2007

FU

X

X

X

X

Haggerty 2007

FU

X

X

X

X

X

CPRG 2014

ST

Indirect

Indirect

Indirect

X

Indirect

X

X

Lochman 2003

ST

X

X

X

X

Lochman 2004

ST

X

X

X

Sanchez 2007

ST

X

X

X

X

Schweinhart 1993

ST

Indirect

Indirect

Indirect

X

X

Shetgiri 2011

ST

X

X

X

X

Beets 2009

SU

X

X

X

X

X

Bond 2004

SU

X

X

X

X

X

X

Bush 1989

SU

X

X

X

D'Amico 2002

SU

X

X

X

DeGarmo 2009

SU

X

X

X

X

Fearnow‐Kenney 2003

SU

X

X

X

X

Flay 2004

SU

X

X

X

Gottfredson 2010

SU

X

X

X

X

Griffin 2006

SU

X

X

X

Griffin 2009

SU

X

X

X

X

Ialongo 1999a

SU

X

X

X

X

X

X

Ialongo 1999b

SU

X

X

X

X

X

X

Kellam

SU

X

X

X

X

Indirect

X

Lewis 2013

SU

X

X

X

X

X

X

LoSciuto 1999

SU

X

X

X

X

X

Matthews 2016

SU

X

X

McNeal 2004

SU

X

X

X

X

X

Melnyk 2013

SU

X

X

X

X

X

X

Nader 1999

SU

X

X

X

O'Neill 2011

SU

X

X

X

X

X

Perry 2003

SU

X

X

X

X

Piper 2000

SU

X

X

X

X

X

Saraf 2015

SU

X

X

X

Shek 2011

SU

X

X

X

X

X

Simons‐Morton 2005

SU

X

X

X

Walter 1989

SU

X

X

X

Wolfe 2009

SU

X

X

X

X

X

TOTAL

42

55

53

53

4

21

9

11

17

19

FT: family‐targeted.

FU: family‐universal.

IT: individual‐targeted.

IU: individual‐universal.

ST: school‐targeted.

SU: school‐universal.

Figuras y tablas -
Table 5. Behaviours targeted by interventions by study type
Table 6. Sensitivity analysis

Outcome

Classification

Subgroup

N studies

N intervention

N control

Estimate, 95% CI

I²

Intracluster correlation coefficient (ICC): use of lowest ICC (studies with short‐term follow‐up only)

Tobacco use

Individual Targeted

Highest ICC

2

280

241

1.28, 0.75 to 2.19

0

Lowest ICC

2

280

241

1.28, 0.75 to 2.19

0

Individual Universal

Highest ICC

2

925

624

1.03, 0.32 to 3.27

38

Lowest ICC

2

925

624

1.03, 0.32 to 3.27

38

Family Targeted

Highest ICC

2

160

153

0.78, 0.40 to 1.53

0

Lowest ICC

2

160

153

0.78, 0.40 to 1.53

0

School Universal

Highest ICC

9

8365

6989

0.77, 0.60 to 0.97

57

Lowest ICC

9

8365

6989

0.76, 0.59 to 0.97

65

Alcohol use

Individual Targeted

Highest ICC

4

1204

840

1.02, 0.79 to 1.30

48

Lowest ICC

4

1204

840

1.02, 0.79 to 1.30

48

Individual Universal

Highest ICC

4

1105

806

0.80, 0.58 to 1.11

0

Lowest ICC

4

1105

806

0.80, 0.58 to 1.11

0

Family Targeted

Highest ICC

3

212

205

0.83, 0.47 to 1.46

29

Lowest ICC

3

212

205

0.83, 0.47 to 1.46

29

School Targeted

Highest ICC

1

615

603

1.03, 0.56 to 1.91

‐‐

Lowest ICC

1

615

603

1.03, 0.56 to 1.91

‐‐

School Universal

Highest ICC

8

4382

4369

0.72, 0.56 to 0.92

58

Lowest ICC

8

4382

4369

0.71, 0.55 to 0.91

60

Binge drinking

Individual Targeted

Highest ICC

3

130

120

0.97, 0.68 to 1.37

0

Lowest ICC

3

130

120

0.97, 0.68 to 1.37

0

School Universal

Highest ICC

5

2825

2669

0.66, 0.41 to 1.06

43

Lowest ICC

5

2825

2669

0.66, 0.45 to 0.99

49

Cannabis use

Individual Targeted

Highest ICC

2

67

59

1.10, 0.71 to 1.97

0

Lowest ICC

1

67

59

1.10, 0.71 to 1.97

0

Individual Universal

Highest ICC

2

180

182

0.69, 0.46 to 1.04

0

Lowest ICC

2

79

83

0.69, 0.46 to 1.04

0

Family Targeted

Highest ICC

3

192

188

1.02, 0.52 to 2.02

43

Lowest ICC

3

192

188

1.02, 0.52 to 2.02

43

School Universal

Highest ICC

5

1924

2216

0.79, 0.62 to 1.01

0

Lowest ICC

5

1924

2216

0.77, 0.61 to 0.97

0

Illicit drug use

Individual Targeted

Highest ICC

3

342

296

0.94, 0.71 to 1.25

0

Lowest ICC

3

342

296

0.94, 0.71 to 1.25

0

Family Targeted

Highest ICC

1

33

36

0.74, 0.42 to 1.31

‐‐

Lowest ICC

1

33

36

0.74, 0.42 to 1.31

‐‐

School Targeted

Highest ICC

4

1431

1023

0.75, 0.53 to 1.06

60

Lowest ICC

4

1431

1023

0.75, 0.53 to 1.06

60

School Universal

Highest ICC

5

4745

6313

0.74, 0.55 to 1.00

69

Lowest ICC

5

4715

6313

0.74, 0.54 to 1.03

82

Tobacco, alcohol, and/or illicit drug use

Family Targeted

Highest ICC

2

115

98

0.81, 0.50 to 1.33

0

Lowest ICC

2

115

98

0.81, 0.50 to 1.33

0

School Targeted

Highest ICC

5

244

98

0.55, 0.24 to 1.25

69

Lowest ICC

5

244

98

0.54, 0.24 to 1.21

71

School Universal

Highest ICC

2

2771

4256

1.13, 0.88 to 1.44

0

Lowest ICC

2

2771

4256

1.10, 1.01 to 1.20

0

Antisocial behaviour

Individual Targeted

Highest ICC

4

409

355

1.21, 0.92 to 1.60

19

Lowest ICC

4

409

355

1.21, 0.92 to 1.60

19

Family Targeted

Highest ICC

6

437

335

0.84, 0.57 to 1.24

42

Lowest ICC

6

437

335

0.84, 0.57 to 1.24

42

Family Universal

Highest ICC

1

208

98

0.87, 0.56 to 1.35

0

Lowest ICC

1

208

98

0.87, 0.56 to 1.35

0

School Targeted

Highest ICC

3

815

716

0.78, 0.59 to 1.05

0

Lowest ICC

3

815

716

0.82, 0.68 to 0.99

0

School Universal

Highest ICC

12

9960

10796

0.81, 0.66 to 0.98

66

Lowest ICC

12

9960

10796

0.82, 0.69 to 0.97

67

Sexual risk behaviour

Individual Targeted

Highest ICC

2

266

228

0.73, 0.49 to 1.08

45

Lowest ICC

2

266

228

0.73, 0.49 to 1.08

45

Individual Universal

Highest ICC

1

79

83

0.42, 0.14 to 1.25

‐‐

Lowest ICC

1

79

83

0.42, 0.14 to 1.25

‐‐

Family Targeted

Highest ICC

3

188

183

0.89, 0.55 to 1.44

0

Lowest ICC

3

188

183

0.89, 0.55 to 1.44

0

School Universal

Highest ICC

6

5757

6876

0.83, 0.61 to 1.12

77

Lowest ICC

6

5757

6876

0.83, 0.61 to 1.12

77

Physical activity

Individual Universal

Highest ICC

2

748

782

1.11, 0.74 to 1.67

0

Lowest ICC

2

748

782

1.11, 0.74 to 1.67

0

Family Targeted

Highest ICC

1

31

30

0.72, 0.29 to 1.79

‐‐

Lowest ICC

1

31

30

0.72, 0.29 to 1.79

‐‐

School Universal

Highest ICC

4

3547

2894

1.32, 1.16 to 1.50

0

Lowest ICC

4

3547

2894

1.33, 1.18 to 1.50

0

Nutrition (BMI)

Individual Universal

Highest ICC

1

421

158

0.80, 0.48 to 1.31

‐‐

Lowest ICC

1

421

158

0.80, 0.48 to 1.31

‐‐

School Universal

Highest ICC

3

2901

2116

0.84, 0.60 to 1.19

61

Lowest ICC

3

2901

2116

0.88, 0.62 to 1.23

69

Nutrition (unhealthy diet)

Individual Universal

Highest ICC

2

925

624

0.76, 0.42 to 1.34

51

Lowest ICC

2

925

624

0.76, 0.42 to 1.34

51

School Universal

Highest ICC

3

3608

2833

0.82, 0.64 to 1.06

49

Lowest ICC

3

3608

2833

0.85, 0.66 to 1.09

63

Educational attainment (academic performance)

Individual Targeted

Highest ICC

1

67

59

1.34, 0.71 to 2.52

‐‐

Lowest ICC

1

67

59

1.34, 0.71 to 2.52

‐‐

School Targeted

Highest ICC

3

619

628

0.91, 0.30 to 2.73

84

Lowest ICC

3

619

628

0.91, 0.39 to 2.14

85

School Universal

Highest ICC

3

602

393

0.94, 0.62 to 1.44

0

Lowest ICC

3

602

393

0.95, 0.74 to 1.22

0

Dichotomous vs continuous outcomes (studies with positive findings; 12‐month follow‐up only)

Tobacco use

School Universal

All

9

8365

6989

0.77, 0.60 to 0.97, P = 0.03

57

Dichotomous

7

7581

6275

0.72, 0.52 to 0.99, P = 0.05

60

Continuous

2

784

714

SMD ‐0.01,‐0.40 to 0.37, P = 0.95

84

Alcohol use

School Universal

All

8

4382

4369

0.72, 0.56 to 0.92, P = 0.009

58

Dichotomous

6

3598

3663

0.68, 0.51 to 0.90, P = 0.008

48

Continuous

2

784

706

SMD ‐0.12, ‐0.46 to 0.22; P = 0.49

79

Illicit drug use

School Universal

All

5

4745

6313

0.74, 0.55 to 1.00, P = 0.05

69

Dichotomous

4

2932

2808

0.67, 0.49 to 0.93

62

Continuous

1

1813

3505

SMD 0.06, 0.00 to 0.12, P = 0.04

‐‐

Cannabis use

School Universal

All

5

1924

2216

0.79, 0.62 to 1.01, P = 0.06

0

Dichotomous

4

1832

2130

0.82, 0.64 to 1.06, P = 0.13

0

Continuous

1

92

86

SMD ‐0.29, ‐0.58 to 0.01, P = 0.06

‐‐

Antisocial behaviour

School Universal

All

13

8445

9277

0.79, 0.64 to 0.97, P = 0.02

68

Dichotomous

4

4042

4339

0.55, 0.30 to 1.01, P = 0.06

87

Continuous

9

5708

6255

SMD ‐0.06,

‐0.11 to ‐0.0, P = 0.03

31

Sexual risk behaviour

School Universal

All

6

5757

6876

0.83, 0.61 to 1.12

77

Dichotomous

4

3020

2635

0.71, 0.39 to 1.30, P = 0.27

84

Continuous

2

2737

4241

SMD ‐0.03, ‐0.08 to 0.02, P = 0.2

0

Studies conducted in all countries vs high‐income countries only (studies with up to 12‐month follow‐up; meta‐analyses incorporating relevant data only)

Tobacco use

Individual Universal

All

2

925

624

1.03, 0.32 to 3.27

38

High‐income countries only

1

504

466

0.74, 0.43 to 1.28

‐‐

Alcohol use

Individual Universal

All

4

1105

806

0.80, 0.58 to 1.11

0

High‐income countries only

3

684

648

0.74, 0.54 to 1.06

0

Antisocial behaviour

School Universal

All

13

9960

10796

0.81, 0.66 to 0.98

66

High‐income countries only

12

8445

9277

0.79, 0.64 to 0.97

68

Sexual risk behaviour

School Universal

All

6

5757

6876

0.83, 0.61 to 1.12

77

High‐income countries only

5

5654

6779

0.81, 0.59 to 1.11

80

Physical activity

Individual Universal

All

2

748

782

1.11, 0.74 to 1.67

0

High‐income countries only

1

504

466

1.40, 0.67 to 2.94

‐‐

School Universal

All

4

3547

2894

1.32, 1.16 to 1.50

0

High‐income countries only

3

2533

1834

1.44, 1.20 to 1.74

0

Unhealthy diet

Individual Universal

All

2

925

624

0.76, 0.42 to 1.34

51

High‐income countries only

1

504

466

0.50, 0.23 to 1.08

‐‐

School Universal

All

3

3608

2833

0.82, 0.64 to 1.06

49

High‐income countries only

2

2594

1773

0.95, 0.76 to 1.19

0

CI: confidence interval.

ICC: intracluster correlation coefficient.

Figuras y tablas -
Table 6. Sensitivity analysis
Comparison 1. Tobacco

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Tobacco Use (short‐term) Show forest plot

15

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Targeted

2

521

Odds Ratio (Random, 95% CI)

0.98 [0.35, 2.73]

1.2 Individual Universal

2

1549

Odds Ratio (Random, 95% CI)

1.03 [0.32, 3.27]

1.3 Family Targeted

2

313

Odds Ratio (Random, 95% CI)

0.78 [0.40, 1.53]

1.4 School Universal

9

15354

Odds Ratio (Random, 95% CI)

0.77 [0.60, 0.97]

2 Tobacco Use (long‐term) Show forest plot

6

Odds Ratio (Random, 95% CI)

Subtotals only

2.1 Individual Targeted

1

397

Odds Ratio (Random, 95% CI)

1.08 [0.56, 2.11]

2.2 Family Targeted

2

1177

Odds Ratio (Random, 95% CI)

0.82 [0.32, 2.14]

2.3 Family Universal

1

237

Odds Ratio (Random, 95% CI)

0.82 [0.38, 1.78]

2.4 School Universal

2

879

Odds Ratio (Random, 95% CI)

0.60 [0.33, 1.09]

Figuras y tablas -
Comparison 1. Tobacco
Comparison 2. Alcohol

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Alcohol Use (short‐term) Show forest plot

19

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Targeted

4

2044

Odds Ratio (Random, 95% CI)

1.02 [0.80, 1.31]

1.2 Individual Universal

4

1911

Odds Ratio (Random, 95% CI)

0.80 [0.58, 1.11]

1.3 Family Targeted

3

417

Odds Ratio (Random, 95% CI)

0.83 [0.47, 1.46]

1.4 School Universal

8

8751

Odds Ratio (Random, 95% CI)

0.72 [0.56, 0.92]

2 Alcohol Use (long‐term) Show forest plot

7

Odds Ratio (Random, 95% CI)

Subtotals only

2.1 Family Targeted

3

1417

Odds Ratio (Random, 95% CI)

1.24 [0.69, 2.24]

2.2 Family Universal

1

237

Odds Ratio (Random, 95% CI)

0.86 [0.47, 1.55]

2.3 School Targeted

2

762

Odds Ratio (Random, 95% CI)

0.73 [0.52, 1.03]

2.4 School Universal

1

566

Odds Ratio (Random, 95% CI)

1.34 [0.55, 3.27]

Figuras y tablas -
Comparison 2. Alcohol
Comparison 3. Binge drinking

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Drunkenness or Excess Drinking (short‐term) Show forest plot

8

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Targeted

3

250

Odds Ratio (Random, 95% CI)

0.97 [0.68, 1.37]

1.2 School Universal

5

5494

Odds Ratio (Random, 95% CI)

0.66 [0.41, 1.06]

2 Drunkenness or Excess Drinking (long‐term) Show forest plot

2

Odds Ratio (Random, 95% CI)

Subtotals only

2.1 Family Targeted

1

240

Odds Ratio (Random, 95% CI)

1.30 [0.79, 2.13]

2.2 School Targeted

1

705

Odds Ratio (Random, 95% CI)

0.75 [0.55, 1.02]

Figuras y tablas -
Comparison 3. Binge drinking
Comparison 4. Illicit drug use

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Illicit drug use (short‐term) Show forest plot

11

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Targeted

3

638

Odds Ratio (Random, 95% CI)

0.94 [0.71, 1.25]

1.2 Family Targeted

1

69

Odds Ratio (Random, 95% CI)

0.74 [0.42, 1.31]

1.3 School Targeted

2

1299

Odds Ratio (Random, 95% CI)

0.96 [0.79, 1.18]

1.4 School Universal

5

11058

Odds Ratio (Random, 95% CI)

0.74 [0.55, 1.00]

2 Illicit drug use (long‐term) Show forest plot

9

Odds Ratio (Random, 95% CI)

Subtotals only

2.1 Family Targeted

4

2032

Odds Ratio (Random, 95% CI)

0.80 [0.52, 1.24]

2.2 School Targeted

2

819

Odds Ratio (Random, 95% CI)

1.07 [0.19, 6.21]

2.3 School Universal

3

3338

Odds Ratio (Random, 95% CI)

0.73 [0.56, 0.95]

Figuras y tablas -
Comparison 4. Illicit drug use
Comparison 5. Cannabis use

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Cannabis Use (short‐term) Show forest plot

12

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Targeted

2

126

Odds Ratio (Random, 95% CI)

1.10 [0.69, 1.76]

1.2 Individual Universal

2

362

Odds Ratio (Random, 95% CI)

0.69 [0.46, 1.04]

1.3 Family Targeted

3

380

Odds Ratio (Random, 95% CI)

1.02 [0.52, 2.02]

1.4 School Universal

5

4140

Odds Ratio (Random, 95% CI)

0.79 [0.62, 1.01]

2 Cannabis Use (long‐term) Show forest plot

6

Odds Ratio (Random, 95% CI)

Subtotals only

2.1 Family Targeted

2

340

Odds Ratio (Random, 95% CI)

0.53 [0.28, 1.02]

2.2 Family Universal

1

237

Odds Ratio (Random, 95% CI)

0.80 [0.44, 1.45]

2.3 School Targeted

2

806

Odds Ratio (Random, 95% CI)

0.82 [0.51, 1.32]

2.4 School Universal

1

566

Odds Ratio (Random, 95% CI)

1.13 [0.40, 3.21]

Figuras y tablas -
Comparison 5. Cannabis use
Comparison 6. Alcohol, tobacco, and/or drug use

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Composite Substance Use (short‐term) Show forest plot

7

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Family Targeted

2

213

Odds Ratio (Random, 95% CI)

0.81 [0.50, 1.33]

1.2 School Targeted

2

342

Odds Ratio (Random, 95% CI)

0.55 [0.24, 1.25]

1.3 School Universal

3

7390

Odds Ratio (Random, 95% CI)

1.03 [0.77, 1.37]

2 Composite Substance Use (long‐term) Show forest plot

6

Odds Ratio (Random, 95% CI)

Subtotals only

2.1 Family Targeted

4

1622

Odds Ratio (Random, 95% CI)

0.69 [0.47, 1.03]

2.2 School Universal

2

2145

Odds Ratio (Random, 95% CI)

1.09 [0.94, 1.27]

Figuras y tablas -
Comparison 6. Alcohol, tobacco, and/or drug use
Comparison 7. Antisocial behaviour and offending

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Antisocial Behaviour and Offending ‐ Any (short‐term) Show forest plot

27

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Targeted

4

764

Odds Ratio (Random, 95% CI)

1.21 [0.92, 1.60]

1.2 Individual Universal

1

200

Odds Ratio (Random, 95% CI)

1.02 [0.62, 1.69]

1.3 Family Targeted

6

772

Odds Ratio (Random, 95% CI)

0.84 [0.57, 1.24]

1.4 Family Universal

1

306

Odds Ratio (Random, 95% CI)

0.87 [0.56, 1.35]

1.5 School Targeted

3

1531

Odds Ratio (Random, 95% CI)

0.78 [0.59, 1.05]

1.6 School Universal

12

20756

Odds Ratio (Random, 95% CI)

0.81 [0.66, 0.98]

2 Violent Offences Show forest plot

13

Odds Ratio (Random, 95% CI)

Subtotals only

2.1 Individual Targeted

2

514

Odds Ratio (Random, 95% CI)

1.11 [0.56, 2.17]

2.2 Family Targeted

1

238

Odds Ratio (Random, 95% CI)

0.95 [0.49, 1.84]

2.3 Family Universal

1

306

Odds Ratio (Random, 95% CI)

0.87 [0.56, 1.35]

2.4 School Targeted

1

158

Odds Ratio (Random, 95% CI)

0.60 [0.31, 1.16]

2.5 School Universal

8

11347

Odds Ratio (Random, 95% CI)

0.86 [0.69, 1.07]

3 School or General Delinquency Show forest plot

14

Odds Ratio (Random, 95% CI)

Subtotals only

3.1 Individual Targeted

2

250

Odds Ratio (Random, 95% CI)

1.07 [0.61, 1.89]

3.2 Family Targeted

4

598

Odds Ratio (Random, 95% CI)

0.80 [0.54, 1.20]

3.3 School Targeted

3

1573

Odds Ratio (Random, 95% CI)

0.79 [0.59, 1.06]

3.4 School Universal

5

10113

Odds Ratio (Random, 95% CI)

0.88 [0.77, 1.00]

4 Antisocial Behaviour and Offending ‐ Any (long term) Show forest plot

11

Odds Ratio (Random, 95% CI)

Subtotals only

4.1 Family Targeted

5

2486

Odds Ratio (Random, 95% CI)

0.74 [0.54, 1.03]

4.2 Family Universal

1

304

Odds Ratio (Random, 95% CI)

0.67 [0.43, 1.04]

4.3 School Targeted

3

1177

Odds Ratio (Random, 95% CI)

0.71 [0.46, 1.10]

4.4 School Universal

2

4146

Odds Ratio (Random, 95% CI)

0.91 [0.63, 1.31]

Figuras y tablas -
Comparison 7. Antisocial behaviour and offending
Comparison 8. Vehicle‐related risk behaviours

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Driving Under the Influence of Alcohol and/or Drugs Show forest plot

2

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Targeted

2

94

Odds Ratio (Random, 95% CI)

0.59 [0.14, 2.48]

Figuras y tablas -
Comparison 8. Vehicle‐related risk behaviours
Comparison 9. Sexual risk behaviours

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Sexual Risk Behaviour (short‐term) Show forest plot

12

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Targeted

2

494

Odds Ratio (Random, 95% CI)

0.73 [0.49, 1.08]

1.2 Individual Universal

1

162

Odds Ratio (Random, 95% CI)

0.42 [0.14, 1.25]

1.3 Family Targeted

3

371

Odds Ratio (Random, 95% CI)

0.89 [0.55, 1.44]

1.4 School Universal

6

12633

Odds Ratio (Random, 95% CI)

0.83 [0.61, 1.12]

2 Sexual Risk Behaviour (long‐term) Show forest plot

8

Odds Ratio (Random, 95% CI)

Subtotals only

2.1 Individual Targeted

1

461

Odds Ratio (Random, 95% CI)

0.93 [0.64, 1.35]

2.2 Family Targeted

2

318

Odds Ratio (Random, 95% CI)

0.47 [0.31, 0.71]

2.3 Family Universal

1

237

Odds Ratio (Random, 95% CI)

1.12 [0.64, 1.96]

2.4 School Targeted

1

650

Odds Ratio (Random, 95% CI)

0.62 [0.47, 0.82]

2.5 School Universal

3

3391

Odds Ratio (Random, 95% CI)

0.74 [0.50, 1.09]

Figuras y tablas -
Comparison 9. Sexual risk behaviours
Comparison 10. Physical activity

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Physical Activity Show forest plot

7

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Universal

2

1530

Odds Ratio (Random, 95% CI)

1.11 [0.74, 1.67]

1.2 Family Targeted

1

61

Odds Ratio (Random, 95% CI)

0.72 [0.29, 1.79]

1.3 School Universal

4

6441

Odds Ratio (Random, 95% CI)

1.32 [1.16, 1.50]

Figuras y tablas -
Comparison 10. Physical activity
Comparison 11. Mental health

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Depressive Symptoms (short‐term) Show forest plot

4

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Targeted

1

124

Odds Ratio (Random, 95% CI)

1.02 [0.54, 1.93]

1.2 School Universal

3

3907

Odds Ratio (Random, 95% CI)

0.92 [0.71, 1.20]

2 Depressive Symptoms (long‐term) Show forest plot

5

Odds Ratio (Random, 95% CI)

Subtotals only

2.1 Family Targeted

4

2386

Odds Ratio (Random, 95% CI)

0.88 [0.80, 0.98]

2.2 School Targeted

1

721

Odds Ratio (Random, 95% CI)

0.68 [0.42, 1.09]

Figuras y tablas -
Comparison 11. Mental health
Comparison 12. Unhealthy diet

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 BMI Show forest plot

4

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Universal

1

579

Odds Ratio (Random, 95% CI)

0.80 [0.48, 1.31]

1.2 School Universal

3

5017

Odds Ratio (Random, 95% CI)

0.84 [0.60, 1.19]

2 Unhealthy Diet Show forest plot

5

Odds Ratio (Random, 95% CI)

Subtotals only

2.1 Individual Universal

2

1549

Odds Ratio (Random, 95% CI)

0.76 [0.42, 1.34]

2.2 School Universal

3

6441

Odds Ratio (Random, 95% CI)

0.82 [0.64, 1.06]

Figuras y tablas -
Comparison 12. Unhealthy diet
Comparison 13. School‐related outcomes

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Academic Performance (short‐term) Show forest plot

5

Odds Ratio (Random, 95% CI)

Subtotals only

1.1 Individual Targeted

1

126

Odds Ratio (Random, 95% CI)

1.34 [0.71, 2.52]

1.2 School Targeted

3

1247

Odds Ratio (Random, 95% CI)

0.91 [0.30, 2.73]

1.3 School Universal

1

579

Odds Ratio (Random, 95% CI)

0.94 [0.62, 1.44]

Figuras y tablas -
Comparison 13. School‐related outcomes