Intervenciones conductuales para mejorar el uso de preservativos en la protección dual
Resumen
Antecedentes
El sexo sin protección es un factor de riesgo principal de enfermedad, discapacidad y mortalidad en muchas áreas del mundo debido a la prevalencia y la incidencia de las infecciones de transmisión sexual (ITS) incluido el VIH. El preservativo masculino es uno de los métodos anticonceptivos más antiguos y es el primer método para prevenir la propagación del VIH. Cuando se utilizan de forma correcta y consistente los preservativos pueden proporcionar protección dual, es decir, contra el embarazo y contra el VIH y las ITS.
Objetivos
Se examinaron los estudios comparativos de intervenciones conductuales para mejorar el uso de preservativos. También hubo interés en la identificación de intervenciones asociadas con el uso efectivo de preservativos según lo medido con evaluaciones biológicas que pueden aportar pruebas objetivas de protección.
Métodos de búsqueda
Hasta septiembre de 2013 se efectuaron búsquedas en bases de datos computarizadas para obtener estudios comparativos de intervenciones conductuales para mejorar el uso de preservativos: MEDLINE, POPLINE, CENTRAL, EMBASE, LILACS, OpenGrey, COPAC, ClinicalTrials.gov y ICTRP. Se escribió a los investigadores para obtener datos faltantes.
Criterios de selección
Los estudios podían ser aleatorios o no aleatorios. Los mismos examinaron una intervención conductual para mejorar el uso de preservativos. La comparación podía ser otra intervención conductual, atención habitual o ninguna intervención. La intervención experimental tenía un componente educacional o de asesoramiento para promover o mejorar el uso de preservativos. Abordó la prevención del embarazo así como de la transmisión del VIH y las ITS. Los objetivos podían ser los preservativos masculinos o femeninos y dirigirse a individuos, parejas o comunidades. Los participantes potenciales incluyeron a mujeres heterosexuales y hombres heterosexuales.
Los estudios debían proporcionar datos de los resultados de las pruebas o registros sobre un resultado biológico: embarazo, VIH/ITS o presencia de semen según lo evaluado con un marcador biológico, p.ej. antígeno prostático específico. No se incluyeron los datos autoinformados sobre el sexo con o sin protección debido a las limitaciones del sesgo de recuerdo y de conveniencia social. Los resultados se midieron al menos tres meses después del inicio de la intervención conductual.
Obtención y análisis de los datos
Dos revisores evaluaron la elegibilidad de los resúmenes y extrajeron los datos de los estudios incluidos. Para los resultados dicotómicos se calculó el odds ratio (OR) de Mantel‐Haenszel con el IC del 95% mediante un modelo de efectos fijos. Los ensayos aleatorios grupales utilizaron diversos métodos para considerar el agrupamiento, como el modelo multinivel. En su mayoría los informes no proporcionaron información para calcular el tamaño de la muestra efectivo. Por lo tanto, los resultados se presentaron según lo informado por los investigadores. No se realizó un metanálisis debido a las diferencias en las intervenciones y las medidas de resultado.
Resultados principales
Siete estudios cumplieron los criterios de elegibilidad. Todos fueron ensayos controlados aleatorios; seis asignaron a grupos y uno asignó al azar a individuos. Los tamaños de muestra de los ensayos aleatorios grupales variaron de 2157 a 15 614; el número de grupos varió entre 18 y 70. Cuatro ensayos se realizaron en países de África, dos en EE.UU. y uno en Inglaterra. Tres se realizaron principalmente en escuelas, dos se realizaron en contextos comunitarios, uno tuvo lugar durante el entrenamiento militar y uno se realizó en un consultorio.
Cinco estudios proporcionaron datos sobre el embarazo a partir de pruebas de embarazo o de los registros nacionales de abortos y nacidos vivos. Cuatro ensayos evaluaron la incidencia o la prevalencia del VIH y el HSV‐2. Tres ensayos examinaron otra ITS. Los ensayos no mostraron ni informaron diferencias significativas entre los grupos de estudio en cuanto al embarazo o el VIH, pero hubo efectos favorables evidentes en algunas ITS. Dos mostraron una incidencia menor de HSV‐2 en el grupo de intervención conductual comparado con el grupo de atención habitual y se informaron cocientes de tasas ajustados (CTA) de 0,65 (IC del 95%: 0,43 a 0,97) y 0,67 (IC del 95%: 0,47 a 0,97), aunque el VIH no difirió significativamente. Uno también informó una menor incidencia de sífilis y prevalencia de gonorrea en el grupo de intervención conductual más tratamiento de la ITS en comparación con el grupo de atención habitual. Los CTA informados fueron 0,58 (IC del 95%: 0,35 a 0,96) y 0,28 (IC del 95%: 0,11 a 0,70), respectivamente. Otro estudio informó un efecto negativo sobre la gonorrea en las pacientes jóvenes del grupo de intervención versus el grupo control (CTA 1,93; IC del 95%: 1,01 a 3,71). La diferencia ocurrió entre las pacientes que recibieron solamente un año de intervención.
Conclusiones de los autores
Se encontraron pocos estudios y pocas pruebas clínicas de la efectividad de las intervenciones para promover el uso de preservativos para la protección dual. No se encontraron resultados favorables para el embarazo ni el VIH y solamente se encontraron algunos para otras ITS. La calidad global de las pruebas fue moderada a baja; las pérdidas durante el seguimiento fueron significativas. Se necesitan intervenciones efectivas para mejorar el uso de preservativos con el objetivo de prevenir el embarazo y la transmisión del VIH y las ITS. Las intervenciones deben ser posibles en contextos de recursos limitados y se deben probar con medidas de resultado válidas y confiables.
PICO
Resumen en términos sencillos
Programas para la prevención del embarazo y las enfermedades mediante la mejoría del uso de preservativos
El sexo sin protección puede provocar enfermedades y muerte en muchas áreas del mundo debido a las infecciones de transmisión sexual (ITS), incluido el VIH. El preservativo masculino es uno de los métodos más antiguos de regulación de la natalidad y el primer método que funciona para prevenir el VIH. Cuando se utilizan correctamente, los preservativos pueden proporcionar protección dual contra el embarazo y el VIH y las ITS. Se examinaron programas conductuales para mejorar el uso de preservativos.
Hasta septiembre de 2013 se efectuaron búsquedas computarizadas de estudios de programas para mejorar el uso de preservativos. Se escribió a los investigadores para obtener datos faltantes. Los estudios podían tener varios diseños. El programa de educación consideró la prevención del embarazo y del VIH y las ITS. La intervención se comparó con un programa diferente, atención habitual o ninguna intervención. Los estudios incluyeron un resultado clínico como pruebas de embarazo, del VIH o de ITS. No se utilizaron los autoinformes del uso de preservativos.
Se encontraron siete ensayos aleatorios. Seis asignaron grupos y uno asignó al azar a individuos. Cuatro ensayos se realizaron en países de África, dos en EE.UU. y uno en Inglaterra. Los estudios se realizaron en escuelas, contextos comunitarios, un consultorio y en el contexto del entrenamiento militar.
Cinco ensayos examinaron el embarazo, cuatro estudiaron el VIH y el HSV‐2 (herpes) y tres evaluaron otra ITS. No se encontraron diferencias importantes entre los grupos de estudio para el embarazo ni el VIH. Se observaron algunos hallazgos para los resultados de las ITS. Dos estudios mostraron menos casos de HSV‐2 con el programa conductual comparado con el grupo control. Uno también informó menos casos de sífilis y gonorrea con el programa conductual más tratamiento de las ITS. Otro estudio informó una tasa mayor de gonorrea en el grupo de intervención. Los investigadores consideraron que el resultado se debió a un subgrupo que no recibió el programa completo.
Se encontró un efecto clínico pequeño en cuanto a la mejoría del uso de preservativos. Los estudios proporcionaron información de calidad moderada a baja. Las pérdidas durante el seguimiento fueron significativas. Se necesitan programas adecuados sobre el uso de preservativos para prevenir el embarazo y el VIH y las ITS. Los programas deben ser útiles en contextos de escasos recursos. Las intervenciones se deben probar con medidas de resultado válidas.
Authors' conclusions
Background
Description of the condition
Unprotected sex is a major risk factor for disease, disability, and mortality in many areas of the world due to the prevalence and incidence of sexually transmitted infections (STI) including HIV (Warner 2012). Millions of women also need protection against unintended pregnancy, especially in lower‐resource areas (Thurman 2011). Multipurpose prevention technologies are being developed or improved to prevent HIV/STI and unplanned pregnancy (Friend 2010; Thurman 2011). Such technologies include physical and chemical barriers alone or in combination. Male and female condoms are existing physical barriers that can provide dual protection against both pregnancy and HIV/STI.
The male condom is one of the oldest contraceptive methods and the earliest method for preventing the spread of HIV (Crosby 2012). Condoms are inexpensive and widely available. In 'more developed' regions globally, the most prevalent forms of contraception, among women who are married or in a union, are condoms and contraceptive pills at 18% each (UN 2011). In 'less developed' areas, female sterilization leads at 21%, followed by use of an intrauterine device (IUD) (15%) and contraceptive pills (7%) (UN 2011). Only 6% of women report using male condoms in those regions. Use of the female condom has been low overall.
While use of two methods has been promoted for preventing pregnancy and disease, such 'dual‐method use' has been relatively low. In the USA, rates may range from 7% to 23%, depending on age and life situation (Sales 2010; Brown 2011; Eisenberg 2012). In Europe, dual‐method use is estimated to be higher, at 15% to 30% (Higgins 2012). Adherence to one contraceptive method is challenging for many people (Halpern 2011; Lopez 2013). Use of two methods can require multiple actions and the processing of multiple messages about risk, i.e., for pregnancy and HIV/STI (O'Leary 2011). Condoms alone may be used more frequently than dual methods, at least among adolescents. From urban clinics in the USA, African American adolescents reported condoms alone as the most frequently used contraceptive method (35%) Brown 2011. Across 24 countries in sub‐Saharan Africa, the range for condom use varied widely for adolescents (Doyle 2012): 5% to 67% for females and 8% to 81% for males. Among never‐married adolescent females, reported condom use at last sex averaged 22% in West Africa, 35% in East Africa, and 60% in Southern Africa.
When used correctly and consistently, condoms can provide dual protection, i.e., against both pregnancy and disease (Steiner 1999; CDC 2010; O'Leary 2011). For the male condom, the estimated first‐year pregnancy rate with typical use is 18% and the rate for perfect use is 2% (Trussell 2011). Typical use among experienced users may be much lower than 18%. Female sex workers in Nevada's legal brothels had very low rates for condom breakage and slippage (Albert 1995); such workers had regular state‐mandated testing for HIV/STI. Educational interventions focused on reducing condom errors may reduce the typical failure rate. For the first version of the female condom approved in the USA (FC1) (FHC 2013), the estimated pregnancy rates were 21% and 5%, respectively (Trussell 2011). In 2009, a second version of this female condom (FC2) (FHC 2013) was approved in the USA, but corresponding effectiveness data are not available. Oral contraceptives (OCs) have first‐year pregnancy rates of 9% with typical use and 0.3% with perfect use (Trussell 2011). Long‐acting reversible contraception (LARC), i.e., IUDs and implants, do not require regular user action. A large cohort study reported pregnancy rates of 0.27 per 100 participant‐years for LARC compared to 4.55 for those using pills, transdermal patch or intravaginal ring (Winner 2012).
Contraceptives that are more effective in typical use for preventing pregnancy do not protect against disease. Only condoms have that additional advantage. Consistent use of latex condoms can reduce risk of HIV transmission from an infected partner by 80% to 90% (USAID 2005). However, failure to use condoms correctly with every sex act can increase risk. Incorrect use includes donning the condom after starting the sex act and removing it before ejaculation (Warner 2012). Furthermore, users can experience condom malfunctioning, such as breakage or slippage. Condom problems appear common; 40% to 50% of users report at least one problem occurring over short time intervals (Hatherall 2007; Warner 2008). Consistent and correct use of condoms decreases the risk of STI, but inconsistent use provides little or no protection (USAID 2005). The female condom protects against both HIV/STI transmission and unintended pregnancy, but few studies have quantified the degree of efficacy (Beksinska 2011; Gallo 2012).
Description of the intervention
Use of condoms as the sole protection may be appropriate when exposure to infection is the primary concern due to the prevalence of HIV/STI or the individual's risk behavior (Cates 2002). If unplanned pregnancy is the major issue, then dual‐method use could be more helpful, given the greater effectiveness of hormonal contraception for preventing pregnancy (Cates 2002). The use of two methods may not be necessary if condoms are used correctly and consistently. Conversely, if condoms are not used correctly for pregnancy prevention, they are unlikely to prevent the transmission of HIV/STI either.
Encouraging effective condom use includes increasing use and promoting consistent and correct use (Warner 2012). Lack of use may still be the limiting factor for condoms (Steiner 1999; Warner 2012). Behavioral interventions to improve condom use often involve counseling, but may also include broader educational programs and communication campaigns. Programs targeted to individuals or couples may be based on direct oral communication and written materials. Social marketing can have greater reach than interventions directed to individuals or groups, and may increase awareness and promote use (Chapman 2012). However, interventions should address the advantages of condoms, barriers to use, and the challenges in using condoms correctly (Maticka‐Tyndale 2012). Interpersonal interaction may help communicate such complex use information and help build skills. Projects may also utilize technology, such as computer‐assisted interviews and mobile phone reminders, while others may engage community workers or peer educators. Identifying effective interventions can be challenging. Evaluations of condom promotion programs could benefit from using valid and reliable outcome measures rather than self‐reports alone (Crosby 2012).
Why it is important to do this review
Reviews of condom interventions often focus on specific types of studies or programs. Carvalho 2011 included RCTs of condom promotion for women with HIV. Others have reviewed programs in less developed areas, e.g., voluntary counseling and testing program (Fonner 2012), social marketing of condoms (Sweat 2012), and peer education (Medley 2009). Some reviews focused on specific populations, i.e., young people 13 to 19 years old (Picot 2012), people who use drugs (Meader 2013), and heterosexual men (Townsend 2013). Many reviews consider biological outcomes such as incidence of HIV/STI, but also examine social, psychological, and behavioral outcomes. Moreover, they do not necessarily require addressing pregnancy prevention, as they often focus on prevention of HIV/STI transmission.
We aimed to identify behavioral interventions associated with improved condom use. Successful promotional or educational programs could be adapted for other groups or locations (Zou 2012). Most interventions to promote condom use focus on prevention of HIV/STI. The motivation for preventing transmission may differ from that for preventing pregnancy. In this review, the behavioral interventions must have addressed contraception as well as prevention of HIV/STI. Studies had to have a clinical (biological) assessment of unprotected sex, which provides a more reliable and valid measure than self‐report. These measures include the incidence of pregnancy, HIV, or other STI and markers of semen exposure, e.g., prostate‐specific antigen (Gallo 2013).
Objectives
We examined comparative studies of behavioral interventions for improving condom use for dual protection. We were interested in identifying interventions associated with effective condom use as measured with biological assessments, which can provide objective evidence of protection.
Methods
Criteria for considering studies for this review
Types of studies
Studies were comparative and could be randomized or nonrandomized. They examined a behavioral intervention for improving condom use. The comparison could be another behavioral intervention, usual care, or no intervention.
Types of participants
Due to the focus on preventing pregnancy as well as HIV/STI transmission, participants could be heterosexual women or heterosexual men. Participants may have been at risk for pregnancy or HIV/STI and could be HIV‐positive or HIV‐negative.
Types of interventions
The behavioral intervention addressed the use of condoms specifically, that is, had an educational or counseling component to encourage or improve condom use. The focus could be on male or female condoms and targeted to individuals, couples, or communities. The intervention addressed preventing both pregnancy and the transmission of HIV/STI. We did not include behavioral interventions promoting dual‐method use, e.g., use of a hormonal method plus condoms.
The report described the content or process of the condom promotion intervention. Condom counseling described as 'standard' or 'routine' was not sufficient nor was standard contraception counseling that covered a range of contraceptive methods without a specific condom component. Such interventions would not be informative about how to improve condom use. However, they were acceptable for the comparison or control condition.
Types of outcome measures
Primary outcomes
The studies had to provide data on one or more of the following:
-
Pregnancy (test result or birth record)
-
HIV (test result)
-
Sexually transmitted infection (test result)
-
Presence of semen as assessed with a biological marker, e.g., prostate‐specific antigen (Gallo 2013).
Outcomes were measured three or more months after the behavioral intervention began, to provide evidence of protected sex over a minimum time period. We did not include self‐reported data on protected or unprotected sex, due to the limitations of recall and social desirability bias.
Search methods for identification of studies
Electronic searches
Through September 2013, we conducted searches of MEDLINE via PubMed, Cochrane Central Register of Controlled Trials (CENTRAL), POPLINE, EMBASE, and LILACS. Further searches for unpublished reports involved OpenGrey and COPAC. We also searched ClinicalTrials.gov and ICTRP for current trials and trials with results or relevant articles. The search strategies are given in Appendix 1.
Searching other resources
We examined reference lists of pertinent papers, including review articles, for additional citations. In addition, we contacted investigators in the field for other relevant published or unpublished studies.
Data collection and analysis
Selection of studies
We assessed for inclusion all titles and abstracts identified during the literature search. Two authors independently examined the search results for potentially eligible studies. Any discrepancies were resolved by discussion. For studies that appeared to be eligible for this review, we obtained and examined the full‐text articles.
Data extraction and management
Two authors extracted the data. One author entered the data into RevMan, and a second author checked accuracy. These data include the study characteristics, risk of bias (quality assessment), and outcomes. Any discrepancies were resolved by discussion.
Assessment of risk of bias in included studies
Intervention fidelity
We used the framework in Borrelli 2011 to assess the quality of the intervention reporting. Domains of treatment fidelity are study design, training of providers, delivery of treatment (intervention), receipt of treatment, and enactment of treatment skills. The framework was intended for assessing current trials. Criteria of interest for our review were as follows:
-
Study design ‐ had a curriculum or treatment manual.
-
Training ‐
-
specified provider credentials;
-
provided standardized training for the intervention.
-
-
Delivery ‐ assessed adherence to the protocol.
-
Receipt ‐ assessed participants' understanding and skills regarding the intervention.
Research design
For randomized controlled trials, we evaluated methodological quality according to recommended principles (Higgins 2011). That is, we examined the information on randomization method, allocation concealment, blinding, and losses to follow up and early discontinuation. For individually randomized trials, adequate methods for allocation concealment include a centralized telephone system and the use of sequentially‐numbered, opaque, sealed envelopes (Schulz 2002). In cluster randomized trials, clusters are usually randomized all at once, making allocation concealment less of an issue (Campbell 2012; Higgins 2011). However, selection bias may be introduced when individuals are approached for consent after the cluster has been randomized. Losses to follow up of 20% or more were considered high losses. Quality assessment also included the length of follow up. While three months was the minimum for inclusion in the review, a six‐month follow up provides more meaningful outcome measures.
We did not find any nonrandomized studies that had biological outcomes and met our other criteria. Therefore, we did not use the Newcastle‐Ottawa Scale (NOS) for nonrandomized studies (Wells 2013).
Measures of treatment effect
Outcomes listed in the Characteristics of included studies address the primary outcomes for this review. Study reports may have included other outcomes of interest to the investigators. If a study had data collection at three or more follow‐up visits, we used the first and last follow‐up assessments to measure short‐ and long‐term changes.
For RCTs with dichotomous outcomes, the Mantel‐Haenszel odds ratio (OR) with 95% confidence interval (CI) was calculated using a fixed‐effect model. An example is the incidence of a sexually transmitted infection. Fixed and random effects give the same result if no heterogeneity exists, as when a comparison includes only one study. For life‐table rates for pregnancy, we had planned to use the rate difference as the effect measure. However, none of the included studies had life‐table rates for pregnancy.
The cluster randomized trials used a variety of strategies to account for the clustering. When available, we used adjusted measures that the investigators considered as the primary effect measures. Odds ratio (OR) is an appropriate effect measure and is commonly provided when adjusted analyses are obtained using logistic regression models. However, if an appropriate adjusted OR was not available from the report, we considered other effect measures, e.g., rate ratio, hazard ratio, or incidence difference. Where multivariate models were used, we did not analyze the treatment effect as that would usually require individual participant data. Rather we presented the results from adjusted models as reported by the investigators.
Unit of analysis issues
If clustering was part of the design, we assessed whether the estimates were properly adjusted to account for clustering effects. The cluster randomized trials used various methods of accounting for the clustering, such as multilevel modeling. The specific methods are given in the results for each trial. Most reports did not provide sufficient information to calculate the effective sample size, so we did not analyze the data but presented the results as reported.
Dealing with missing data
If reports were missing data needed for analysis, we wrote to the study investigators. However, we limited our data requests to studies less than 10 years old. Researchers are unlikely to have access to data from older studies.
Assessment of heterogeneity
Given the diversity of design features and behavioral interventions, we did not conduct meta‐analysis for pooled estimates. Behavioral interventions can vary widely in design and content. In such cases, we assessed sources of heterogeneity without pooling the data. We addressed heterogeneity due to differences in study design, analysis strategies, and confounding adjustment. To interpret the intervention results, we examined the intervention location, i.e., country and setting (clinic, school, community); participant characteristics; and the intervention content, implementation, and fidelity information.
Data synthesis
We applied principles from GRADE to assess the evidence quality and address confidence in the effect estimates (Balshem 2011). However, when a meta‐analysis is not viable due to varied interventions, a summary of findings table is not feasible. Therefore, we did not conduct a formal GRADE assessment with an evidence profile and summary of findings table (Guyatt 2011).
Our assessment of the body of evidence was based on the quality of evidence from the included studies. We included our assessment of intervention fidelity in the overall quality assessment. Evidence quality could be high, moderate, low, or very low. RCTs were considered high quality initially, then downgraded a level for each of the following: a) randomization sequence generation and allocation concealment: no information on either, or one was inadequate; b) intervention fidelity information for three or fewer criteria; c) follow up less than 6 months; d) losses to follow up 20% or more.
Results
Description of studies
Results of the search
The database searches resulted in 5960 citations due to the broad nature of the search for educational interventions (Figure 1). A total of 1118 duplicates were removed (912 electronically and 206 by hand). This left 4842 items. We also identified 9 reports from other sources, such as reference lists. We discarded 4741 items based on the titles and abstracts. We reviewed the full text of 110 papers for eligibility as original studies or related articles. Many studies did not have an intervention, an appropriate study design, or a biological outcome. Our search also identified 96 unduplicated listings in ClinicalTrials.gov and ICTRP; no ongoing trials met our inclusion criteria.
Study flow diagram.
Included studies
Seven studies met our Criteria for considering studies for this review, along with eight secondary articles (two of which were relevant for both Ross 2007 and Cowan 2010). We examined another six papers related to these studies, but did not extract any information from them. Four of the included trials had been included in our review of theory‐based interventions (Lopez 2013). However, this review utilized different outcome data.
All seven studies were RCTs; six assigned clusters (platoons, villages, communities, schools) and one assigned individuals. Four cluster randomized trials were conducted in Africa: Uganda (Kamali 2003), Tanzania (Ross 2007), South Africa (Jewkes 2008), and Zimbabwe (Cowan 2010). The other two were conducted in the USA (Boyer 2005) and England (Stephenson 2008). The trial that randomized individuals was conducted in the USA (Petersen 2007). The settings varied: three were mainly based in schools with two also having community activities. Two others were conducted in community settings, one took place during training for military recruits, and one was clinic‐based. The ages of the target populations included adolescents for three trials, young women for two trials, and wider age ranges for the remaining two studies. Two trials focused on women; another included both male and female participants, but the biological data came from women (abortions and live births).
The trials were published from 2003 to 2010. Sample sizes for the six cluster‐randomized trials ranged from 2157 to 15,614; the number of clusters ranged from 18 to 70. Therefore, the effective sample sizes would be smaller due to the assignment of groups rather than individuals. The one individually‐randomized trial had 764 participants. All trials provided details on sample size calculations.
For the intervention, six trials provided multiple sessions in a group format. The remaining trial had one individual session and a booster follow‐up contact. Planned follow up ranged from 14 months to six years after baseline; the average was three years.
Comparisons
The interventions can be classed as follows:
-
Multi‐component intervention versus
-
standard program (Kamali 2003; Ross 2007; Jewkes 2008) or
-
deferred intervention (Cowan 2010);
-
-
Risk reduction versus health promotion (Boyer 2005);
-
Motivational interviewing versus general health counseling (Petersen 2007);
-
Peer‐led education versus standard (teacher‐led) program (Stephenson 2008).
Outcomes
Available outcome data are summarized below.
| Biological outcomes reported by study | |||
| Study | Pregnancy1 | HIV and HSV‐2 | Other STI |
| ‐‐‐ | x | x | |
| x | ‐‐‐ | x | |
| x | ‐‐‐ | x | |
| x | x | x | |
| ‐‐‐ | x | ‐‐‐ | |
| x | ‐‐‐ | ‐‐‐ | |
| x | x | ‐‐‐ | |
| Total studies | 5 | 4 | 4 |
1Results of pregnancy tests, except Stephenson 2008 (linked records from national data on abortions and live births).
Excluded studies
We excluded 39 reports for which we examined the full text; these represented 25 studies and 14 related papers. Two were completed trials for which manuscripts were being prepared for publication. We obtained additional information from an investigator for one completed trial (Bachanas 2013) and from a conference presentation for the other (Grossman 2012). Some excluded studies did not provide a behavioral intervention focused on condom use that also addressed contraception. Others were not comparative or did not have the outcome data we sought. Specifics can be found in Characteristics of excluded studies.
Risk of bias in included studies
Intervention fidelity
The reported fidelity information is presented by study (Table 1).
| Study | Curriculum or manual | Provider credentials | Training for intervention | Assessed adherence | Assessed intervention | Fidelity |
| No information | Experienced health educators trained for this project | 30 to 40 hours | Random observation of sessions and feedback from the project manager | 'Booster session' focused on progress toward risk reduction and effective method use | 4 | |
| 3‐year school curriculum with 10 to 15 lessons each school year | Teachers (T); class peer educators (CPE) selected by teachers and research staff; health workers (HW) from government health unit; condom promoters and deliverers (CPD), elected in village meetings | T: 1 week each year; | Supervision visits to intervention schools and clinics; observe in‐class and clinic sessions and clubs; check exercise books for sessions taught; annual feedback to teachers | No information | 4 | |
| Manual with 13 core sessions and 3 meetings; content and activities | Slightly older than participants; post‐school qualification; open‐mindedness and gender sensitivity | 3 weeks: course as participants, week of didactic; facilitated practice groups with manual, included observation and feedback | Not specific: | 3 meetings: treatment groups gather, present exercises, have dialogue | 4 | |
| 3 sessions with identified topics and activities. Peer‐educators prepared lessons. Intervention 'standardized as far as possible'. | Peer educators: 11th‐grade students. Trainers: expertise in sexual health promotion. | 2 days: sexual health information, participatory learning methods, classroom management and group facilitation. Plus 3 sessions pre‐training (needs assessment) and 1 follow up (lesson planning) | Not specific: Observation for process data (not supervision). | Participatory sessions were interactive; involved practice. | 4 | |
| Youth: 3‐year school curriculum (from Ross 2007); year 4, 24‐session program for out‐of‐school youth. | Not specific: Community, carefully selected community facilitators | Youth program, 'trained'; Nurses trained for clinic intervention ran some youth and parent sessions; Community facilitators 'trained' | Clinic: monthly visits, additional training as needed; assessed against standards Team of social scientists regularly assessed delivery; feedback provided and modifications made as needed | No information | 3 | |
| 16‐topic guide for large group discussions | Not specific: community educators (CE) self‐selected then approved by local committee | CE: 1 week plus 2‐day retraining 18 months later; Drama groups: 2 days plus monthly visits | Not specific: | No information | 2 | |
| 4 sessions with educational objectives and strategies; activities and materials | Not specific: | Not specific: Trained | No information | Last session involved describing, practicing, discussing | 2 | |
| Studies meeting criteria | 6 | 4 | 6 | 3 | 4 | 3.3 overall fidelity |
1Assessed participants' understanding and skills regarding the intervention.
-
Six studies specified how the intervention was standardized.
-
For training, four studies noted the qualifications of the providers and six mentioned specific training related to the study, though one did not specify the content or length.
-
Three studies had a means to assess delivery adherence.
-
Four had means to assess intervention receipt by participants.
Allocation
Of the seven included trials, all provided information on the randomization process, such as 'computer‐generated' or the use of permuted blocks. Kamali 2003 used shuffled cards. For Cowan 2010, the design was changed to a cross‐sectional survey due to out‐migration of the cohort. Four trials mentioned stratification (Ross 2007; Jewkes 2008; Stephenson 2008; Cowan 2010).
The individually randomized trial used sealed envelopes for allocation concealment (Petersen 2007). The cluster randomized trials identified the clusters prior to randomization except Jewkes 2008; all individuals meeting the inclusion criteria were eligible. Allocation concealment was considered unclear if the report did not indicate whether the recruiters of individuals or the potential participants were aware of the cluster allocation prior to the consent process.
Blinding
Two trials used some blinding (Kamali 2003; Jewkes 2008). The evaluators or interviewers were masked to the participant's assignment. Two other studies noted the participants were not blinded to study arm and three trial reports did not mention blinding. Double‐blinding is often not feasible for participants or providers in educational interventions, but the assessors could have been blinded to study arm.
Incomplete outcome data
Losses to follow up were 20% or more for five trials: Kamali 2003 (28% at 18 to 24 months; 39% at 36 to 48 months); Boyer 2005 (59% at 14 months); Ross 2007 (27% at 3 years); Jewkes 2008 (22% at 12 months; 25% at 24 months); and Stephenson 2008 (25% missing postal codes for matching national health records). In addition, for Cowan 2010, an interim survey showed nearly half of the cohort had migrated out of the area. Those remaining were determined to be lower risk. The investigators, with the data and safety monitoring board, changed the design to a cross‐sectional survey.
High losses to follow up threaten validity (Strauss 2005). Differential losses between treatment and control groups did not appear to be a major factor in these studies. Losses did not differ substantially across treatment arms.
Effects of interventions
Pregnancy
Five studies provided data on pregnancy, either from pregnancy tests (Boyer 2005; Cowan 2010; Petersen 2007; Ross 2007) or from national records of abortions and live births (Stephenson 2008). No significant difference was reported between the study arms in these trials. In Petersen 2007, individuals were randomized to the study groups. Pregnancy rates were not significantly different at 12 months for the two groups (OR 0.88; 95% CI 0.55 to 1.42) (Analysis 1.1). The other four studies were cluster randomized trials:
-
For Boyer 2005, the investigators calculated robust standard errors using the Huber‐White sandwich estimator in regression models assessing intervention effectiveness. Reportedly, no significant differences were found between the study groups in unplanned pregnancy by 14 months (Table 2).
-
In Ross 2007, a random‐effects model was used to account for the cluster effects in the analysis. Reportedly, pregnancy was not significantly different between the study arms at three years (Table 3). A secondary paper (Doyle 2010) reported on a cross‐sectional survey at nine years that did not include pregnancy tests.
-
The investigators in Stephenson 2008 used general estimating equations, accounting for correlation between schools, with robust standard errors. Reportedly, the study arms were not significantly different for the study's primary outcome of abortion by age 20 nor for live births by age 20.5 (Table 4).
-
Cowan 2010 assessed pregnancy prevalence in a cross‐sectional survey due to design change (Table 5). The investigators used generalized estimating equations with robust standard errors to account for the clustering. Reportedly, the comparison groups did not differ significantly for pregnancy prevalence in the main analysis nor in the subgroup analysis of respondents who attended a trial school during the study (Table 5).
| Outcomes by 14 months1,2 | Experimental | Control | Odds ratio (95% CI) |
| Unplanned pregnancy or | 17.9 | 23.9 | 1.41 (1.01 to 1.98) |
| Unplanned pregnancy (%) | 6.7 | 7.3 | not reported |
1Cluster randomized trial accounted for the cluster effects. Researchers calculated robust standard errors using the Huber‐White sandwich estimator in regression models. Independent variables were intervention group, sexual history, and time between assessments.
2Results as reported by researchers; insufficient data for analysis in this review.
3STI included chlamydia, gonorrhea, and trichomonas.
| Outcomes at 3 years1,2 | Gender | Intervention | Comparison | Adjusted |
| Pregnancy prevalence (%) | females | 19.2 | 18.0 | 1.09 (0.85 to 1.40) |
| HIV incidence (per 1000 person‐years) | females | 3.18 | 4.73 | 0.75 (0.34 to 1.66) |
| males | 0.43 | 0.30 | NA4 | |
| HSV‐2 prevalence (%) | females | 21.3 | 20.8 | 1.05 (0.83 to 1.32) |
| males | 11.3 | 12.5 | 0.92 (0.69 to 1.22) | |
| Syphilis prevalence (%) | females | 3.3 | 3.6 | 0.99 (0.67 to 1.46) |
| males | 1.4 | 1.8 | 0.78 (0.46 to 1.30) | |
| Chlamydia prevalence (%) | females | 4.9 | 3.6 | 1.37 (0.98 to 1.91) |
| males | 0.5 | 0.5 | 1.14 (0.53 to 2.43) | |
| Gonorrhea prevalence (%) | females | 2.4 | 1.2 | 1.93 (1.01 to 3.71) |
| males | 0.4 | 0.1 | NA4 | |
| Trichomonas prevalence (%) | females | 28.6 | 25.8 | 1.13 (0.92 to 1.37) |
1Cluster randomized trial accounted for the cluster effects by using a random‐effects model.
2Results as reported by researchers; insufficient data for analysis in this review.
3Adjusted for age group, stratum, ethnic group, and number of lifetime partners at baseline.
4NA = not applicable; reportedly, numbers were too small to justify comparison.
| Outcomes by age 201,2 | Intervention | Control | Adjusted odds ratio |
| Abortion by age 20 | 5.0 (4.0 to 6.3) | 5.0 (4.0 to 6.4) | 1.07 (0.80 to 1.42) |
| Live births by age 20.5 | 7.5 (5.9 to 9.6) | 10.6 (6.8 to 16.1) | 0.77 (0.51 to 1.15) |
1Cluster randomized trial accounted for cluster effects. Investigators used general estimating equations, accounting for correlation between schools, with robust standard errors.
2Results as reported by researchers; insufficient data for analysis in this review.
3CI reportedly account for clustering by school.
| Outcomes at 4 years1,2 | Gender | Intervention | Comparison | Adjusted odds |
| HIV prevalence (%) | females | 8.1 | 7.2 | 1.15 (0.81 to 1.64) |
| males | 1.7 | 1.3 | 1.20 (0.66 to 2.18) | |
| HSV‐2 prevalence (%) | females | 11.9 | 9.8 | 1.24 (0.93 to 1.65) |
| males | 1.8 | 1.5 | 1.23 (0.69 to 2.18) | |
| Pregnancy prevalence (%) | females | 7.7 | 8.1 | 0.92 (0.70 to 1.19) |
| Subgroup analysis4 | ||||
| HIV prevalence (%) | females | 6.5 | 3.8 | 1.65 (0.90 to 3.03) |
| males | 1.5 | 1.4 | 0.91 (0.35 to 3.24) | |
| HSV‐2 prevalence (%) | females | 7.5 | 5.9 | 1.21 (0.71 to 2.05) |
| males | 1.4 | 0.8 | 1.34 (0.51 to 3.53) | |
| Pregnancy prevalence (%) | females | 5.3 | 5.9 | 0.83 (0.50 to 1.35) |
1Cluster randomized trial accounted for the cluster effects. Investigators used generalized estimating equations
with robust standard errors.
2Results as reported by researchers; insufficient data for analysis in this review.
3Adjusted for age, strata, marital status, and education.
4Attended trial school and lived in trial community for duration of intervention (40% of survey respondents).
HIV and HSV‐2
Four trials assessed the incidence or prevalence of HIV and HSV‐2. The investigators reportedly did not find any significant difference in HIV outcomes between the study groups. Two trials reportedly showed lower rates for HSV‐2 in the intervention group compared to the control group. In Kamali 2003, incident events and person‐years at risk were combined across three rounds of data collection. Incidence rate ratios accounted for the matching of communities by comparing observed to expected (from a regression model). The incidence of HSV‐2 was reportedly lower by four years in the behavioral intervention group compared to the control group (reported adjusted rate ratio 0.65; 95% CI 0.43 to 0.97) (Table 6). For Jewkes 2008, the investigators used general linear mixed models with clusters treated as random effect. Incidence was analyzed for HIV and HSV‐2 by two years (Table 7). HSV‐2 was reportedly lower for the intervention group compared to the control group (reported adjusted rate ratio 0.67; 95% CI 0.47 to 0.97). The addition of the STI program did not appear to make a significant difference.
| Outcomes at 4 years1 | Behavioral | Intervention | Control | Adjusted rate ratio2 (95% CI) | |
| Intervention | Intervention + STI program | ||||
| HIV incidence (per 100 person‐years) | 0.61 | 0.81 | 0.80 | 0.92 (0.62 to 1.37) | 0.91 (0.56 to 1.47) |
| HSV‐2 incidence (per 100 person‐years) | 2.3 | 3.6 | 3.5 | 0.65 (0.43 to 0.97) | 1.02 (0.69 to 1.50) |
| Active syphilis incidence, any titre | 2.9 | 2.1 | 2.9 | 1.13 (0.70 to 1.82) | 0.81 (0.64 to 1.04) |
| Active syphilis incidence, titre >= 1/8 | 0.5 | 0.3 | 0.6 | 1.13 (0.61 to 2.07) | 0.58 (0.35 to 0.96) |
| Gonorrhea prevalence (%) | 1.02 | 0.51 | 1.17 | 0.74 (0.30 to 1.82) | 0.28 (0.11 to 0.70) |
| Chlamydia prevalence (%) | 1.35 | 1.92 | 1.81 | 0.59 (0.20 to 1.72) | 0.99 (0.71 to 1.39) |
1Results as reported by researchers; insufficient data for analysis in this review.
2Adjusted for age and sex. HIV, HSV‐2, and syphilis adjusted for respective baseline prevalence. Gonorrhea and chlamydia data pooled from 2 follow‐up assessments.
| Outcomes at 2 years1,2 | Gender | Intervention | Control | Adjusted rate |
| HIV incidence (per 100 person‐years) | overall | 3.46 | 4.07 | 0.95 (0.67 to 1.35) |
| women | 5.65 | 6.95 | ‐‐‐ | |
| men | 1.40 | 1.29 | ‐‐‐ | |
| HSV‐2 incidence (per 100 person‐years) | overall | 3.24 | 4.62 | 0.67 (0.47 to 0.97) |
| women | 5.35 | 7.71 | ‐‐‐ | |
| men | 1.46 | 2.04 | ‐‐‐ |
1Cluster randomized trial accounted for the cluster effects. Investigators used general linear mixed models with clusters treated as random effect.
2Results as reported by researchers; insufficient data for analysis in this review.
3Adjusted for stratum, sex, participant’s age, and baseline cluster prevalence of HIV or HSV‐2, respectively.
Two of the trials reported no significant differences in HIV or HSV‐2 outcomes. Ross 2007 assessed incidence at three years (Table 3). A cross‐sectional survey for this study examined prevalence at nine years (Table 8); respondents attended trial schools during the intervention period but did not necessarily participate in the study. For Cowan 2010, the investigators examined the prevalence in a cross‐sectional survey due to design change. The study groups reportedly did not differ significantly for either measure at four years (Table 5).
| Outcomes at 9 years1,2 | Gender | Intervention | Comparison | Adjusted prevalence |
| HIV prevalence (%) | females | 3.9 | 4.2 | 1.07 (0.68 to 1.67) |
| males | 2.0 | 1.7 | 0.91 (0.50 to 1.65) | |
| HSV‐2 prevalence (%) | females | 40.3 | 42.5 | 0.96 (0.87 to 1.06) |
| males | 25.0 | 26.7 | 0.94 (0.77 to 1.15) | |
| Syphilis seroprevalence (%) | females | 6.3 | 7.5 | 0.86 (0.62 to 1.21) |
| males | 5.8 | 5.3 | 1.06 (0.74 to 1.52) | |
| Active syphilis prevalence (%) | females | 4.5 | 5.2 | 0.91 (0.65 to 1.28) |
| males | 3.8 | 3.3 | 1.11 (0.72 to 1.72) | |
| Chlamydia prevalence (%) | females | 2.6 | 2.1 | 1.27 (0.87 to 1.86) |
| males | 2.1 | 2.1 | 1.24 (0.66 to 2.33) | |
| Gonorrhea prevalence (%) | females | 0.3 | 0.4 | 0.73 (0.20 to 2.63) |
| males | 0.3 | 0.4 | 0.71 (0.21 to 2.41) |
1Cross‐sectional survey; reportedly analyzed for stratified cluster randomized trial. Survey participants attended trial schools during the intervention period.
2Results as reported by researchers; insufficient data for analysis in this review.
3Adjusted for age group, stratum, and ethnic group.
Sexually transmitted infections
Three trials examined other STI. In Petersen 2007, chlamydia incidence reportedly was 1% overall and did not differ significantly between the study groups. The actual data were not reported. Two trials reported some intervention effect. Kamali 2003 assessed syphilis, gonorrhea, and chlamydia at four years (Table 6). While the behavioral intervention group reportedly did not differ significantly from the control, some effect was noted for the behavioral intervention plus STI program (Intervention‐plus). The incidence of active syphilis (high‐titer) was lower for the Intervention‐plus compared to the control group (reported adjusted rate ratio 0.58; 95% CI 0.35 to 0.96). The prevalence of gonorrhea was also lower for the Intervention‐plus versus the control group (reported adjusted rate ratio 0.28: 95% CI 0.11 to 0.70). For prevalence estimates, the investigators adjusted the standard errors in the logistic models using the Huber‐White sandwich estimator. Ross 2007 analyzed the prevalence of syphilis, gonorrhea, chlamydia, and trichomonas at three years (Table 3). Among the young women, the prevalence of gonorrhea was reportedly higher for the intervention group compared to the control (adjusted rate ratio 1.93; 95% CI 1.01 to 3.71). The groups did not differ significantly for syphilis or chlamydia. The cross‐sectional survey at 9 years also examined the prevalence of these STI among young people who attended trial schools during the intervention (Table 8). Reportedly, the comparison groups did not differ significantly.
Discussion
Summary of main results
Although we considered comparative studies of various designs, the only studies identified were randomized controlled trials. Our criteria for having a biological outcome likely limited the type of study. We summarized the outcome measures and evidence of effect (Table 9). As noted earlier (Unit of analysis issues), we were not able to analyze outcomes for the cluster randomized trials, so we presented them as reported by the investigators.
| Study | Pregnancy | HIV | HSV‐2 | Syphilis | Gonorrhea | Chlamydia | Trichomonas | Other |
| ND | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | |
| ND | ND | ND | ND | Females, I > C | ND | ND | ‐‐‐ | |
| ND1 | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | |
| ND | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ND2 | |
| ND | ND | ND | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | |
| ‐‐‐ | ND | I < C | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | |
| ‐‐‐ | ND | I < C | I‐plus < C3 | I‐plus < C | ND | ‐‐‐ | ‐‐‐ | |
| Studies with some | ‐‐‐ | ‐‐‐ | 2 | 1 | 2 | ‐‐‐ | ‐‐‐ | ‐‐‐ |
ND = No significant difference between study groups
I = Behavioral intervention
I‐plus = Behavioral intervention plus STI syndromic management
C = Comparison
1Abortions and live births
2Chlamydia, gonorrhea, trichomonas (grouped)
3High‐titre syphilis only
The trials showed or reported no significant differences between study groups for pregnancy or HIV, but favorable effects were evident for STI in some studies. Two studies showed a lower incidence of HSV‐2 in the intervention group, while HIV did not differ significantly (Kamali 2003; Jewkes 2008). One of those trials also reported lower syphilis incidence and gonorrhea prevalence for the group with the behavioral intervention plus STI management versus the usual‐care group. Another study reported one difference in an STI and that was a negative effect (Ross 2007). Young women in the special intervention group had a higher prevalence of gonorrhea than the control group. The investigators' analysis by school year at baseline indicated the difference occurred among those who had only one year of the intervention.
Overall completeness and applicability of evidence
Study characteristics. The trials were conducted in community settings, schools, a clinic, and a military training setting. Ages of the target populations varied. Six of the seven trials provided multiple sessions in a group format, which is more common for public health programs such as HIV prevention than for clinic‐based models of contraceptive counseling. Four trials took place in African countries, two in the USA, and one in England.
Applicability. Relevance of the few successful interventions to traditional contraceptive counseling may be limited. Contraceptive counseling typically focuses on individual women. Contact time might be a few minutes within a clinic visit or a separate session of 10 to 15 minutes. In such situations, expectations for behavior change should be limited. Effective interventions are needed, including some that can be adapted to clinical settings.
Theory base. Nearly all of these trials had an identified theoretical or model basis, which is also more common in HIV and STI prevention than in contraception counseling. High‐quality research on behavior change has been limited for reproductive health (Lopez 2013). A USA study explored attitudes and beliefs of clinicians about reproductive counseling. Most respondents believed they influenced their patients through their medical authority and the presentation of information (Henderson 2011). Those views are not consistent with current thinking about behavior change and patient‐centered counseling.
Adverse effects. These studies did not report on adverse effects or events (AE) as clinical trials would. Because these were educational interventions and not drug trials, side effects like physical pain or nausea were unlikely. However, unintended pyschosocial effects might occur, e.g., anxiety about discussing sexual behavior or concern about confidentiality. Such issues were not addressed in these reports. Unexpected behavioral effects were also possible, e.g., an increase in unsafe sexual behavior. For most studies, change in sexual behavior was a planned outcome measure, and might not be considered an AE. We did not include behavioral outcomes in our review. Some studies reported on negative trends in the planned outcome measures, such as more 'transactional sex' at the interim measure or more unintended pregnancies in the intervention group (Jewkes 2008).
Relevant subgroups. We did not find any analysis of relevant subgroups, such as those at high risk for pregnancy or HIV/STI. Outcomes were reported by demographic variables such as age or marital status, but not examined within group. Some differences were noted regarding changes in knowledge or behavior but not for biological outcomes. These studies were not likely powered to examine our outcomes by subgroup.
Quality of the evidence
As noted in Data synthesis, we did not conduct a formal GRADE assessment with an evidence profile and summary of findings table. Meta‐analysis was not viable due to the varied interventions, which is typical with these types of programs. Without a meta‐analysis, a summary of findings table is not feasible. We did use principles from GRADE to assess the evidence quality.
Our assessment was based on details from the individual studies. Most trials provided sufficient information. In many cases, design articles rather than outcome reports provided the detail. The studies also provided substantial information on the fidelity of implementation (Table 1). Four trials met four of our five fidelity criteria. The data most often lacking were provider credentials, assessing adherence to the protocol, and assessing intervention receipt. The overall quality of evidence is considered moderate to low for this review (Table 10). Of the three trials with evidence of high or moderate quality, one had evidence of an intervention effect and that was negative (Table 9). Losses to follow up were high in most trials. The only study with low losses was the clinic‐based trial with a 12‐month follow up, shorter than several others.
| Study | Randomization and | Intervention | Follow up | Losses < 20% | Quality of evidence1 |
| + | + | + | + | High | |
| + | + | + | ‐ | Moderate | |
| + | + | + | ‐ | Moderate | |
| + | ‐ | + | ‐ | Low | |
| + | ‐ | + | ‐ | Low | |
| ‐ | + | + | ‐ | Low | |
| ‐ | ‐ | + | ‐ | Very low | |
| Studies meeting criteria | 5 | 4 | 7 | 1 | Moderate to low overall |
1Quality could be high, moderate, low, or very low. RCTs were considered high quality then downgraded a level for each of the following: a) randomization sequence generation and allocation concealment: no information on either, or one was inadequate; b) intervention fidelity information for <= 3 criteria; c) follow up < 6 months; d) losses to follow up >= 20%.
2Due to out‐migration of cohort, losses were > 20%. Design was changed to cross‐sectional survey for final data collection.
Most studies reported an a priori sample size calculation for the biological outcomes of STI (Boyer 2005), HIV (Kamali 2003; Ross 2007; Jewkes 2008; Cowan 2010), or pregnancy (Stephenson 2008). The individually randomized trial (Petersen 2007) was powered to detect a difference in contraceptive use rather than our outcome of pregnancy.
Potential biases in the review process
The assessment of evidence quality involves judgment. We focused on five criteria for intervention fidelity that we believed to be pertinent in reviewing behavioral interventions and have found useful in conducting similar reviews. Other researchers might emphasize different criteria. In addition, we focused on certain design features for the overall quality assessment that are relevant to these types of studies. Again, others may have chosen different factors or cutoffs. Our approach and the evidence are presented for the reader.
Most trials were cluster randomized and the investigators conducted adjusted analyses. We could not analyze the data within the review, but presented the data as reported. Since most of the studies found no significant effect of the intervention, using the reported results is unlikely to have positively biased our conclusions. Given the differences in interventions and populations, we would not have conducted meta‐analysis even if we had the individual participant data.
We excluded many studies because they did not have pregnancy prevention as an intervention objective, although some used pregnancy as an outcome measure. Pregnancy incidence is a useful measure of unprotected sex, regardless of whether the intervention addresses contraception. While our eligibility criterion limited the number of studies, it helped focus the review and is unlikely to have introduced bias.
The interventions could have affected outcome measures that we did not include, such as contraceptive use or sexual behavior. We had extracted other outcome data from four trials for a different project (Lopez 2013). One trial showed a positive effect for the intervention group on condom use and on knowledge of pregnancy prevention (Ross 2007), but three did not show any difference in use of condoms or other contraceptives (Boyer 2005; Petersen 2007; Cowan 2010). Still, some studies assessed behavioral outcomes that we did not examine. The interventions might have had some effect on those measures.
Agreements and disagreements with other studies or reviews
Findings from related reviews may vary by eligibility criteria for interventions and outcomes. Many studies examine behavioral measures of condom use along with biological outcomes. Further, investigators may use several condom use measures, e.g., first condom use, condom use with last sex, or condom use with new or regular sex partner. We limited the outcomes to biological assessments, which are more reliable and valid indicators of unprotected sex than self‐report. The available evidence was further limited with our criterion of pregnancy prevention as part of the intervention along with prevention of HIV/STI.
Several reviews of behavioral interventions have included pregnancy prevention as an intervention or an outcome measure. Chin 2012 reviewed comprehensive risk‐reduction programs and reported favorable results for all their outcomes. However, most were self‐reported, including pregnancy and HIV incidence. The review of Blank 2012 focused on contraceptive service provision and use, and included pregnancy as an outcome. The interventions that promoted condom use had some favorable results for condom use but did not address pregnancy.
Four trials in the current review are also in a review of theory‐based interventions to improve contraceptive use (Lopez 2013). The outcome criteria for the earlier review included use of condoms and other contraceptives. Still, no significant differences were found in the behavioral measures in three studies included here (Boyer 2005; Petersen 2007; Cowan 2010). Ross 2007 reported some positive differences for the intervention group in condom use and knowledge of pregnancy prevention.
Others reviews have focused on promotion of condom use. Carvalho 2011 identified five eligible studies focused on women with HIV; the meta‐analysis did not show a difference in condom use. Free 2011 found 139 RCTs of interventions to promote condom use, but few studies had their primary outcome measures. Of 10 trials with data on 'any STI,' seven showed some reductions in STI for the intervention group (Free 2011). Four studies had self‐reported pregnancy, and three of those showed reductions in pregnancy. For 'condom use at last sex', 9 of 18 trials reported an increase in use. Many other reviews, such as Wariki 2012, focus on preventing HIV/STI transmission and did not address pregnancy prevention.
Study flow diagram.
Comparison 1 Pregnancy and STI prevention: motivational interviewing versus general health counseling, Outcome 1 Pregnancy (by 12 months).
| Study | Curriculum or manual | Provider credentials | Training for intervention | Assessed adherence | Assessed intervention | Fidelity |
| No information | Experienced health educators trained for this project | 30 to 40 hours | Random observation of sessions and feedback from the project manager | 'Booster session' focused on progress toward risk reduction and effective method use | 4 | |
| 3‐year school curriculum with 10 to 15 lessons each school year | Teachers (T); class peer educators (CPE) selected by teachers and research staff; health workers (HW) from government health unit; condom promoters and deliverers (CPD), elected in village meetings | T: 1 week each year; | Supervision visits to intervention schools and clinics; observe in‐class and clinic sessions and clubs; check exercise books for sessions taught; annual feedback to teachers | No information | 4 | |
| Manual with 13 core sessions and 3 meetings; content and activities | Slightly older than participants; post‐school qualification; open‐mindedness and gender sensitivity | 3 weeks: course as participants, week of didactic; facilitated practice groups with manual, included observation and feedback | Not specific: | 3 meetings: treatment groups gather, present exercises, have dialogue | 4 | |
| 3 sessions with identified topics and activities. Peer‐educators prepared lessons. Intervention 'standardized as far as possible'. | Peer educators: 11th‐grade students. Trainers: expertise in sexual health promotion. | 2 days: sexual health information, participatory learning methods, classroom management and group facilitation. Plus 3 sessions pre‐training (needs assessment) and 1 follow up (lesson planning) | Not specific: Observation for process data (not supervision). | Participatory sessions were interactive; involved practice. | 4 | |
| Youth: 3‐year school curriculum (from Ross 2007); year 4, 24‐session program for out‐of‐school youth. | Not specific: Community, carefully selected community facilitators | Youth program, 'trained'; Nurses trained for clinic intervention ran some youth and parent sessions; Community facilitators 'trained' | Clinic: monthly visits, additional training as needed; assessed against standards Team of social scientists regularly assessed delivery; feedback provided and modifications made as needed | No information | 3 | |
| 16‐topic guide for large group discussions | Not specific: community educators (CE) self‐selected then approved by local committee | CE: 1 week plus 2‐day retraining 18 months later; Drama groups: 2 days plus monthly visits | Not specific: | No information | 2 | |
| 4 sessions with educational objectives and strategies; activities and materials | Not specific: | Not specific: Trained | No information | Last session involved describing, practicing, discussing | 2 | |
| Studies meeting criteria | 6 | 4 | 6 | 3 | 4 | 3.3 overall fidelity |
| 1Assessed participants' understanding and skills regarding the intervention. | ||||||
| Outcomes by 14 months1,2 | Experimental | Control | Odds ratio (95% CI) |
| Unplanned pregnancy or | 17.9 | 23.9 | 1.41 (1.01 to 1.98) |
| Unplanned pregnancy (%) | 6.7 | 7.3 | not reported |
| 1Cluster randomized trial accounted for the cluster effects. Researchers calculated robust standard errors using the Huber‐White sandwich estimator in regression models. Independent variables were intervention group, sexual history, and time between assessments. | |||
| Outcomes at 3 years1,2 | Gender | Intervention | Comparison | Adjusted |
| Pregnancy prevalence (%) | females | 19.2 | 18.0 | 1.09 (0.85 to 1.40) |
| HIV incidence (per 1000 person‐years) | females | 3.18 | 4.73 | 0.75 (0.34 to 1.66) |
| males | 0.43 | 0.30 | NA4 | |
| HSV‐2 prevalence (%) | females | 21.3 | 20.8 | 1.05 (0.83 to 1.32) |
| males | 11.3 | 12.5 | 0.92 (0.69 to 1.22) | |
| Syphilis prevalence (%) | females | 3.3 | 3.6 | 0.99 (0.67 to 1.46) |
| males | 1.4 | 1.8 | 0.78 (0.46 to 1.30) | |
| Chlamydia prevalence (%) | females | 4.9 | 3.6 | 1.37 (0.98 to 1.91) |
| males | 0.5 | 0.5 | 1.14 (0.53 to 2.43) | |
| Gonorrhea prevalence (%) | females | 2.4 | 1.2 | 1.93 (1.01 to 3.71) |
| males | 0.4 | 0.1 | NA4 | |
| Trichomonas prevalence (%) | females | 28.6 | 25.8 | 1.13 (0.92 to 1.37) |
| 1Cluster randomized trial accounted for the cluster effects by using a random‐effects model. | ||||
| Outcomes by age 201,2 | Intervention | Control | Adjusted odds ratio |
| Abortion by age 20 | 5.0 (4.0 to 6.3) | 5.0 (4.0 to 6.4) | 1.07 (0.80 to 1.42) |
| Live births by age 20.5 | 7.5 (5.9 to 9.6) | 10.6 (6.8 to 16.1) | 0.77 (0.51 to 1.15) |
| 1Cluster randomized trial accounted for cluster effects. Investigators used general estimating equations, accounting for correlation between schools, with robust standard errors. | |||
| Outcomes at 4 years1,2 | Gender | Intervention | Comparison | Adjusted odds |
| HIV prevalence (%) | females | 8.1 | 7.2 | 1.15 (0.81 to 1.64) |
| males | 1.7 | 1.3 | 1.20 (0.66 to 2.18) | |
| HSV‐2 prevalence (%) | females | 11.9 | 9.8 | 1.24 (0.93 to 1.65) |
| males | 1.8 | 1.5 | 1.23 (0.69 to 2.18) | |
| Pregnancy prevalence (%) | females | 7.7 | 8.1 | 0.92 (0.70 to 1.19) |
| Subgroup analysis4 | ||||
| HIV prevalence (%) | females | 6.5 | 3.8 | 1.65 (0.90 to 3.03) |
| males | 1.5 | 1.4 | 0.91 (0.35 to 3.24) | |
| HSV‐2 prevalence (%) | females | 7.5 | 5.9 | 1.21 (0.71 to 2.05) |
| males | 1.4 | 0.8 | 1.34 (0.51 to 3.53) | |
| Pregnancy prevalence (%) | females | 5.3 | 5.9 | 0.83 (0.50 to 1.35) |
| 1Cluster randomized trial accounted for the cluster effects. Investigators used generalized estimating equations | ||||
| Outcomes at 4 years1 | Behavioral | Intervention | Control | Adjusted rate ratio2 (95% CI) | |
| Intervention | Intervention + STI program | ||||
| HIV incidence (per 100 person‐years) | 0.61 | 0.81 | 0.80 | 0.92 (0.62 to 1.37) | 0.91 (0.56 to 1.47) |
| HSV‐2 incidence (per 100 person‐years) | 2.3 | 3.6 | 3.5 | 0.65 (0.43 to 0.97) | 1.02 (0.69 to 1.50) |
| Active syphilis incidence, any titre | 2.9 | 2.1 | 2.9 | 1.13 (0.70 to 1.82) | 0.81 (0.64 to 1.04) |
| Active syphilis incidence, titre >= 1/8 | 0.5 | 0.3 | 0.6 | 1.13 (0.61 to 2.07) | 0.58 (0.35 to 0.96) |
| Gonorrhea prevalence (%) | 1.02 | 0.51 | 1.17 | 0.74 (0.30 to 1.82) | 0.28 (0.11 to 0.70) |
| Chlamydia prevalence (%) | 1.35 | 1.92 | 1.81 | 0.59 (0.20 to 1.72) | 0.99 (0.71 to 1.39) |
| 1Results as reported by researchers; insufficient data for analysis in this review. | |||||
| Outcomes at 2 years1,2 | Gender | Intervention | Control | Adjusted rate |
| HIV incidence (per 100 person‐years) | overall | 3.46 | 4.07 | 0.95 (0.67 to 1.35) |
| women | 5.65 | 6.95 | ‐‐‐ | |
| men | 1.40 | 1.29 | ‐‐‐ | |
| HSV‐2 incidence (per 100 person‐years) | overall | 3.24 | 4.62 | 0.67 (0.47 to 0.97) |
| women | 5.35 | 7.71 | ‐‐‐ | |
| men | 1.46 | 2.04 | ‐‐‐ | |
| 1Cluster randomized trial accounted for the cluster effects. Investigators used general linear mixed models with clusters treated as random effect. | ||||
| Outcomes at 9 years1,2 | Gender | Intervention | Comparison | Adjusted prevalence |
| HIV prevalence (%) | females | 3.9 | 4.2 | 1.07 (0.68 to 1.67) |
| males | 2.0 | 1.7 | 0.91 (0.50 to 1.65) | |
| HSV‐2 prevalence (%) | females | 40.3 | 42.5 | 0.96 (0.87 to 1.06) |
| males | 25.0 | 26.7 | 0.94 (0.77 to 1.15) | |
| Syphilis seroprevalence (%) | females | 6.3 | 7.5 | 0.86 (0.62 to 1.21) |
| males | 5.8 | 5.3 | 1.06 (0.74 to 1.52) | |
| Active syphilis prevalence (%) | females | 4.5 | 5.2 | 0.91 (0.65 to 1.28) |
| males | 3.8 | 3.3 | 1.11 (0.72 to 1.72) | |
| Chlamydia prevalence (%) | females | 2.6 | 2.1 | 1.27 (0.87 to 1.86) |
| males | 2.1 | 2.1 | 1.24 (0.66 to 2.33) | |
| Gonorrhea prevalence (%) | females | 0.3 | 0.4 | 0.73 (0.20 to 2.63) |
| males | 0.3 | 0.4 | 0.71 (0.21 to 2.41) | |
| 1Cross‐sectional survey; reportedly analyzed for stratified cluster randomized trial. Survey participants attended trial schools during the intervention period. | ||||
| Study | Pregnancy | HIV | HSV‐2 | Syphilis | Gonorrhea | Chlamydia | Trichomonas | Other |
| ND | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | |
| ND | ND | ND | ND | Females, I > C | ND | ND | ‐‐‐ | |
| ND1 | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | |
| ND | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ND2 | |
| ND | ND | ND | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | |
| ‐‐‐ | ND | I < C | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | ‐‐‐ | |
| ‐‐‐ | ND | I < C | I‐plus < C3 | I‐plus < C | ND | ‐‐‐ | ‐‐‐ | |
| Studies with some | ‐‐‐ | ‐‐‐ | 2 | 1 | 2 | ‐‐‐ | ‐‐‐ | ‐‐‐ |
| ND = No significant difference between study groups | ||||||||
| Study | Randomization and | Intervention | Follow up | Losses < 20% | Quality of evidence1 |
| + | + | + | + | High | |
| + | + | + | ‐ | Moderate | |
| + | + | + | ‐ | Moderate | |
| + | ‐ | + | ‐ | Low | |
| + | ‐ | + | ‐ | Low | |
| ‐ | + | + | ‐ | Low | |
| ‐ | ‐ | + | ‐ | Very low | |
| Studies meeting criteria | 5 | 4 | 7 | 1 | Moderate to low overall |
| 1Quality could be high, moderate, low, or very low. RCTs were considered high quality then downgraded a level for each of the following: a) randomization sequence generation and allocation concealment: no information on either, or one was inadequate; b) intervention fidelity information for <= 3 criteria; c) follow up < 6 months; d) losses to follow up >= 20%. | |||||
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Pregnancy (by 12 months) Show forest plot | 1 | 737 | Odds Ratio (M‐H, Fixed, 95% CI) | 0.88 [0.55, 1.42] |
