School‐based education programmes for the prevention of unintentional injuries in children and young people

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

●

‐

‐

●

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐: not applicable; ●: age range covered.

Nineteen (70%) studies specified that boys and girls were included in the analysis (Buckley 2010; Campbell 2001; Carmel 1991; Chapman 2013; Collard 2010; Cook 2006; Falavigna 2012; Gresham 2001; Kendrick 2007; Lee 2004; Martinez 1996; Morrongiello 1998; Reed 2001; Terzidis 2007; Twisk 2013; Wang 2009; Wesner 2003; Wright 1995; Zirkle 2005), but the remaining eight (30%) did not make this clear (Azeredo 2003; Frederick 2000; Grant 1992; Greene 2002; Lu 2000; Richards 1991; Sun 2004; Zhao 2006). A total of 13 (48%) studies specified that the intervention was delivered in state‐funded schools (Buckley 2010; Campbell 2001; Carmel 1991; Collard 2010; Grant 1992; Kendrick 2007; Lee 2004; Morrongiello 1998; Richards 1991; Sun 2004; Wang 2009; Wesner 2003; Wright 1995), three (11%) were in both state and independent schools (Azeredo 2003; Cook 2006; Falavigna 2012), and 11 (41%) did not specify the type of school (Chapman 2013; Frederick 2000; Greene 2002; Gresham 2001; Lu 2000; Martinez 1996; Reed 2001; Terzidis 2007; Twisk 2013; Zhao 2006; Zirkle 2005). Five (5%) programmes were only in urban schools (Carmel 1991; Falavigna 2012; Gresham 2001; Kendrick 2007; Sun 2004) and three (9%) were only in a rural school (Grant 1992; Lee 2004; Reed 2001). For the others, three (9%) were based in a rural and urban setting (Azeredo 2003; Richards 1991; Zhao 2006); one (4%) was in urban and suburban areas (Collard 2010); one (4%) was in urban, suburban and rural areas (Wright 1995); and for 14 (52%) studies details of the setting were not provided (Buckley 2010; Campbell 2001; Chapman 2013; Cook 2006; Frederick 2000; Greene 2002; Lu 2000; Martinez 1996; Morrongiello 1998; Terzidis 2007; Twisk 2013; Wang 2009; Wesner 2003; Zirkle 2005).

Types of interventions

All studies included primary prevention in the intervention. In addition, 16 (59%) studies also had secondary prevention components (Azeredo 2003; Cook 2006; Falavigna 2012; Grant 1992; Greene 2002; Gresham 2001; Kendrick 2007; Lee 2004; Martinez 1996; Morrongiello 1998; Reed 2001, Richards 1991; Wesner 2003; Wright 1995; Zhao 2006; Zirkle 2005), and three (11%) had elements of tertiary prevention (e.g. first aid) but these components were not included in the analysis (Buckley 2010; Campbell 2001; Zhao 2006). The injury mechanisms that each intervention was aimed at are described in Table 2.

Open in table viewer

Table 2. Injury mechanisms targeted for included studies

Study ID

Injury mechanism

Pedestrian

Cycle

Motorcycle

Vehicle (non‐specific)

Falls

Swimming/ drowning/ diving

Smoke/ fire/ flames

Heat and hot surfaces

Sport/ physical activity

Sun

Household safety

Agricultural injuries

Poisoning

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

‐

●

‐

‐

●

‐

●

‐

●

‐

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

‐

●

‐

●

‐

●

‐

●

‐

‐

●

‐

●

‐

●

‐

‐

●

‐

‐

●

‐

●

‐

‐

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

See: Summary of findings for the main comparison School injury prevention programmes compared to controls for the prevention of unintentional injuries in children and young people

●

‐

‐: outcome not measured; ●: outcome measured.

Eighteen (66%) interventions were targeted at children alone (Carmel 1991; Chapman 2013; Cook 2006; Falavigna 2012; Frederick 2000; Grant 1992; Greene 2002; Lee 2004; Martinez 1996; Morrongiello 1998; Reed 2001; Richards 1991; Terzidis 2007; Twisk 2013; Wang 2009; Wesner 2003; Wright 1995; Zirkle 2005), and nine (34%) were targeted at children and families (Azeredo 2003; Buckley 2010; Campbell 2001; Collard 2010; Gresham 2001; Kendrick 2007; Lu 2000; Sun 2004; Zhao 2006). While the major component of all interventions was based in the school setting, three (11%) also had a component in the family home (Campbell 2001; Lu 2000; Terzidis 2007), two (7%) in the community (Lee 2004; Lu 2000), and one (4%) in a hospital (Frederick 2000). Sixteen (59%) studies were delivered in whole or part by a teacher (Azeredo 2003; Buckley 2010; Carmel 1991; Chapman 2013; Collard 2010; Frederick 2000; Grant 1992; Greene 2002; Gresham 2001; Kendrick 2007; Lu 2000; Morrongiello 1998; Richards 1991; Terzidis 2007; Wesner 2003; Zhao 2006), and nine (33%) were delivered by professionals with an injury prevention role (Campbell 2001; Cook 2006; Falavigna 2012; Lee 2004; Martinez 1996; Reed 2001; Wang 2009; Wright 1995; Zirkle 2005). One (4%) study trained students to be peer educators (Wang 2009), and two (7%) studies did not explicitly report who delivered the intervention in the school (Sun 2004; Twisk 2013).

Studies used a variety of components as part of their education programmes: nine (33%) interventions utilised information giving (Carmel 1991; Falavigna 2012; Greene 2002; Richards 1991; Terzidis 2007; Twisk 2013; Wesner 2003; Wright 1995; Zirkle 2005), one (4%) used peer education (Wang 2009), and the other 17 studies (63%) had multi‐component programmes (Azeredo 2003; Buckley 2010; Campbell 2001; Chapman 2013; Collard 2010; Cook 2006; Frederick 2000; Grant 1992; Gresham 2001; Kendrick 2007; Lee 2004; Lu 2000; Martinez 1996; Morrongiello 1998; Reed 2001; Sun 2004; Zhao 2006). Multi‐component programmes included combinations of elements such as information giving, safety equipment provision, skills training and testing, physical training, cognitive behaviour change methods, telling of 'real life' stories by or about injured individuals and interactive learning (e.g. developing campaign materials).

Seven (26%) studies involved single education sessions (Cook 2006; Falavigna 2012; Morrongiello 1998; Terzidis 2007; Twisk 2013; Wesner 2003; Wright 1995); nine (33%) involved between three and eight sessions (Buckley 2010; Campbell 2001; Chapman 2013; Greene 2002; Gresham 2001; Lu 2000; Martinez 1996; Richards 1991; Zhao 2006); one (4%) involved 18 to 27 sessions (Azeredo 2003), and one (4%) involved over 50 sessions (Collard 2010). Nine (33%) studies did not document the number of sessions (Carmel 1991; Frederick 2000; Grant 1992; Kendrick 2007; Lee 2004; Reed 2001; Sun 2004; Wang 2009; Zirkle 2005).

Interventions were varied in terms of the duration of individual sessions, the frequency of these sessions and the overall duration of the intervention. Seven (26%) interventions were of very short duration, lasting only one day or one or two sessions (Cook 2006; Falavigna 2012; Morrongiello 1998; Terzidis 2007; Twisk 2013; Wesner 2003; Wright 1995). Two (7%) interventions lasted for between one and four weeks (Campbell 2001; Martinez 1996), seven (26%) lasted between one and six months (Azeredo 2003; Buckley 2010; Campbell 2001; Chapman 2013; Greene 2002; Gresham 2001; Richards 1991), and six (22%) lasted longer than six months (Collard 2010; Lu 2000; Sun 2004; Wang 2009; Zhao 2006; Zirkle 2005). Five (19%) studies did not document the length of the intervention (Frederick 2000; Grant 1992; Kendrick 2007; Lee 2004; Reed 2001).

Thirteen (48%) studies described 'branded' programmes including Think First and Think Well (brain and spinal cord injury prevention programmes: Falavigna 2012; Greene 2002; Gresham 2001; Wesner 2003; Wright 1995; Zirkle 2005), SPIY (Buckley 2010), I‐Play (Collard 2010), IMPS (Frederick 2000) based in schools and emergency departments, Learn Not To Burn (Grant 1992), Risk Watch (delivered by the fire service but covering a range of injury mechanisms: Kendrick 2007), Go AHEAD, including a range of road safety and sports injury prevention (Morrongiello 1998), and AgDARE, aimed at young people working on farms (Reed 2001).

The studies covered a range of injury mechanisms: 18 (67%) reported transport safety (including 12 (44%) pedestrian (Buckley 2010; Cook 2006; Frederick 2000; Greene 2002; Kendrick 2007; Lu 2000; Martinez 1996; Richards 1991; Twisk 2013; Wright 1995; Zhao 2006; Zirkle 2005), 17 (63%) cycling (Azeredo 2003; Buckley 2010; Chapman 2013; Cook 2006; Falavigna 2012; Frederick 2000; Greene 2002; Gresham 2001; Kendrick 2007; Lu 2000; Martinez 1996; Morrongiello 1998; Richards 1991; Twisk 2013; Wesner 2003; Wright 1995; Zirkle 2005), nine (33%) motor cycle (Buckley 2010; Chapman 2013; Falavigna 2012; Frederick 2000; Lu 2000; Martinez 1996; Wright 1995; Zhao 2006; Zirkle 2005), and 17 (63%) non‐specific vehicle injuries (Azeredo 2003; Buckley 2010; Chapman 2013; Cook 2006; Falavigna 2012; Frederick 2000; Greene 2002; Gresham 2001; Lu 2000; Martinez 1996; Morrongiello 1998; Richards 1991; Wesner 2003; Wright 1995; Zhao 2006; Zirkle 2005)), five (19%) reported falls safety (Falavigna 2012; Kendrick 2007; Lu 2000; Richards 1991; Sun 2004), 10 (37%) reported water/drowning safety (Azeredo 2003; Falavigna 2012; Frederick 2000; Greene 2002; Gresham 2001; Morrongiello 1998; Richards 1991; Terzidis 2007; Wesner 2003; Zhao 2006), seven (26%) reported smoke/fire safety (Azeredo 2003; Campbell 2001; Carmel 1991; Frederick 2000; Grant 1992; Kendrick 2007; Lu 2000), seven (26%) reported sports safety (Campbell 2001; Collard 2010; Greene 2002; Gresham 2001; Lu 2000; Morrongiello 1998; Richards 1991), three (11%) reported household safety (Campbell 2001; Frederick 2000; Zhao 2006), two (7%) reported agricultural safety (Lee 2004; Reed 2001), and five (19%) reported poisoning safety (Buckley 2010; Campbell 2001; Frederick 2000; Kendrick 2007; Zhao 2006). Table 2 shows the injury mechanisms targeted by each study.

Twenty‐one (78%) studies documented that the control groups had no intervention or the usual curriculum (not related to injury prevention) with either no access to the intervention or delayed until the end of the study (Azeredo 2003; Buckley 2010; Carmel 1991; Chapman 2013; Collard 2010; Cook 2006; Falavigna 2012; Frederick 2000; Grant 1992; Greene 2002; Kendrick 2007; Lee 2004; Lu 2000; Martinez 1996; Morrongiello 1998; Reed 2001; Terzidis 2007; Twisk 2013; Wang 2009; Wesner 2003; Wright 1995). In three studies (11%), the control groups received an alternative educational programme (e.g. tobacco and alcohol prevention programmes or general information on food hygiene and disease prevention) (Campbell 2001; Sun 2004; Zhao 2006), and in the remaining three (11%), it was unclear what the control group received (Gresham 2001; Richards 1991; Zirkle 2005).

Types of outcome measures

For the primary outcomes, five studies (19%) included non‐medically attended injury occurrence (Chapman 2013; Collard 2010; Lee 2004; Sun 2004; Wang 2009), one (4%) included medically attended injury occurrence (Zhao 2006), and one (4%) included both medically and non‐medically attended injuries (Lu 2000). Of these, five studies reported injury incidence rates (Collard 2010; Lu 2000; Sun 2004; Wang 2009; Zhao 2006), and two reported injury incidence proportion (Chapman 2013; Lee 2004).

For the secondary outcome measures, two studies (7%) reported observed safety skills (Frederick 2000; Kendrick 2007), four (15%) reported observed safety behaviours (Azeredo 2003; Reed 2001; Wright 1995; Zirkle 2005), 19 (70%) reported self‐reported behaviour and practices (Azeredo 2003; Buckley 2010; Campbell 2001; Chapman 2013; Collard 2010; Falavigna 2012; Frederick 2000; Gresham 2001; Kendrick 2007; Lee 2004; Martinez 1996; Morrongiello 1998; Reed 2001; Richards 1991; Sun 2004; Twisk 2013; Wesner 2003; Wright 1995; Zirkle 2005), and 21 (78%) reported safety knowledge (Azeredo 2003; Campbell 2001; Carmel 1991; Collard 2010; Cook 2006; Falavigna 2012; Frederick 2000; Grant 1992; Greene 2002; Gresham 2001; Kendrick 2007; Lee 2004; Lu 2000; Martinez 1996; Morrongiello 1998; Richards 1991; Terzidis 2007; Wang 2009; Wesner 2003; Wright 1995; Zirkle 2005). One study reported health economic outcomes (Lu 2000).

Excluded studies

See Characteristics of excluded studies table.

Risk of bias in included studies

The risk of bias across all included studies is summarised in Figure 3 and Figure 4 and is included in the Characteristics of included studies table. Many of the risks of bias across all studies were unclear due to inadequate reporting. Major sources of bias within included studies were from inadequate blinding of both participants and personnel: in most studies, it was not possible to blind participants to allocation, and teachers delivering the intervention were responsible for administering and marking outcome assessments.

Figure 3

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

Figure 4

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

Allocation

Out of the 27 included studies, 13 were RCTs and 14 were non‐RCTs and CBAs. We only assessed random sequence generation and allocation concealment for the RCTs. Of the RCTs, we judged 11 (85%) to have an unclear risk of adequate random sequence generation (Campbell 2001; Carmel 1991; Chapman 2013; Collard 2010; Falavigna 2012; Gresham 2001; Lee 2004; Reed 2001; Sun 2004; Wang 2009; Zhao 2006), and two (15%) a high risk (Grant 1992; Kendrick 2007). For allocation concealment, we concluded that all 13 RCTs had an unclear risk of bias, mostly due to inadequate reporting.

For the 14 non‐RCTs, we judged that seven (50%) had a high risk of allocation bias (selection bias), mostly due to schools selecting themselves to be part of the study (Azeredo 2003; Buckley 2010; Cook 2006; Frederick 2000; Twisk 2013; Wesner 2003; Wright 1995), and seven (50%) had an unclear risk of allocation bias, again largely due to inadequate reporting (Greene 2002; Lu 2000; Martinez 1996; Morrongiello 1998; Richards 1991; Terzidis 2007; Zirkle 2005).

Blinding

We judged 18 (67%) studies to be at high risk of performance bias due to inadequate blinding of participants and personnel. Of these, eight were RCTs (Campbell 2001; Chapman 2013; Collard 2010; Falavigna 2012; Grant 1992; Kendrick 2007; Lee 2004; Wang 2009), and 10 were non‐RCTs (Azeredo 2003; Buckley 2010; Cook 2006; Frederick 2000; Martinez 1996; Morrongiello 1998; Richards 1991; Twisk 2013; Wesner 2003; Wright 1995). We judged that a further eight (30%) had an unclear risk, and of these four were randomised (Carmel 1991; Gresham 2001; Reed 2001; Sun 2004), and four were non‐randomised (Greene 2002; Lu 2000; Terzidis 2007; Zirkle 2005). We considered only one study (3%), an RCT, to be at low risk of bias (Zhao 2006).

Regarding blinding of those assessing study outcomes, we judged 12 (44%) to be at high risk of bias due to non‐blinding, 14 (52%) had an unclear risk of bias and only one (randomised) study (Campbell 2001) had a low risk of bias. For the high risk of bias, four were randomised (Grant 1992; Kendrick 2007; Lee 2004; Reed 2001), and eight were non‐randomised (Azeredo 2003; Buckley 2010; Cook 2006; Frederick 2000; Richards 1991; Twisk 2013; Wesner 2003; Zirkle 2005). For the studies at unclear risk of bias (again mostly due to non‐reporting) eight were randomised (Carmel 1991; Chapman 2013; Collard 2010; Falavigna 2012; Gresham 2001; Sun 2004; Wang 2009; Zhao 2006) and six were non‐randomised (Greene 2002; Lu 2000; Martinez 1996; Morrongiello 1998; Terzidis 2007; Wright 1995). We assessed all studies measuring self‐reported outcomes to be at high risk of bias.

Incomplete outcome data

The risk of bias due to incomplete outcome data varied across studies. We assessed 10 (36%) studies at unclear risk (three randomised (Chapman 2013; Gresham 2001; Sun 2004) and seven non‐randomised (Azeredo 2003; Cook 2006; Greene 2002; Lu 2000; Morrongiello 1998; Richards 1991; Zirkle 2005)), we judged eight (30%) at low risk (four randomised (Collard 2010; Kendrick 2007; Wang 2009; Zhao 2006) and four non‐randomised (Buckley 2010; Frederick 2000; Terzidis 2007; Twisk 2013)), and we assessed nine (33%) at high risk (six randomised (Campbell 2001; Carmel 1991; Falavigna 2012; Grant 1992; Lee 2004; Reed 2001) and three non‐randomised (Martinez 1996; Wesner 2003; Wright 1995)). Often, our assessment of unclear risk was due to the number of participants allocated at baseline not being clearly defined. We made judgements of high risk when less than 80% of the baseline sample provided outcome data at follow‐up (Wesner 2003; Wright 1995), when the studies lacked matched data (Lee 2004; Martinez 1996; Reed 2001), contained incomplete data returns from schools (Grant 1992), or when studies contained participants that had not received parental consent (Falavigna 2012).

Selective reporting

We considered the risk of selective outcome reporting to be unclear in 12 (44%) studies; four were randomised (Campbell 2001; Chapman 2013; Lee 2004; Zhao 2006) and eight were non‐randomised (Buckley 2010; Cook 2006; Frederick 2000; Greene 2002; Martinez 1996; Morrongiello 1998; Richards 1991; Zirkle 2005). We assessed the risk to be high in seven (26%) studies, of which five were randomised (Carmel 1991; Falavigna 2012; Grant 1992; Gresham 2001; Reed 2001) and two were non‐randomised (Azeredo 2003; Wright 1995). The risk of selective outcome reporting was low in eight studies (30%) (four randomised (Collard 2010; Kendrick 2007; Sun 2004; Wang 2009) and four non‐randomised (Lu 2000; Terzidis 2007; Twisk 2013; Wesner 2003)). The reasons for our 'high risk' assessment included incomplete reporting for some outcomes of interest (Azeredo 2003; Carmel 1991; Falavigna 2012; Grant 1992), no clear description of method of scoring for assessment tools used (Reed 2001), and only selected items being reported in tables (Wright 1995).

Other potential sources of bias

We judged five (18%) studies to be at high risk of other bias, for reasons including: failure to adjust for clustering effects (Buckley 2010 (non‐randomised); Lee 2004 (randomised)), misclassification bias (Grant 1992 (randomised)), lack of signed parental consent (Chapman 2013 (randomised)) and differences in test instrument used between intervention and control group (Wright 1995 (non‐randomised)). We judged 14 (52%) studies at unclear risk: five were randomised (Carmel 1991; Kendrick 2007; Reed 2001; Sun 2004; Wang 2009) and nine were non‐randomised (Azeredo 2003; Cook 2006; Frederick 2000; Greene 2002; Lu 2000; Martinez 1996; Morrongiello 1998; Richards 1991; Zirkle 2005). We judged eight (29%) to be at low risk of other bias, five being randomised (Campbell 2001; Collard 2010; Falavigna 2012; Gresham 2001; Zhao 2006) and three being non‐randomised (Terzidis 2007; Twisk 2013; Wesner 2003).

Risk of bias due to confounding (non‐randomised and controlled before‐and‐after studies)

We assessed four (29%) of the non‐RCT/CBA studies at high risk of bias due to confounding (Azeredo 2003; Lu 2000; Martinez 1996; Wright 1995). This was due to a lack of matching in the study design or adjustment in the analysis or statistically significant differences in characteristics between groups of participants. We judged six (43%) to have an unclear risk (Buckley 2010; Greene 2002; Morrongiello 1998; Richards 1991; Terzidis 2007; Twisk 2013) and four (29%) to be at low risk (Falavigna 2012; Frederick 2000; Wesner 2003; Zirkle 2005).

Effects of interventions

Self‐reported or medically attended unintentional injuries or injuries with an unspecified intent

Five (19%) studies reported non‐medically attended injury occurrence (Chapman 2013; Collard 2010; Lee 2004; Sun 2004; Wang 2009); one (4%) study reported medically attended injury occurrence (Zhao 2006), and one (4%) study reported both medically and non‐medically attended injuries (Lu 2000). Five studies reported injury incidence rates (Collard 2010; Lu 2000; Sun 2004; Wang 2009; Zhao 2006), and two reported injury incidence proportions (Chapman 2013; Lee 2004).

Lu 2000 was a clustered CBA, and the remaining six were cluster RCTs. The interventions included: a programme using cognitive behavioural change methods to modify risk‐taking behaviour in young people aged 13 to 14 years (Chapman 2013); an exercise‐based intervention to prevent sport injuries in children aged 10 to 12 years (Collard 2010); an intervention to prevent agricultural injuries in young farmers aged 13 to 18 years (Lee 2004); a teacher‐led health education programme with content designed by participating children to prevent a range of injuries in primary and middle school students in China (age range six to 16 years) (Lu 2000; Sun 2004; Zhao 2006); and a peer‐educator intervention for preventing a range of injuries in middle school children in China (Wang 2009).

Four studies, all conducted in China, were similar in terms of the interventions, adopting a health education approach for primary or middle school‐age children (Lu 2000; Sun 2004; Wang 2009; Zhao 2006, and three of these studies were included in a meta‐analysis of injury incidence rates (Lu 2000; Wang 2009; Zhao 2006; see Figure 5). We were unable to include Sun 2004 as this was a PhD thesis and only the abstract was available. We sought further information from the author but were unsuccessful in obtaining this. For the studies in the meta‐analysis, the participants (12,977 in total, with 2073 person‐years after adjusting for clustering) were aged from six to 16 years (Lu 2000), 12 to 15 years (Wang 2009), and seven to 13 years (Zhao 2006). The interventions comprised providing students with booklets and classes on injury prevention, letters to families encouraging co‐operation with safety education, a mass media campaign and training teachers to enhance safety in physical activity classes and at school drop‐off and pick‐up times (Lu 2000); weekly injury prevention sessions for students provided by peer educators which included activities, presentations, games or themed discussions (Wang 2009); and lectures and leaflets provided to teachers, parents and students, plus safety posters and a safety course provided to children before summer and winter school holidays (Zhao 2006). In two studies, the control arm received no intervention (Lu 2000; Wang 2009), and in one study, they received disease prevention education (pneumonia, iron‐deficiency anaemia, rickets and common communicable diseases) (Zhao 2006). The three studies had 2073 person‐years of follow‐up after adjusting for clustering. The pooled IRR was 0.73 (95% CI 0.49 to 1.08) and there was substantial heterogeneity between effect sizes (Chi² = 5.46, degrees of freedom (df) = 2, P = 0.07; I² = 63%) (Analysis 1.1). We assessed the quality of evidence as low, due to the high heterogeneity and wide CIs.

Figure 5

Forest plot of comparison: 1 Injury rates, outcome: 1.1 Injury rates at follow‐up.

We undertook sensitivity analysis and excluded the non‐RCT (Lu 2000). This had the effect of lowering the injury IRR in the intervention versus control groups (IRR 0.59, 95% CI 0.49 to 0.72) and there was no heterogeneity between effect sizes (Chi² = 0.97, df = 1, P = 0.33; I² = 0%). This implies that the non‐RCT design may explain the heterogeneity in the pooled analysis.

Sun 2004 reported a reduction in injury in the intervention schools that delivered an injury prevention campaign, including the distributing booklets and information leaflets, helping children with "blackboard bulletins" and offering posters on safety education to schools (adjusted risk ratio 0.65, 95% CI 0.57 to 0.76).

The remaining three studies that were not included in the meta‐analysis described self‐reported (non‐medically attended) injury. Chapman 2013 reported the proportion of students who had at least one transport injury in a three‐month period before and after the intervention for both the control and intervention groups. While the proportions of students who experienced at least one transport injury decreased by 6.0% from baseline in the intervention group and increased by 4.8% from baseline in the control group, the odds ratio (OR) showed no evidence of an effect (using intervention group as a reference, OR 1.42, 95% CI 0.62 to 3.17, P = 0.387). Collard 2010 investigated the I‐Play programme, a physical activity injury prevention scheme. They found weak evidence of a reduction in injury incidence (hazard ratio 0.81, 95% CI 0.41 to 1.59, adjusted for clustering) and this effect was stronger in children with lower initial levels of activity (hazard ratio 0.47, 95% CI 0.21 to 1.06). Lee 2004 described the impact of the Future Farmers of America (FFA) programme, comparing a standard intervention (promotional material, guides, newsletters and training for trainers) and enhanced intervention (with additional national trainer conventions and contact with a programme facilitator) with a control group. The control group received marketing and promotion of the Partners programme only (via newsletters, videos or conferences). In the standard intervention group, at three‐month follow‐up, 22% of children reported an injury in the past three months, as did 24% of children in the enhanced intervention group, and 24% in the control group. No P values were reported.

Observed safety skills

Two studies reported on observed safety skills for 1506 children after the intervention had been delivered (Kendrick 2007; Frederick 2000). While both studies found an improvement in the observed safety skills, we assessed the quality of evidence for this outcome to be low. This was due to the high risk of selection bias of included schools (as one of the two studies was a CBA), inconsistency of interventions and outcome measures, and imprecision as there were wide CIs in one of the two included studies and a paucity of data. However, the effect sizes were large. This rating of the evidence for this outcome was low quality means that our confidence in these effect estimates was limited.

Kendrick 2007 studied the Risk Watch programme targeting cycle and pedestrian injuries, falls, fire and burns, and poisonings. Children's safety skills were observed in scenarios including 'stop, drop and roll', road safety (including cycle helmet wearing) and poisoning secondary prevention. This study found some evidence of improvement in combined scores for all safety skills (mean difference 11.9, SD 1.4 to 22.5, P = 0.03) but only fire and burn prevention skills individually showed an improvement (OR 8.93, 95% CI 1.67 to 47.78, P = 0.01).

Frederick 2000 evaluated the impact of the IMPs programme. This was a multi‐component intervention based in the classroom and in the emergency department at local hospitals and aimed at preventing and minimising the impact of a range of injury mechanisms including road safety, accidents in the home, fire, electricity, poisoning and waterway injuries. They used scenarios to test skills practices and retention, and found a higher percentage of children in the intervention group compared to the control group exhibited correct basic life support skills at the five‐month post‐intervention test (e.g. assessment of danger: 36% in the intervention group versus 3% in the control group, P < 0.0005; assessment of responsiveness: 58% in the intervention group versus 12% in the control group, P < 0.0005; assessment of circulation: 7% in the intervention group versus 1% in the control group, P < 0.0005).

Behaviour

We divided this outcome into observed and self‐reported behaviour. All four studies (five articles) that reported observed safety behaviour showed improved practices, and 13 out of 19 studies describing self‐reported behaviour showed improved practices.

However, we assessed the quality of the evidence for this outcome to be very low, due to the high risk of selection bias of included schools, inconsistency in the methods of collecting data and intervention types, and wide CIs in the results presented. This means we have very little confidence in the effect estimates presented.

Observed behaviour

Four studies reported directly observed behaviour for 7022 children after the intervention had been delivered (Azeredo 2003; Reed 2001; Wright 1995; Zirkle 2005).

Azeredo 2003, studying school children aged five to 11 years and their families, introduced multiple interventions in 18 to 27 lessons, depending on the grade of child. The intervention included smoke‐alarm giveaways, school cycle fairs with helmet giveaways, safety pen‐pal letters, a letter to parents and injury prevention talks at parent‐teacher meetings. The control group had no intervention. Seat‐belt use was observed before, during and two weeks after the intervention. In the intervention group, observed seat‐belt use for passengers increased from 21% before the intervention, to 36% after completion. However, there was noted to be a 1% decrease in observed seat‐belt use within the control group. Among the intervention group, observed cycle helmet use increased from 0% to 10% (no P values given), although it is unclear to what extent this was affected by free helmet giveaways. There were no data given on the control group.

Reed 2001 implemented the AgDARE programme among high‐school agriculture students in the ninth and tenth grades in Kentucky, US. This incorporated two elements; narrative (cognitive) simulations, involving problem‐solving activities and a written component, and physical simulations whereby they could gain a better understanding of the difficulty of performing certain farm jobs with a disability. The control group received no intervention. After one year, researchers and agricultural teachers made a farm visit to 29 students who had received the intervention. Seventy‐six percent of students had made at least one positive safe work behaviour change, although no P values were presented. The control group participants were not visited.

Wright 1995 studied the implementation of the Think First programme aimed at reducing brain and spinal cord injuries, among school children aged between 11 and 15 years. The intervention programme was delivered by project staff and a person who had a brain/spinal cord injury. It included a film, lecture and talk by the injured person describing the traumatic injury. The programme covered areas such as seat‐belt use, use of motorbike and cycle helmets, avoiding drugs and alcohol when driving or participating in sport, and checking the depth of water when swimming or diving. The control group received the same programme after the study had finished. Behaviour was observed at one control school, one intervention middle school, and one intervention high school. Assessment took place before the intervention, two weeks and three months after the intervention for the intervention group, and two weeks before and one day before a Think First assembly for the control group. Statistical comparisons were within‐group only. Among middle‐school children in the intervention group there was very weak evidence of an increase in observed seat‐belt use at two weeks from 31% to 36% (P = 0.03), but decrease to 27% by three months (P < 0.05). There was no evidence of a change in cycle helmet use although the numbers of children observed were small. There was also no evidence of a change in the control group between the first and second observations of helmet or seat‐belt use.

Zirkle 2005 also reported on the Think First programme among primary school children. Behaviour was observed by parents, who noted a range of pedestrian, sport, interpersonal and car safety behaviours and the intervention was reported to have resulted in positive behaviour changes, although there were no data presented.

Self‐reported behaviour and practices

Nineteen studies assessed the impact of interventions on self‐reported behaviour and practices for an approximated 48,000 children following the intervention delivery (Azeredo 2003; Buckley 2010; Campbell 2001; Chapman 2013; Collard 2010; Falavigna 2012; Frederick 2000; Gresham 2001; Kendrick 2007; Lee 2004; Martinez 1996; Morrongiello 1998; Reed 2001; Richards 1991; Sun 2004; Twisk 2013; Wesner 2003; Wright 1995; Zirkle 2005).

Safety equipment wearing

Eight studies reporting on safety equipment wearing (Azeredo 2003; Falavigna 2012; Martinez 1996; Morrongiello 1998; Richards 1991; Wesner 2003; Wright 1995; Zirkle 2005).

Azeredo 2003, in a study incorporating a safety curriculum, smoke‐alarm giveaway, school cycle fairs, safety pen‐pal letters and school letters to parents, found that among the intervention group, self‐reported passenger seat‐belt use increased from 42% at baseline to 65% after the programme in children from grades kindergarten to grade one (P = 0.04). Self‐reported passenger seat‐belt use was not reported in the control group. Driver seat‐belt use increased by 2% in the intervention group, but there was no increase in the control group (no P values given). Among the children in the intervention group, there was some evidence of an increase in cycle helmet use (in children in kindergarten to grade one (P = 0.03) and children in grades two to five (P < 0.01)). The study did not mention the effects of the smoke‐alarm giveaway component of the programme.

Falavigna 2012 reported the Think Well programme, which is similar to the Think First For Kids programme. This involved a 60‐minute session provided by trained researchers, and comprised watching a video of a child describing their injury and its impact, and a brain and spinal cord trauma prevention programme. This study reported weak evidence of an increase in cycle helmet wearing from 8.6% at baseline to 14.5% at one‐week postintervention (P = 0.039) and 17.7% at five months postintervention (P = 0.034). The study reported some evidence of an increase in use of skateboard and rollerblading safety equipment one‐week postintervention (from 28% at baseline to 33.6% at one week, P = 0.460), and the effect was stronger at five months (from 33.6% at one‐week postintervention to 40.4% at five months, P = 0.037). There was no evidence of a decrease in use of skateboarding and rollerblading equipment in the control group (from 28.4% at baseline to 27.3% at one week, P = 0.992, and 22.6% at five months, P = 0.421). They did not report evidence of a change in the proportion of young people using seat belts or motorcycle helmets. However, there was some evidence that the proportion of young people in the intervention group who would sometimes ride in a car when someone had drunk alcohol increased after the intervention (36.9% before intervention to 42.1% after intervention, P = 0.046). No data were presented for the control group.

Martinez 1996 introduced a five‐component course into grades 10, 11 and 12 in a US high school. This consisted of audio‐visual aids, a physical demonstration and a didactic lecture. The sessions covered a basic lesson on types of energy, particularly kinetic energy; safety features of vehicles; occupant kinematics and forces/crash prevention; a demonstration of a rollover, followed by designing and testing of crash vehicles. The control group received their normal physics lesson. Self‐reported behaviour was collected at baseline, two weeks and six months after the end of the intervention. After adjusting for baseline measures, they found evidence of a difference between the control and intervention groups at six months in favour of the intervention for wearing a seat belt as a passenger (P < 0.001) but no evidence of an effect of wearing a seat belt as the driver (P = 0.63).

Morrongiello 1998 looked at several injury mechanisms, particularly focusing on head injuries caused by cycling, sport and vehicles. The intervention covered general safety education, including swimming, sun safety and stranger danger and was delivered by a teacher in the presence of a project co‐ordinator. It consisted of four activity stations. The control group received no intervention. Among the intervention group, there was a higher self‐reported use of helmets when cycling and rollerblading post‐intervention, compared to pre‐intervention (from 62% to 83% when cycling and from 45% to 65% when rollerblading, P < 0.05), but not in the use of a seat belt (from 87% to 95%, no P values given). Outcomes from the control group were not reported.

Richards 1991 implemented an injury prevention curriculum delivered over three months by a teacher. There was an eight‐component curriculum which included spinal cord awareness and water safety. For each component, teachers could choose from a variety of activities. It was not clear what the control group received. There was an increase in seat‐belt use for children in different school grades. Baseline levels of seat‐belt use in the intervention group ranged from 60% to 80%, and rose to between 72% and 86% post‐intervention. In the control group, baseline levels of seat‐belt use were between 74% and 84%, which decreased to between 60% and 82% at the second observation.

Wesner 2003 used the Think First programme, which involved the intervention group receiving a one‐hour session, incorporating a video of teenagers with a spinal cord injury, an educational session with audio‐visual aids, a description of brain anatomy and pathophysiology, and an account from a teenager with a spinal cord injury. This study reported (using within‐group analyses) an increase in self‐reported cycle helmet use among the intervention group from 32.6% before to 40.1% after the intervention (P = 0.05), but a decrease in the control group from 40.9% to 40.0% (no P values given); an increase in protective equipment wearing while rollerblading in the intervention group from 70.6% to 72.8% (P = 0.049) compared to from 66.5% to 68.4% in the control group (no P value given); and an increase in protective equipment wearing while skateboarding in the intervention group from 1.4% to 3.0% (P = 0.01) compared to from 4.8% to 7.4% in the control group (no P value given).

Wright 1995, in an analysis of the Think First programme, found little impact on self‐reported behaviours, with no evidence of an increase in seat‐belt wearing or motorcycle helmet wearing, even though baseline levels of motorcycle helmet wearing were lower than in the Falavigna study at 60% to 80% (Falavigna 2012).

Zirkle 2005 described an increase in self‐reported safety behaviours following the Think First For Kids programme but no data were provided (this was a PhD thesis and only the summary was available).

Risk behaviour relating to the road

Four studies reported self‐reported risky behaviour relating to the road (Chapman 2013; Gresham 2001; Martinez 1996; Twisk 2013).

Chapman 2013 studied the impact of the SPIY programme aimed at reducing transport injuries among 13‐ to 14‐year‐old school children in Australia. The intervention group was presented with risk‐taking injury scenarios, incorporating activities such as role play and discussion. These used cognitive behavioural change principles. The control group received no intervention. The outcome of interest for this review was self‐reported transport risk behaviour. Comparing the control with the intervention (reference) group there was no evidence of a difference in the post‐test reporting of transport risk behaviour (OR 1.37, 95% CI 0.64 to 2.93, P = 0.415).

In the Think First For Kids programme, Gresham 2001 found that both the intervention and control groups showed an improvement in risk behaviour scores (within‐group before‐and‐after comparison), with scores increasing significantly more in the intervention than control groups in all grades. This varied between the different grades. In grade one, there was a score increase of 3.06 (95% CI 2.76 to 3.35, P < 0.01) in the intervention group compared to 1.70 (95% CI 1.35 to 2.06, P < 0.01) in the control group. In grade two, this was 2.80 (95% CI 2.46 to 3.14, P < 0.01) in the intervention group versus 1.10 (95% CI 0.74 to 1.46, P < 0.01) in the control group and in grade three it was 3.27 (95% CI 2.92 to 3.63, P < 0.01) in the intervention group versus 1.55 (95% CI 1.17 to 1.93, P < 0.01) in the control group.

Martinez 1996, incorporating road safety into a physics curriculum, also assessed driving‐related risky behaviour (e.g. speeding and drink driving) with data collected at baseline, two weeks and six months after the end of the intervention. After adjusting for baseline measures, they found some evidence of a difference between the control and intervention groups at six months in favour of the intervention for speeding (P < 0.001) but no evidence of a difference for drink driving (P = 0.7).

Twisk 2013 used a lorry at primary schools to give pedestrian and cyclist safety instruction. Students could visualise the driver's field of view and were given information on safe road behaviour. There were two intervention groups, one targeting awareness of blind spot programmes (addressing carelessness) and the other targeting competency (addressing blind spot hazards). The control group received no intervention. Compared to the control group, there was no evidence of a change in the self‐reported correct positioning of cycle or self as a pedestrian, in either intervention group. For the awareness group, mean scores (for correct positioning) increased from a baseline of 9.5 (SD 1.4) to 9.7 (SD 1.1) in the intervention group compared to a decrease in mean scores from 9.6 (SD 1.2) at baseline to 9.3 (SD 1.5) in the control group (analysis of covariance (ANCOVA) P = 0.84). For the competency group, the mean score increased from 8.2 (SD 2.5) to 9.5 (SD 1.8) in the intervention group compared to a decrease in the mean score from 8.7 (SD 1.4) to 8.5 (SD 1.8) in the control group (ANCOVA P = 0.30).

Agriculture‐related injuries

Two studies (reported in three articles) described self‐reported behaviour related to agriculture‐related injury prevention (Lee 2004; Reed 2001).

Lee 2004 described the impact of the FFA Partners programme, with a standard (promotional material, guides, newsletters and training for trainers) and enhanced intervention (with additional national trainer conventions and contact with a programme facilitator). The control group received marketing and promotion of the Partners programme only (via newsletters, videos or conferences). This study found no evidence of a difference between the intervention and control groups for self‐reported safety consciousness (standard post‐test mean score 2.9 (SD 0.72), enhanced post‐test mean score 3.0 (SD 0.69), control post‐test mean score 3.0 (SD 0.77); P = 0.47)), or dangerous risk taking (standard post‐test mean score 3.2 (SD 0.73), enhanced post‐test mean score 3.2 (SD 0.72), control post‐test mean score 3.2 (SD 0.78); P = 0.38).

Reed 2001 implemented the AgDARE programme among high‐school agriculture students in the ninth and tenth grades in Kentucky. This incorporated two elements; narrative (cognitive) simulations, involving problem‐solving activities and a written component, and physical simulations, whereby they could gain a better understanding of the difficulty of performing certain farm jobs with a disability. This study used a 10‐item Stages of Change instrument to assess the agricultural students' transitions from contemplation to action to make a positive work behaviour change. They found that there was some evidence of a higher mean Stages of Change score in the intervention compared to the control groups (mean: 31.1 (SE 0.6) in the intervention group, 21.4 (SE 0.5) in the control group, F statistic (df) 134.5 (2;604) P < 0.001). A higher Stages of Change score indicates that those participants were more likely to make a behaviour change.

Sports injuries

Collard 2010, in a study from the Netherlands, looked at the impact of the I‐PLAY programme on self‐reported safety behaviours among 10‐ to 12‐year‐old children. This involved two physical exercise (PE) lessons per week, over eight months, delivered by a teacher with monthly newsletters and access to online information. The control group received their normal PE classes. There was no evidence of a difference between the intervention and control group according to self‐reported safety behaviours (measured on a Likert scale). The differences between means were 0.05 (95% CI ‐0.04 to 0.14) for wearing protective equipment during organised sport, 0.01 (95% CI ‐0.21 to 0.19) for wearing protective equipment during leisure activities and 0.07 (95% CI ‐0.13 to 0.27) for wearing appropriate footwear during PE.

Multiple injury types

Five studies, reporting on interventions aimed at preventing multiple injury types, included self‐reported behaviour and practices (Buckley 2010; Campbell 2001; Frederick 2000; Kendrick 2007; Sun 2004).

Buckley 2010 reported on the SPIY programme aimed at 13‐ to 14‐year‐old children in a deprived urban area in Australia. The intervention schools received teacher training, a teacher's manual and a student workbook for eight sessions. These were 50 minutes in duration and included presentations about risk‐taking and injury scenarios, introduction to first aid and cognitive behavioural activities to prevent the risk‐taking behaviour, including protecting friends. The control schools continued with their normal curriculum, but could use the SPIY programme after the end of the study. The intervention group had a greater positive change in their self‐reported risk‐taking score compared to the control group (from 2.39 to 1.93 with intervention versus 1.67 to 1.85 with control, P = 0.014).

Campbell 2001 reported on the impact of a first aid and home safety programme among 11‐ to 18‐year‐old children with a Hispanic background and their parents. The intervention focused on prevention of injuries and responding to emergency situations and included household safety, emergency care, controlling bleeding and treating burns, presented in eight sessions. The control group received an alcohol and tobacco prevention programme. At one‐year follow‐up both the intervention and control groups reported making home safety behaviour changes. For example, 81% of all those surveyed reported removing small objects, 90% reported keeping chemicals and 95% reported keeping medicines out of reach, and 43% had purchased fire extinguishers. The only evidence of difference between intervention and control groups was in the percentage of families who had practised a fire escape plan (47% in the intervention group, 31% in the control group, P < 0.01).

To assess self‐reported behaviour following the IMPs programme, Frederick 2000 used a video showing dangerous scenarios to provide students with the opportunity to describe what they would do in those situations. There appeared to be some weak evidence of a difference between the intervention and control groups in the percentages of children reporting a range of safe practices. These included, for example, situations such as not playing near water (25% in the intervention group versus 17.8% in the control group, P < 0.01) and identifying dangers for electrocution risk (18.1% in the intervention group versus 7.3% in the control group, P < 0.01).

For the Risk Watch programme, Kendrick 2007 found a difference in the adjusted OR between the intervention and control groups in the number of children self‐reporting using matches (83.0% with intervention versus 74.7% with control, adjusted OR 1.84, 95% CI 1.06 to 3.20, P = 0.031). However, on other measures (e.g. having smoke‐alarms, cooking food, getting medicine without an adult present and road safety), there was no evidence of a difference between groups.

Sun 2004 studied the impact of a safety intervention among over 7000 Chinese school pupils. This intervention included distributing booklets and information leaflets, helping children with "blackboard bulletins" and offering posters on safety education to schools. The control group received general information on food hygiene and disease prevention. In the questionnaire follow‐up at one year, there was reported to be a greater decrease in the score of risk‐taking behaviour for the intervention group than the control group (no P values given).

Safety knowledge

Twenty‐one studies reported on changes in safety knowledge for 46,550 children following the intervention (Azeredo 2003; Campbell 2001; Carmel 1991; Collard 2010; Cook 2006; Falavigna 2012; Frederick 2000; Grant 1992; Greene 2002; Gresham 2001; Kendrick 2007; Lee 2004; Lu 2000; Martinez 1996; Morrongiello 1998; Richards 1991; Terzidis 2007; Wang 2009; Wesner 2003; Wright 1995; Zirkle 2005). Of these, 19 reported an improvement in at least one question domain in the intervention compared to the control group.

However, we assessed the quality of this evidence as very low, meaning that we have little confidence these results. This is because 11 studies were CBA or non‐RCT design with a high risk of selection bias of included schools. There was also considerable inconsistency in knowledge tested, questionnaire designs and methods of collecting data, and a wide range of intervention types and pedagogical approaches, which may have contributed to the wide range of effect sizes. In addition, results were presented in a way that often made precision difficult to compare.

Vehicle and road safety knowledge

Four studies reporting on road and vehicle safety included safety knowledge as an outcome (Cook 2006; Martinez 1996; Wright 1995; Zirkle 2005).

Cook 2006 reported the effect on class safety knowledge when an injured classmate participated in safety education sessions. In the intervention group, the child gave a presentation and then interacted with his or her classmates. A nurse gave a presentation on injury occurrence and prevention, which involved discussions, short videos, written materials and a workbook. The first control group received the same information and the injured child's scenario, but without an actual injured child (enhanced control), while the second control group only received some anatomy education (basic control). Six classrooms were in the intervention group, and six in each of the control groups. There was improvement between the pre‐ and post‐test scores in both the intervention and enhanced control group, with five out of six classrooms in each group showing improvement (P < 0.05). At one‐month post‐intervention, 100% of the intervention classes maintained significantly higher test scores (P < 0.05) compared to 66% of the enhanced control group classes.

Martinez 1996 used a seven‐item questionnaire, which included questions about the physics of crashes, demographics of people involved in crashes and characteristics of automobiles. Compared to the control group, the intervention group showed higher safety knowledge scores two weeks' post‐intervention (0.77 with intervention versus 0.94 with control, P < 0.001), and six months' post‐intervention (0.75 with intervention versus 0.89 with control, P < 0.001).

Wright 1995 used a 22‐item questionnaire for the intervention schools, and a five‐item questionnaire for the control schools. For the intervention schools, these were administered before the intervention, at two weeks' post‐intervention, and three months' post‐intervention and for control schools two weeks before, and one day before the assembly where they were given the same curriculum information as the intervention group. From baseline to three months' post‐intervention there was a significant increase in safety knowledge in the middle school (from 15.0/22 to 16.2/22, P < 0.05), but a decrease in scores in the high school (from 16.1/22 to 14.6/22, P < 0.05). There were no significant changes in knowledge scores in the control school from two weeks before to one day before (from 4.6/5 to 4.6/5).

Zirkle 2005 implemented the Think First programme in five control and five intervention schools. There was very limited information on the actual content of the curriculum in the article, but similar Think First programmes include activities or teaching aimed at reducing brain and spinal cord injuries. This study reported a significantly greater improvement in safety knowledge in the intervention versus the control group, although they provided no data or P values.

Multiple injury safety knowledge

Twelve studies involving multiple injury mechanisms reported on safety knowledge (Azeredo 2003; Campbell 2001; Falavigna 2012; Frederick 2000; Greene 2002; Gresham 2001; Kendrick 2007; Lu 2000; Morrongiello 1998; Richards 1991; Wang 2009; Wesner 2003).

Azeredo 2003 used two different instruments to measure children's safety knowledge (a four‐item activity and written questions for children in kindergarten to grade one and a 20‐item multiple choice questionnaire for children in grades two to five). These included questions on emergency telephone numbers, good safety habits, intersection/stop signs, train warnings and good swimming habits for the younger children, and how to handle emergencies, pedestrian rules, cycle rules, water safety rules and home fire safety for the older children. The study showed an increase between the baseline and post‐test scores in the intervention schools (P values all < 0.01). In both age groups, there also was a difference in the post‐test scores between the intervention and control schools (P values between < 0.01 and 0.04).

Campbell 2001 reported that a higher proportion of students in the intervention group could correctly identify more items in a first aid kit than students in the control group. The baseline number of items correctly identified in the control group was 3.8, rising to 4.1 immediately post‐intervention, and 4.3 at the one‐year follow‐up. The baseline number of items correctly identified in the intervention group was slightly lower than the control at 3.7. This rose to 4.6 post‐intervention and remained at 4.6 at the one‐year follow‐up (P < 0.001 immediately post‐intervention and P < 0.01 at one‐year follow‐up).

Among intervention group children, Falavigna 2012 found no evidence of a difference in the effects of traumatic brain injury, occipital lobe function and seat‐belt use knowledge scores between the control and intervention groups at baseline, but there was an improvement in traumatic brain injury (P < 0.001) and occipital lobe function (P < 0001) immediately post‐intervention and at five months in the intervention group. Knowledge about seat‐belt use was similar in both the intervention and control groups at baseline and did not change during the intervention (no P values provided).

Frederick 2000 assessed the impact of the IMPS programme on students' knowledge by a quiz, administered before and five months after the intervention. There was an increase in safety knowledge scores from baseline in both groups (P < 0.01), although the intervention group demonstrated greater knowledge than controls in calling 999 (UK's emergency telephone number), first aid for burns and for choking. Data comparing scores for intervention and control groups were not presented.

Greene 2002 reported on Phase III of the Think First For Kids curriculum. This was a six‐week, six‐subject curriculum which was integrated into the usual school curriculum. The units covered the structure and function of the brain and spinal cord, road traffic safety, conflict resolution, and water, sports, playground and recreational safety. The control group received no intervention. In the post‐intervention questionnaire, administered one week after the six‐week teaching period, all school grades in the intervention group had higher safety knowledge scores related to brain and spinal cord injury than the controls (between the grades, P values ranged from 0.0001 to 0.0037) and there was an indication of increased water safety knowledge (P values between 0.0001 and 0.0489). In cycle safety questions, the intervention group had higher scores than the control groups in two grades (grade three, P = 0.02; grade one, P = 0.07; grade two not significant (no P value given)). Regarding the motor vehicle/pedestrian safety questions, there was no evidence of a difference between the intervention and control groups in grade one. However, among children in grades two and three, the scores for the intervention group were higher than those of the control group (P = 0.0143 for grade two and 0.0134 for grade three). There was no evidence of a difference in safety knowledge regarding playground and sports safety between the intervention and control groups.

Gresham 2001, also using the Think First For Kids programme, reported that the intervention group had a larger increase in their knowledge score than the control group (P < 0.001). In the intervention group, there was an increase from pre‐test scores of between 19% and 23%, measured by individual grades of children. Scores for the control group were not reported.

In the Risk Watch programme, Kendrick 2007 used an age‐appropriate questionnaire to assess knowledge, with questions illustrated pictorially. In the post‐test questionnaire, compared to the control group, the intervention group children answered more fire and burn protection questions correctly (difference between means 7.0%, 95% CI 1.5% to 12.6%, P = 0.01) but there was no evidence of a difference in safety knowledge relating to poisoning prevention (P = 0.57), cycle and pedestrian safety (P = 0.66) or falls prevention (OR 0.48, 95% CI 0.21 to 1.10, P = 0.08). Analysis of specific questions indicated that intervention group children may have been more aware of action to take if clothes caught fire (difference between means 35.3%, 95% CI 22.7% to 47.9%, no P values given), and were more likely to know the correct way to wear a cycle helmet (difference between mean 6.3%, 95% CI 1.4% to 11.1%, no P values given).

Lu 2000 reported on an intervention incorporating classes, letters home and family involvement. There was also teacher training and involvement in supervision of potentially risky behaviour, and a reporting system. This study found a higher post‐test score in students' safety knowledge in the intervention group compared to the control group (Chi² = 56.63, P < 0.001). The study did not report on the content of the questionnaire, but stated that it comprised 10 questions.

Morrongiello 1998 reported the impact on safety knowledge of the activity‐based sessions in the intervention group. These sessions focused on four safety topics; bicycling and road use, reducing sports injury, creating a safety banner, and vehicle and road safety. Postintervention, Morrongiello found that the intervention group correctly answered 89% of the questions on knowledge of safety facts compared to 55% in the control group. The intervention group also correctly answered 84% of questions about which safety equipment should be worn for which sport, compared to 64% in the control group (no P values given).

Richards 1991 introduced an intervention which was tailored to the different grades of children. The intervention group received an eight‐component curriculum, which included topics such as spinal cord awareness and water safety. The authors did not report whether the control group received an intervention. The study found some evidence of a difference between the intervention group and control group in the post‐test questionnaire (P values ranged from P < 0.0001 and P < 0.05, between different school grades).

Wang 2009 measured safety knowledge with an eight‐point questionnaire, including topics such as the term unintentional injury itself, sports, falls, traffic and burns. The intervention group answered a higher number of questions correctly compared to children in the control group (82.7% in the intervention group versus 57.2% in the control group, P < 0.05).

Wesner 2003 asked a set of 13 questions to test students' knowledge and found an increase in the intervention group compared to baseline in knowledge of motor vehicle accidents being the most common cause of brain and spinal cord injury (from 37.7% to 61.0%, P < 0.001), the severe and permanent nature of brain and spinal cord injuries (from 60.3% to 69.4%, P = 0.022) and that injuries are most common in 15‐ to 24‐year‐old people (from 42.9% to 81.4%, P < 0.001). There was no evidence of a change in their scores relating to the laws about the use of helmets on motorbikes (from 74.9% to 74.5%, P = 0.71) and cycles (from 64.9% to 62.7%, P = 0.54), or that brain injury affects walking, talking and thinking (from 88.3% to 93.2%, P = 0.75). In the control group, there was no evidence of a change from baseline in scores on any of the questions.

Burn safety knowledge

Two studies reported on burns safety knowledge (Carmel 1991; Grant 1992).

Carmel 1991 used multiple methods of teaching to deliver a burns prevention programme, including a presentation, a home checklist, a set of photographs and a colouring book. The aim of this programme was to raise awareness, increase knowledge, and change attitudes and behaviour related to burn prevention. Compared to the control group, who received no intervention, the study authors reported that the intervention group could answer a higher number of safety questions correctly, both immediately at the end of the programme (mean values: 84.74 (SD 12.65) in the intervention group versus 69.85 (SD 14.18) in the control group, P < 0.001), and 10 weeks after the end of the programme (84.46 (SD 12.31) in the intervention group versus 71.73 (SD 13.33) in the control group, P < 0.001).

However, this was not found by Grant 1992, who introduced the "Learn Not To Burn" curriculum to children in grades three and four in US primary schools. This covered 22 key behaviours for burn prevention. The control groups either used "other methods of fire safety education" or "no established fire safety education". There was no evidence of a difference in percentage scores between the intervention and control group (from 81.37 to 90.75 in the intervention group versus from 81.10 to 90.58 in the control group, P = 0.96).

Sports safety knowledge

Collard 2010 investigated the I‐PLAY programme to prevent sport‐related injuries by introducing regular exercises. They used a nine‐item questionnaire to assess safety knowledge and found weak evidence of a difference, with the intervention group scoring higher than the control groups at follow‐up (mean difference 0.49, 95% CI 0.20 to 0.78).

Agricultural safety knowledge

Lee 2004 implemented a school‐based educational programme to reduce unintentional injury rates among youths involved in agricultural work. The standard intervention involved working with big agricultural businesses, marketing and promotion of safety information aimed at reducing agriculture‐related injuries, and working with trainers. For the enhanced intervention group, there was more support for trainers, with regular conferences and access to the public health office. The study did not show evidence of a difference in mean safety knowledge scores between the groups (mean score 2.7 (SD 0.75) in the standard group versus 2.8 (SD 0.88) in the enhanced group versus 2.7 (SD 0.82) in the control group, P = 0.43).

Water safety knowledge

Terzidis 2007 studied prevention of water‐related injuries. The intervention group attended a day event, which consisted of a short audio‐visual presentation, a discussion about pupils' personal experiences, comments on how relevant events could have been averted and drama plays. There were also take‐home materials. The control group received no intervention. Among children in kindergarten and grade one, children in the intervention groups showed higher water safety knowledge postintervention compared to children in the control group (mean difference 17.40%, 95% CI 6.41% to 28.39%, P = 0.005). However, among children in higher grades in elementary school and in high school there was no evidence of a difference (P = 0.08 in elementary school and P = 0.92 in high school).

The quality of the evidence relating to both observed and self‐reported safety behaviours and practices was very low. This was due to studies being highly heterogeneous in their methods and with very poor reporting of methodologies, and many studies with high or unclear risk of bias across domains.

Health economic outcomes

Only one study included an economic analysis (Lu 2000). This was a multi‐component intervention involving classes, a booklet on injury prevention, letter to families, a mass media campaign, training of teachers in being alert to unsafe behaviour and a school injury reporting system. The overall cost of the programme was 9000RMB (Chinese Yuan). Before the intervention, there were 2016 cases of injuries, representing 12,418 lost school days. After the intervention, there were 708 lost school days (a decrease of 1162 from baseline) though this was not described in terms of cost‐effectiveness. The cost of treating injuries reduced from 145,152RMB before the intervention to 11,019RMB after the intervention, resulting in a saving of 134,136RMB. This was a 92.4% decrease from baseline and resulted in a net benefit of 125,136RMB after subtracting programme costs, giving a cost:benefit ratio of 1:13.90, that is, for every 1RMB spent, 13.90RMB was saved.

We assessed the evidence for this outcome to be very low, due to its limited applicability and high or unclear risk of bias across multiple domains.

Discussion

Summary of main results

The review includes 30 articles from 27 studies. Seven studies assessed injury occurrence, of which three were included in a meta‐analysis. This showed no strong evidence of a lower incidence rate in the intervention than control groups, and there was substantial heterogeneity (summary of findings Table for the main comparison). We rated this evidence as low quality because of this heterogeneity (inconsistency) and imprecision in the results. However, when we performed a sensitivity analysis and removed the non‐RCT from the analysis, there was stronger evidence of an effect and no heterogeneity, suggesting study design may explain the heterogeneity that occurred in the meta‐analysis. There was insufficient evidence to determine whether school‐based educational programmes can prevent unintentional injuries.

We were unable to pool data for our secondary outcomes, so our conclusions were limited, as they were drawn from highly diverse single studies and the body of evidence was rated as low (safety skills) or very low (behaviour, safety knowledge, health economic outcomes).

Two studies reported observed safety skills and both showed evidence of an improvement in the intervention group. However, again our confidence in this effect was limited as the evidence was of low quality (serious risk of selection bias, imprecision and inconsistency, but a large effect size countering this to some extent).

All four studies reporting observed safety behaviours and 13 of 19 studies describing self‐reported behaviour presented very low quality evidence in favour of the intervention. There was very low evidence in favour of an improvement in safety knowledge in the intervention group versus the control, with 19 of the 21 studies reporting improvements in all or some of the safety knowledge scores in the intervention groups. Studies which showed stronger evidence of an increase in knowledge comprised broadly similar elements as those which did not, so it was not possible to determine which elements might be most effective. The duration of the intervention did not appear to be associated with effectiveness.

Studies that examined self‐reported behaviour showed greater variation in findings, with over two‐thirds (13 out of 19, 68%) reporting an improvement and the remainder showing no evidence of a difference. The reason for this heterogeneity was not clear, as most incorporated similar elements (interactive sessions and audio‐visual equipment) into their curricula. Only one study reported economic outcomes and estimated a positive return on investment (very low quality evidence).

More high‐quality studies are needed to evaluate the impact of educational programmes on injury occurrence.

Overall completeness and applicability of evidence

We searched multiple large databases, in addition to handsearching, to obtain as many relevant full‐text reports as possible. We also included full‐text reports and abstracts written in languages other than English, and placed no time restrictions on the searches.

The review included studies from several high‐ and middle‐income countries, with rural and urban populations and children aged four to 18 years. This spread of populations and countries is likely to improve the generalisability of the review to similar populations, although there were only five studies from low‐ or middle‐income countries (China, Israel and Brazil). While some studies covered a broad range of mechanisms, others focused on more specific mechanisms or injury types (e.g. burns, pedestrian or agricultural injuries), thus potentially limiting their generalisability to other injury mechanisms.

Seven of the 27 studies reported the primary outcome. Of these, only three studies were similar enough to allow inclusion in the meta‐analysis. All three were based in China and so it is unclear how generalisable the results were to other countries and school systems. There was a wide range of injury mechanisms studied and types of interventions aimed at improving outcomes. These included a range of approaches (e.g. the provision of safety equipment, inclusion of families with letters and homework and in‐school lessons), often with some student involvement (peer teaching) and over differing times and intensities of one session to multiple sessions over one year. Only seven studies in our review referred to having used behavioural change or learning theories (or both) to develop the intervention they evaluated and how they were applied was often not described in detail. Specific examples included using the theory of planned behaviour which was used to develop the SPIY programme (Buckley 2010), social learning theory which was used to develop the i‐Play programme (Collard 2010), and self‐determination theory which capitalised on the 'teachable moment' after an injury to develop a post‐trauma programme (Cook 2006). Other authors referred more generally to having used applied behaviour or social theories, without describing their application in detail (e.g. Azeredo 2003; Greene 2002; Gresham 2001; Reed 2001).

Similarly, very few studies described the educational components of their interventions (e.g. learning theories, learning objectives, teaching methods, techniques or communication vehicles) in sufficient detail to enable us to assess how these impacted on our outcomes. While some studies did report using different pedagogical approaches for different age children, again there was insufficient detail reported to be able to define what worked best for children at different developmental stages.

Quality of the evidence

The quality of evidence was low for our primary outcome and low or very low for the secondary outcomes, indicating that further research is very likely to have an important impact on our confidence in the estimate of the effect. There are several reasons why the quality of the evidence was rated in this way.

Risk of bias has a serious impact on the overall quality of studies. For many of the included studies, it was difficult to assess the risk of bias due to inadequate reporting. Consequently, we judged a large number of studies to have an unclear risk of each source of bias. The majority of relevant studies had an unclear risk of random sequence generation and of allocation bias, with the remaining studies having high risk. The risk of performance and detection bias was high in most studies. The risk of attrition bias and reporting bias was split quite evenly between low, unclear and high risk. This was also true of confounding bias in the relevant studies. Most studies were at unclear risk of other bias, with the remaining studies split evenly between high and low risk. When only RCTs were examined for risk of bias, the findings were very similar to when we included all study designs. Common limitations in study design and reporting included: reporting only the number of participants for whom outcomes were measured rather than defining the selected groups of participants to be followed up, failing to include adequate detail on randomisation and failing to adequately take account of clustering, not undertaking intention‐to‐treat analyses and failing to present participant flow charts. Major sources of bias across most studies arose from an inability to blind participants to their allocation due to the nature of the interventions and a lack of blinding during outcome ascertainment.

Inconsistency was also a major factor in our lack of confidence in these study findings. The quality of the evidence for our primary outcome of medically or non‐medically attended injuries was downgraded for inconsistency, in part due to substantial statistical heterogeneity (I² = 63%). When we performed sensitivity analysis and restricted the meta‐analysis to only RCTs, this had a substantial impact on the results, with the IRR reducing (from 0.73 (95% CI 0.49 to 1.08) to 0.59 (95% CI 0.49 to 0.72)) and the heterogeneity disappearing (I² = 0%). This suggests that the non‐random study design of Lu 2000 may have contributed substantially to the heterogeneity we observe in the pooled analysis.

However, this may also be due to other differences between studies; for example, the intervention in the non‐RCT included a mass media campaign which the two RCTs did not, and which could have resulted in contamination between treatment groups (Lu 2000). Outcomes were also measured using different data collection methods, with varying potential for detection bias in the three studies: Lu 2000 collected injury data via weekly student injury events being entered into a central reporting system; Wang 2009 collected injury data via student self‐completed questionnaires administered at baseline and follow‐up and Zhao 2006 used a school injury surveillance system based on hospital and insurance company reports of injuries. In addition, it was unclear when injury data were collected in relation to the end of the intervention and difference in this may have contributed to heterogeneity.

For each of the other outcomes, there was substantial heterogeneity in terms of the size and in some cases, direction of the effect. This may be explained by variation in terms of what was measured and how. For example, there were 21 studies that assessed safety knowledge and 19 of these reported an improvement in at least one knowledge domain. However, the knowledge tested was extremely variable, as were the methods of data collection and instruments used.

Imprecision was also an issue for some of our outcomes, particularly safety skills, where we considered it to be serious due to the wide CIs in one of the two included studies and the paucity of data.

The quality of the evidence was not downgraded due to directness or generalisability, as most studies included all students in the school or class, and schools were mostly state schools which are likely to have children of a broad range of backgrounds, abilities and affluence.

Potential biases in the review process

We undertook a comprehensive search that included 28 bibliographic databases and websites. Although the database searches were conducted in English, LILACS includes studies from Latin American countries, and no language restrictions were placed on the search results. Several potential studies screened were translated from other languages including Russian, French, German and Chinese by native speakers prior to assessment for inclusion. We searched conference abstracts and the grey literature for unpublished studies. We were unable to assess publication bias using a funnel plot as only three studies were included in the meta‐analysis. It is possible that our searches failed to find some studies eligible to be included in our review.

Agreements and disagreements with other studies or reviews

There has been one previous review of school‐based injury prevention programmes (Mulvaney 2012), and several Cochrane and non‐Cochrane reviews of school‐based interventions aimed at changing safety behaviours relating to single injury mechanisms (Duperrex 2009; Mytton 2006; Owen 2011).

Mulvaney 2012 described safety education for a range of injury mechanisms but only one of the included studies in the review (Collard 2010) fulfilled the criteria for inclusion in our review. This was because the other school‐based studies in Mulvaney 2012 either did not have an appropriate control, or were focused on a single injury mechanism. They found no papers reporting the impact of safety education on injury rates but, consistent with our review, found a positive impact on knowledge, behaviour and skills although knowledge did not always translate into behaviour change. The authors highlighted the importance of involving children, families and communities in formulating injury prevention programmes, as well as using a multi‐disciplinary approach to deliver them.

Owen 2011 undertook a Cochrane Review of non‐legislative interventions to increase uptake of cycle helmet use. Although community‐based programmes were most effective in increasing helmet use, the review also reported evidence of an improvement among school‐based interventions (8 studies; OR 1.73, CI 95% 1.03 to 2.91). Free helmet giveaways were most effective in increasing helmet use, while programmes with subsidised helmets or education only had limited impact (3 studies; OR 1.43, 95% CI 1.09 to 1.88). This was echoed in a systematic review by Nauta 2014 of both community‐ and school‐based programmes, which also found that free safety equipment (e.g. cycle helmets) was most effective at increasing use of safety devices. Only one study in our review reported on the effect of free helmet giveaways, but did not find evidence of an impact of that intervention on observed behaviour (Azeredo 2003).

One Cochrane Review of school‐based education initiatives to prevent dog bites by Duperrex 2009 was not able to assess the primary outcome of reduced bites because of a lack of studies reporting this outcome. However, they did report changes in behaviour and found that education involving a 30‐minute session and letters to parents could improve children's knowledge, attitudes and behaviour when around dogs.

Mytton 2006 conducted a systematic review of school‐based initiatives aimed at children who were at risk of aggressive behaviour. The review found that elements (e.g. teaching relationship skills, as well as to a lesser extent, non‐response to provocative situations) could reduce aggressive behaviour in both primary‐ and secondary‐aged children although again, there was no clear evidence that this reduced violence‐related injury. This suggests that school‐based interventions can have a positive impact on behaviour, and mirrors the findings in some of the studies in our review.

There are also systematic reviews of school‐based interventions to reduce pupils' risky behaviour. Thomas 2006, in a Cochrane Review of school‐based programmes for preventing smoking, highlighted the importance of incorporating elements focused on social influences and social competence when designing a programme. This was also a finding by Faggiano 2005, who reported that programmes combining social influences and social competence were most effective at preventing marijuana use. In contrast, one Cochrane Review by Foxcroft 2011 found no conclusive evidence that school‐based interventions to prevent alcohol misuse were effective. In that review, in agreement with our review, some studies showed a positive effect in the intervention group, and others showed no effect.

Our review was unable to identify how useful surrogate measures, such as safety knowledge, were in predicting the impact of the intervention on the primary income of injury rates.

Finally, though not specifically assessing the impact on injury prevention outcomes, the results of one Cochrane Review of the World Health Organization's Health Promoting School framework provided evidence for the effectiveness this approach for some health behaviour interventions but not others, further supporting the school as a suitable setting for health improvement (Langford 2014).

Figure 1

Logic model, based on Langford 2014

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 2

PRISMA flow chart detailing the process of study selection for all studies included in the review. CBA: controlled before‐and‐after study; RCT: randomised controlled trial.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 3

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

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 4

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

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 5

Forest plot of comparison: 1 Injury rates, outcome: 1.1 Injury rates at follow‐up.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Analysis 1.1

Comparison 1 Medically attended or non‐medically attended injury rates, Outcome 1 Injury rates at follow‐up, adjusted for baseline injury rates in non‐randomised studies.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Summary of findings for the main comparison. School injury prevention programmes compared to controls for the prevention of unintentional injuries in children and young people

School injury prevention programmes compared to controls for the prevention of unintentional injuries in children and young people
Patient or population: children and young people Setting: schools Intervention: school injury prevention programmes aimed at preventing multiple injury mechanisms Comparison: control
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Injury rate, adjusted for clustering, with control	Injury rate, adjusted for clustering, with School injury prevention programmes
Self‐reported medically or non‐medically attended unintentional injuries or injuries with an unspecified intent (injury rates adjusted for clustering) assessed with: self‐report follow‐up: range 12 months to 24 months	367 per 1000 person‐years	243 per 1000 person‐years	Rate ratio 0.76 (0.49 to 1.17)	12,977 (2073 person‐years adjusted for clustering) (2 RCTs, 1 CBA)	⊕⊕⊝⊝ Low ¹	We excluded 4 studies from the meta‐analysis due to varied interventions (e.g. sports, agriculture or risk‐taking injury prevention). Their findings were consistent with the meta‐analysis studies.
Safety skills assessed with: observations follow‐up: range 4 months to 5 months	Both studies found an improvement in observed safety skills (Kendrick 2007 ‐ fire and burn prevention skills: odds ratio 8.93 (95% CI 1.67 to 47.78, P = 0.01); Frederick 2000 ‐ basic life support skills, P < 0.005 for assessment of danger, responsiveness and circulation).		‐	1751 (1 RCT, 1 CBA)	⊕⊕⊝⊝ Low ²	Interventions and safety skills observed were varied in these 2 studies.
Behaviour assessed with: observations and self‐reported follow‐up: range 2 weeks to 36 months	All 4 studies (5 articles) reported that observed safety behaviour showed evidence of improved practices and 13/19 studies describing self‐reported behaviour reported improved practices.		‐	52,950 (9 RCTs, 4 non‐RCTs, 6 CBA)	⊕⊝⊝⊝ Very low³	Behaviours included safety equipment wearing, road risk‐taking behaviour, agriculture and sports‐related injury behaviours.
Safety knowledge assessed with: surveys and self‐completion tests follow‐up: range 1 month to 36 months	Of the 21 studies assessing changes in safety knowledge, 19 reported an improvement in at least 1 question domain in the intervention compared to the control group.		‐	55,732 (9 RCTs, 5 non‐RCTs, 7 CBA)	⊕⊝⊝⊝ Very low⁴	Outcomes included a wide range of knowledge testing instruments and topics.
Cost‐effectiveness assessed with: cost:benefit ratio follow‐up: 1 years	For every 1 Chinese Yuan spent, 13.90 was saved (cost:benefit ratio 1:13.9)		‐	4639 (1 CBA)	⊕⊝⊝⊝ Very low⁵	Only 1 study reported economic outcomes and should, therefore, be interpreted with caution.
CI: confidence interval; CBA: controlled before‐and‐after study; RCT: randomised controlled trial; non‐RCT: non‐randomised controlled trial.
GRADE Working Group grades of evidence High quality: We are very confident that the true effect lies close to that of the estimate of the effect Moderate quality: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low quality: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low quality: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹ Downgraded twice because of inconsistency as the I² = 63%, indicating substantial heterogeneity and because there was imprecision in the results (the rate ratio was 0.76 but the confidence intervals spanned 1). Two of the studies were RCTs with a low risk of bias. ² Downgraded twice because of a serious risk of bias (one of the two studies was a CBA resulting in selection biases relating to which schools participated in studies) and inconsistency, which was rated as serious because the two studies had different intervention types with different outcome measures. Imprecision was serious as there were wide confidence intervals in one of the two included studies, as well as a paucity of data. However, the effect sizes were classed as large as there was a nearly nine times greater odds of fire and burn prevention skills in the Risk Watch programme (Kendrick 2007), and 33% greater skills assessment in the Injury Minimisation Programme for Schools (IMPS) relating to assessment of danger (Frederick 2000). ³ Downgraded three times because 10 studies were CBA or non‐RCT design with high risk of selection bias of included schools, there was serious inconsistency in methods of collecting data and intervention types, and this may have contributed to the wide range of effect sizes and directions seen (no effect or some effect), There was often wide confidence intervals in results presented. ⁴ Downgraded three times because 11 studies were CBA or non‐RCT design with high risk of selection bias of included schools; there was serious inconsistency in knowledge tested, questionnaire designs and methods of collecting data, a wide range of intervention types and pedagogical approaches and this may have contributed to the wide range of effect sizes and directions seen (no effect or some effect). Results were presented in a way that often made precision difficult to compare. ⁵ Downgraded three times because this was one study, limited in applicability with high or unclear risk of bias across multiple domains.

Summary of findings for the main comparison. School injury prevention programmes compared to controls for the prevention of unintentional injuries in children and young people

Table 1. Age ranges included in the studies

Study ID

Age (years)

4/5

5/6

6/7

7/8

8/9

9/10

10/11

11/12

12/13

13/14

14/15

15/16

16/17

17/18

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

●

‐

‐

●

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐

●

‐

‐: not applicable; ●: age range covered.

Table 1. Age ranges included in the studies

Table 2. Injury mechanisms targeted for included studies

Study ID

Injury mechanism

Pedestrian

Cycle

Motorcycle

Vehicle (non‐specific)

Falls

Swimming/ drowning/ diving

Smoke/ fire/ flames

Heat and hot surfaces

Sport/ physical activity

Sun

Household safety

Agricultural injuries

Poisoning

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

‐

●

‐

‐

●

‐

●

‐

●

‐

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

‐

●

‐

●

‐

●

‐

●

‐

‐

●

‐

●

‐

●

‐

‐

●

‐

‐

●

‐

●

‐

‐

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

‐

●

●

‐

‐: outcome not measured; ●: outcome measured.

Table 2. Injury mechanisms targeted for included studies

Comparison 1. Medically attended or non‐medically attended injury rates

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Injury rates at follow‐up, adjusted for baseline injury rates in non‐randomised studies Show forest plot	3	2073	Rate Ratio (Random, 95% CI)	0.73 [0.49, 1.08]

Comparison 1. Medically attended or non‐medically attended injury rates