Abstract
Background. Schools are sites of dating and relationship violence (DRV) and of gender-based violence (GBV) victimization and perpetration. School-based interventions can reach a broad range of students, targeting both individual and group processes that may underpin DRV and GBV. Considering DRV and GBV jointly is important because of their shared etiologies. Comparing the effectiveness of interventions using network meta-analysis (NMA) can support decision-making on optimal resource use.
Objectives. To evaluate the comparative effectiveness of school-based interventions for children aged 5 to 18 years on DRV and GBV victimization, perpetration, and related mediators.
Search Methods. We searched 21 databases in July 2020 and June 2021, alongside extensive supplementary search methods, including gray literature searches, forward and backward citation chasing, and searches on first and last author names.
Selection Criteria. We included randomized-controlled trials of interventions for children of compulsory school age implemented within the school setting, and either partially or wholly aimed at changing DRV or GBV outcomes.
Data Collection and Analysis. Pairwise meta-analyses using random-effects robust variance estimation considered intervention effectiveness on DRV and GBV victimization and perpetration using odds ratios, and on mediators (e.g., knowledge and attitudes) using standardized mean differences. Effects were divided into short-term (< 12 months postbaseline) and long-term (≥ 12 months postbaseline). NMAs on victimization and perpetration outcomes compared interventions categorized by breadth of mechanism and complexity of delivery and implementation. Meta-regression tested sensitivity to percentage of girls in the trial sample and country context.
Main Results. Our analysis included 68 trials. Evidence was stronger overall for effects on DRV than for GBV, with significant long-term impacts on DRV victimization (odds ratio [OR] = 0.82; 95% confidence interval [CI] = 0.68, 0.99) and DRV perpetration (OR = 0.78; 95% CI = 0.64, 0.94). Knowledge and attitudinal effects were predominantly short-term (e.g., for DRV-related violence acceptance, d = 0.16; 95% CI = 0.08, 0.24). NMAs did not suggest the superiority of any intervention type; however, most analyses for GBV outcomes were inconsistent. A higher proportion of girls in the sample was associated with increased effectiveness on long-term victimization outcomes.
Author’s Conclusions. Evidence is stronger for DRV than for GBV, despite considerable heterogeneity. Certainty of findings was low or very low overall.
Public Health Implications. Violence reductions may require more than 1 school year to become apparent. More extensive interventions may not be more effective. A possible reason for stronger effectiveness for DRV is that whereas GBV is ingrained in school cultures and practices, DRV is potentially more open to change via addressing individual knowledge and attitudes. (Am J Public Health. 2023;113(3):320–330. https://doi.org/10.2105/10.2105/AJPH.2022.307153)
Plain-Language Summary
Dating and relationship violence and gender-based violence in adolescents and young people remain major issues for school health, especially given that schools are major sites for perpetration and victimization of both DRV and GBV. School-based prevention of DRV and GBV has been tested in many forms, but patterns of effectiveness across both types of outcomes have not been considered. This is especially important because of growing social awareness of how DRV and GBV are linked by toxic patriarchal norms. We searched 21 databases to find randomized trials of school-based interventions for DRV and GBV, and meta-analyzed them by short-term (< 12 months from baseline) and long-term (≥ 12 months from baseline). We included 68 trials. These trials suggested that long-term, but not short-term, impacts on victimization and perpetration for DRV were in evidence, but did not offer clear evidence of effectiveness for GBV outcomes. These trials also suggested that interventions could have short-term impacts on knowledge and attitudes, such as violence acceptance. An additional analysis that compared types of interventions did not find that more extensive (more components, broader implementation) interventions were necessarily more effective. This means that schools may need to wait longer than 1 school year to see impacts.
Conservative estimates suggest that between a quarter and a third of school-age children experience dating and relationship violence (DRV), such as physical, sexual, and psychological abuse (including online abuse and coercive control),1 although rates of DRV in excess of two thirds of students have been reported in some contexts.2 Students also describe gender-based violence (GBV) as “commonplace” in schools, including sexual harassment and homophobic and transphobic bullying, with sexual assaults reported in school spaces.3 DRV and GBV share risk factors and antecedent attitudes,4,5 including patriarchal gender norms at the societal level, inconsistently enforced violence prevention policies at the school level, and, at the individual level, exposure to and reinforcement of antisocial GBV-related norms.1,2,6 They also share pervasive consequences for both survivors and perpetrators, including poor mental health, low self-esteem, and risky sexual behavior2,7; consequences for academic performance and school engagement8,9; and elevated risk of intimate partner violence as adults.
Schools are sites of DRV and GBV victimization and perpetration, but they are also important venues for intervention. School-based interventions can reach a broad range of students, targeting both individual and group processes that may underpin DRV and GBV.10–12 Previous reviews11,13,14 have evaluated the effectiveness of interventions for DRV and GBV but have not considered how these affect DRV and GBV outcomes jointly, despite overlap in antecedents. Understanding the effectiveness on both outcomes together supports greater knowledge of approaches for each outcome and informs joint implementation of interventions to reduce DRV and GBV concurrently. Several older reviews require updating to assess newer interventions, but the most recent major review12 also missed relevant studies because of an unduly narrow approach to literature searches. Evaluations of school-based interventions are often published within gray literature (i.e., not in mainstream databases), and therefore reviews without rigorous searches of gray literature sources may exclude relevant data. Finally, to date, no previous review has undertaken a network meta-analysis (NMA) on DRV or GBV outcomes, capitalizing on a mature evidence base to estimate the comparative effectiveness of interventions. An NMA able to identify patterns in the effectiveness of interventions—such as in the breadth or level of delivery, mechanisms of action, and implementation efforts required—would be of value for policymakers seeking to select an intervention for their schools, particularly given sustained policy interest in whole-school approaches despite their complexity.3,6 Thus, this systematic review sought to evaluate the effectiveness of school-based interventions on DRV and GBV victimization and perpetration among children aged 5 to 18 years, as well as the factors—including knowledge and attitudes—that might mediate reductions in victimization and perpetration. It also presents, for the first time, an NMA of the comparative effectiveness of intervention types on DRV and GBV victimization and perpetration.
METHODS
This review was registered on PROSPERO (CRD42020190463).
Search Methods
In July 2020, we searched the following databases without limitation on date or language: MEDLINE, Embase, PsycINFO, Social Policy and Practice (Ovid); CINAHL, ERIC, British Education Index, Education Research Complete, EconLit, Criminal Justice Abstracts (EBSCOhost); Cochrane Database of Systematic Reviews and the Cochrane Central Register of Controlled Trials (via the Cochrane Library, Wiley); NHS Economic Evaluation Database (via the Centre for Reviews and Dissemination); Social Science Citation Index and Conference Proceedings Citation Index (Web of Science, Clarivate Analytics); Australian Education Index, ProQuest Dissertations & Theses Global, Sociological Abstracts including Social Services Abstracts, Applied Social Sciences Index and Abstracts (ProQuest); Trials Register of Promoting Health Interventions and Bibliomap (EPPI-Centre); and Campbell Systematic Reviews (Campbell Collaboration). We updated the bibliographic database searches in June 2021 and added further free-text search terms for named interventions. The timing of searches was chosen to coincide with the requirements of the funder and in preparation for submission of the funder report.
Our database searches included free-text terms and subject headings for schools and for DRV and GBV. We used forward and backward citation chasing on included studies in Scopus (Elsevier), Web of Science, and Google Scholar, and we reviewed the reference lists of relevant systematic reviews and reports. To identify linked studies and further gray literature, we conducted targeted searches in Web of Science and Scopus using first and last author names, and searched Google Scholar for specific intervention names (e.g., Project Respect, Shifting Boundaries). We also searched or browsed publication lists on key Web sites, and searched clinical trial registers (ClinicalTrials.gov, WHO ICTRP). Where missing data from trial publications was expected to affect the analysis, we contacted authors to request additional information.
All search results were downloaded into EndNote ×9 (Clarivate Analytics, London, UK) for deduplication. Further details are provided in Appendix A (available as a supplement to the online version of this article at http://www.ajph.org).
Selection Criteria
Randomized-controlled trials (RCTs) were eligible for inclusion, including cluster trials. The population was restricted to children of compulsory school age (5–18 years). Relevant interventions were implemented within the school setting (including out of school hours, provided these were conducted with school cohorts), and either partially or wholly aimed at changing DRV or GBV outcomes. We excluded interventions that might have had only opportunistic effects on DRV or GBV outcomes—for example, through another health promotion effect (e.g., healthy eating). No restriction was placed on the content of interventions, which may have involved delivery to individual or groups of students, training of staff or school personnel, and interventions targeting local and school policy changes. Interventions may have been delivered by school staff or by an external organization, or entirely peer-led (e.g., through a computerized module). Comparisons with control or other active intervention were included.
Search records were screened by 2 reviewers at both the title and abstract level and full-text level. Publications were not excluded at the title and abstract level based on outcome. Disagreements were resolved through discussion and with a third reviewer where required. A reviewer extracted data into a data extraction form developed and piloted a priori and checked by a second reviewer. Data extracted included details about the study design, study sample, intervention characteristics, analysis methods, and outcome data.
Outcomes
Outcomes included victimization or perpetration of DRV or GBV. DRV included physical violence, emotional violence (including isolation, coercive control and cyber abuse), and sexual assault within a dating relationship. Where physical and sexual DRV were considered jointly in an outcome, this was treated as a separate outcome type. GBV included violence outside of a relationship, such as harassment and bullying on the basis of gender or sexuality (including homophobic and transphobic bullying), cyber abuse (including unwanted sexting or forwarding of sexts), unwanted sexual contact (such as groping or “upskirting”), sexual assault, and rape. Trials varied in the measurement of DRV and GBV outcomes, and a pragmatic decision was taken to group together outcomes across studies based on the types of violence measured. Groupings were informed by outcome descriptions in the original studies and, where available, inspection of measurement items. For both DRV and GBV, “omnibus” measures were overall measures without differentiation (e.g., by emotional, physical, or cyber abuse). In addition, knowledge, attitudes, and behaviors related to DRV and GBV were included, such as rape myth acceptance, bystander attitudes, and GBV-condoning norms; these were grouped by similarity of construct. We did not include outcomes related to “honor”-based violence, forced marriage, or female genital cutting. Outcomes were quantitative, and included categorical, count, and continuous measures, using bespoke or validated measures. We extracted relevant moderators in included trials.
Pairwise Meta-Analysis
Analyses were based on intention-to-treat data reported by trials; per protocol data were only included if intention-to-treat analyses were not available, and were downgraded during quality appraisal. Outcomes were grouped, by length of follow-up, as short-term (< 1 year) or long-term (≥ 1 year). Pairwise meta-analyses of comparisons against control were conducted, grouped by outcome (DRV or GBV type) as per the review protocol and availability of evidence in the included trials.
The key metric for primary outcomes was the odds ratio (OR); where outcome measures were continuous, we converted these to ORs using a logistic transformation. We meta-analyzed mediators using standardized mean differences. Meta-analyses used robust variance estimation. This approach improves on previous strategies for dealing with multiple relevant effect sizes per study (e.g., from several treatment arms or effect estimates), such as artificially splitting meta-analyses or choosing 1 effect size, by including all relevant effect sizes but adjusting for interdependencies within studies.15 As heterogeneity across study designs and interventions was anticipated, meta-analyses used a random-effects model as default. We assessed heterogeneity in part using I2, defined as substantial (> 60%), moderate (31%–60%), little (6%–30%), and minimal (≤ 5%). For cluster trials, where the intracluster correlation coefficient was not explicitly modeled or reported, we imputed an estimate of 0.05 based on other studies used within the review, as recommended by Cochrane guidance.16 Following adjustment, data from cluster trials were pooled with RCTs.
Network Meta-Analysis
We conducted NMAs of study effects, including trials of head-to-head comparisons, to compare the effectiveness of intervention types on DRV and GBV perpetration and victimization outcomes. On the basis of a components analysis informed by stakeholder consultation and policy priorities for school health, we grouped interventions according to delivery type, breadth of mechanism, and implementation (single-component, curriculum, multicomponent, and multilevel interventions; Table 1). We used a frequentist framework via “network” in Stata version 17 (StataCorp LP, College Station, TX). We included correlations between arms in multiarm trials using estimates from trial reports, and a common between-study variance parameter was used across the network. Because of unresolved heterogeneity in effects across trials identified in pairwise meta-analyses, only random-effects models were fitted. We explored analyses for inconsistency using design-by-treatment interaction models, and transitivity was assessed and explored by considering known effect modifiers (e.g., network meta-regression) and the similarity of interventions in each node with respect to the intervention groupings. We then ranked interventions in consistent models using 1000 bootstrap draws, with rankings summarized using the surface under the cumulative ranking curve (SUCRA). SUCRA values balance the precision of, and numerical differences between, estimates and integrate the probability of each intervention type at each rank. SUCRA values produce estimates of how interventions compare with a hypothetical situation where each intervention had 100% probability of ranking first. Where trials reported multiple effect sizes for the same outcome (e.g., different types of DRV victimization), we assumed outcomes to be correlated with ρ = 0.8.15
TABLE 1—
Typology of Delivery, Mechanisms, and Implementation for School-Based Interventions for Dating and Relationship Violence and Gender-Based Violence
| Type | Description |
| Single-component interventions | |
| Delivery | Generally brief (e.g., 25–30 min) single sessions or a few sessions (≤ 5). May or may not require in-person facilitators. |
| Mechanisms | Focuses on a single, or very narrow range of, change mechanism. |
| Implementation | Often delivered through a key technology as integral to effectiveness (e.g., video game, online, immersive virtual environments). |
| Curriculum-based interventions | |
| Delivery | Generally delivered in more sessions (≥ 6) and over a longer term (ranging from several weeks to several years), by extensively trained external in-person facilitators following specific manuals, lesson plans, or scripts for each session. |
| Mechanisms | Focuses on a narrow range of change mechanisms at 1 or 2 levels but does not address higher-level (i.e., structural) change mechanisms. |
| Implementation | Can be integrated into existing school curriculum (personal, social, and health education, etc.) or else delivered in a classroom environment in place of existing subjects for a short period of time. |
| Multicomponent interventions | |
| Delivery | Generally delivered using a variety of modes of intervention for varying durations, including but not limited to curriculum, theater productions, videos, presentations, group and pair discussions, individual work, and the Internet. |
| Mechanisms | Can address multiple change mechanisms across multiple levels but does not extensively address structural change mechanisms. |
| Implementation | Requires some school staff investment and external facilitation. |
| Multilevel interventions | |
| Delivery | Uses a variety of modes over several ecological levels in schools, beyond just instructing students or school personnel. Integrates explicit components relating to social structural or structural environmental domains. |
| Mechanisms | Addresses a range of change mechanisms over multiple ecological levels. |
| Implementation | Requires a combination of school staff investment and external facilitation. |
Quality Appraisal and Sensitivity Analysis
Two reviewers (C. F. and a research assistant) appraised all trials for quality using an adapted Cochrane risk of bias tool.16 In the main, appraisals were guided by the tool; however, trials were not downgraded for unblinded outcome assessors within the outcome measurement domain. This decision avoided a floor effect in quality-appraisal ratings as in most trials it was infeasible for study authors to blind or obscure study aims from students. Appraisal decisions were quality assured by a third reviewer (G. J. M.-T.) and disagreements resolved through discussion. We generated comparison-adjusted funnel plots to investigate publication bias for primary outcomes. We sensitivity analyzed primary outcomes using meta-regression on country context (high-income vs low-income and middle-income) and percentage of girl children in the trial sample. These were most commonly identified by stakeholders as likely moderators of effectiveness. Pairwise meta-regressions used common between-study variance parameters between groups. Network meta-regressions additionally assumed a common coefficient across all comparisons against control.
RESULTS
Characteristics of studies included in the review are provided in online Appendix B. Following de-duplication, we screened 40 160 records on title and abstract, and 788 records on full-text (Figure 1). Of these, we included 68 RCTs evaluating 80 interventions for DRV or GBV. These included 14 RCTs and 54 cluster RCTs that compared interventions against a control intervention (n = 66, including an active control intervention, usual practice, waitlist, or no intervention) or another active intervention (n = 8). Head-to-head comparisons were of different interventions (n = 4), of additional components (n = 3), or of different methods of implementation (n = 3). More interventions were identified as targeting DRV (n = 43) than GBV (n = 15), and 14 interventions were identified as targeting both. The intended target was unclear for 8 interventions, although these trials were included because the intervention content included topics considered relevant to either DRV or GBV.
FIGURE 1—
Flowchart of Studies in the Review: School-Based Interventions to Prevent Dating and Relationship Violence and Gender-Based Violence
Most studies (n = 42) were conducted in North America, with the remaining split across Europe (n = 9), Asia (n = 8), Africa (n = 6), and South America (n = 3). Across these trials, 50 were undertaken in high-income country contexts. Sample sizes ranged from 47 to 89 707 participants (median = 839). Studies were mostly conducted in middle or high schools (i.e., ages 11–18 years). Only 4 studies also or solely included students within primary or junior schools. Most trials were conducted with male and female students, whereas 4 and 6 studies, respectively, were conducted exclusively with male or female students. Only 2 studies permitted students to record gender beyond the binary, and only 5 studies included students’ self-reported sexuality. No studies included solely LGBTQ+ (lesbian, gay, bisexual, transgender, or queer) students. Only half of included studies (52.9%) reported student race or ethnicity; of these, more than 50% of students identified as White or Caucasian (37.8%), Hispanic or Latino (18.9%), and Black or African American (10.8%). School or students’ socioeconomic status (SES) was reported for 35 studies, of which 11 included more than 50% of students from lower SES backgrounds (e.g., free or subsidized school lunches, or in areas with high economic deprivation). No identified studies exclusively included students who had experienced DRV or GBV; however, 2 studies included only participants considered at risk for DRV.
Interventions included single-component interventions (n = 22 RCTs), curriculum interventions (n = 11), multicomponent interventions (n = 15), and multilevel interventions (n = 22). Half of all interventions included full or partial implementation by external agencies (50.1%). A minority of interventions included a self-study (12.5%) or digital (15.0%) component (e.g., use of virtual reality games). DRV and GBV interventions were not clearly different in choice of facilitator or delivery method.
Quality of Included Studies
Critical appraisals for included outcome evaluations are reported in online Appendix B. Only 1 included trial17 was appraised at overall low risk for bias (1.5%). The other trials were split between those appraised as having “some concerns” (54.4%) and those considered to be at high risk for bias (44.1%). The main risk of bias issues in the included trials were as follows:
-
•
unclear allocation concealment, with most trials using simple randomization procedures that can be open to manipulation;
-
•
potential for contamination in schools where students may mix with those in other intervention arms, or teachers trained to deliver the intervention may alter their behavior toward students in the control arm; and
-
•
loss of clusters following randomization without evidence that drop-out was unrelated to trial outcomes.
Pairwise Meta-Analyses
Pairwise meta-analyses for interventions compared with controls are reported in Table 2, with forest plots in online Appendix C. Findings suggested that school-based interventions were effective compared with controls in reducing the victimization and perpetration of DRV. A reduction in DRV was shown across subtypes of violence, and was greater at long-term follow-up. However, effect estimates were substantially heterogeneous, with wide confidence intervals (CIs) typically crossing the line of null effect. School-based interventions may be effective for reducing victimization and perpetration of GBV; however, effects were smaller than for DRV, and all effects were highly imprecise. Findings were similar across subtypes of GBV, and heterogeneity was also substantial in these analyses. GRADE for pairwise meta-analyses (Appendix C) led to all outcomes rated as low or very low certainty of evidence, owing primarily to substantial unexplained heterogeneity and risk of publication bias.
TABLE 2—
Pairwise Meta-Analyses of School-Based Interventions for Dating and Relationship Violence (DRV) and Gender-Based Violence (GBV) Compared to Controls
| Outcome | Short-Term Follow-Up (< 1 y) | Long-Term Follow-Up (≥ 1 y) | ||||||
| k | No. | OR (95% CI) | I2 (%) | k | No. | OR (95% CI) | I2 (%) | |
| DRV victimization | ||||||||
| All outcomes | 17 | 118 | 0.90 (0.80, 1.02) | 81 | 13 | 79 | 0.82 (0.68, 0.99) | 80 |
| Omnibus | 10 | 45 | 0.88 (0.69, 1.12) | 84 | 5 | 12 | 0.85 (0.63, 1.15) | 52 |
| Emotional/psychological | 8 | 16 | 0.84 (0.55, 1.27) | 90 | 9 | 21 | 0.81 (0.59, 1.12) | 88 |
| Physical | 5 | 14 | 0.93 (0.69, 1.25) | 64 | 6 | 21 | 0.84 (0.61, 1.16) | 82 |
| Sexual | 7 | 29 | 0.97 (0.88, 1.08) | 76 | 5 | 13 | 0.88 (0.59, 1.31) | 78 |
| Physical/sexual | 4 | 8 | 0.85 (0.43, 1.69) | 76 | 5 | 9 | 0.90 (0.53, 1.51) | 73 |
| Cyber | 3 | 6 | 0.82 (0.31, 2.16) | 87 | 2 | 3 | 0.57 (0.45, 0.72) | 0 |
| DRV perpetration | ||||||||
| All outcomes | 18 | 118 | 0.91 (0.80, 1.04) | 83 | 16 | 79 | 0.78 (0.64, 0.94) | 79 |
| Omnibus | 11 | 43 | 0.95 (0.85, 1.07) | 70 | 7 | 15 | 0.74 (0.52, 1.06) | 75 |
| Emotional/psychological | 9 | 19 | 0.77 (0.54, 1.11) | 90 | 9 | 21 | 0.77 (0.59, 1.01) | 85 |
| Physical | 7 | 16 | 0.91 (0.71, 1.18) | 83 | 7 | 22 | 0.83 (0.59, 1.18) | 80 |
| Sexual | 7 | 30 | 0.99 (0.86, 1.13) | 79 | 4 | 9 | 0.85 (0.37, 1.92) | 60 |
| Physical/sexual | 3 | 6 | 0.82 (0.13, 5.29) | 76 | 5 | 9 | 0.77 (0.42, 1.43) | 78 |
| Cyber | 2 | 4 | 0.96 (0.77, 1.18) | 71 | 2 | 3 | 0.49 (0.38, 0.63) | 50 |
| GBV victimization | ||||||||
| All outcomes | 13 | 72 | 0.88 (0.76, 1.02) | 75 | 11 | 58 | 0.93 (0.80, 1.08) | 66 |
| Omnibus | 7 | 29 | 1.00 (0.91, 1.10) | 60 | 7 | 17 | 0.93 (0.79, 1.10) | 41 |
| Emotional/verbal | 2 | 2 | 0.94 (0.82, 1.08) | 0 | 3 | 11 | 0.92 (0.56, 1.52) | 76 |
| Physical | 9 | 40 | 0.76 (0.62, 0.93) | 78 | 6 | 25 | 0.91 (0.68, 1.23) | 66 |
| Homophobic | 1 | 1 | 1.01 (0.77, 1.33) | … | 1 | 5 | (Not estimable) | |
| GBV perpetration | ||||||||
| All outcomes | 11 | 67 | 0.95 (0.85, 1.07) | 66 | 9 | 58 | 0.90 (0.73, 1.12) | 67 |
| Omnibus | 9 | 30 | 0.97 (0.88, 1.06) | 55 | 6 | 15 | 0.98 (0.73, 1.30) | 57 |
| Emotional/verbal | 2 | 2 | 0.85 (0.40, 1.80) | 76 | 4 | 12 | 0.86 (0.60, 1.24) | 63 |
| Physical | 5 | 33 | 0.87 (0.62, 1.23) | 77 | 5 | 24 | 0.79 (0.48, 1.28) | 68 |
| Homophobic | 2 | 2 | 1.06 (0.85, 1.32) | 0 | 2 | 7 | 0.95 (0.89, 1.02) | 38 |
Note. CI = confidence interval; OR = odds ratio.
Meta-regression sensitivity analyses (Appendix C) also showed that country context moderated effects for DRV or GBV, particularly at long-term follow-up. At 1 year or longer after baseline, interventions in high-income contexts were associated with larger reductions in the odds of DRV and GBV victimization and perpetration (ORs = 0.71–0.86; all Ps < .05). Furthermore, the proportion of girls in the trial sample moderated effects for DRV and GBV victimization, but not for DRV or GBV perpetration. With each additional 10% points of girls in the sample, the odds of DRV victimization decreased by 22% (although the effect was marginally nonsignificant; OR = 0.78; 95% CI = 0.59, 1.04) and the odds of GBV victimization decreased by 9% (OR = 0.91; 95% CI = 0.85, 0.97).
Analyses of Knowledge and Attitudes
Meta-analyses of knowledge and attitudes are presented in online Appendix C. Overall, interventions were effective at improving short-term DRV-focused violence acceptance (d = 0.16; 95% CI = 0.08, 0.24), knowledge (d = 0.69; 95% CI = 0.18, 1.20), attitudes to intervening (d = 0.14; 95% CI = 0.01, 0.26), and attitudes to personal help-seeking (d = 0.14; 95% CI = 0.06, 0.22), but none of these effects was maintained in long-term analyses. Interventions improved GBV-focused violence acceptance (d = 0.29; 95% CI = 0.11, 0.33), knowledge (d = 0.68; 95% CI = 0.26, 1.11), and individual self-efficacy (d = 0.16; 95% CI = 0.08, 0.25) in the short term, with only violence acceptance having a credible long-term effect.
Network Meta-Analyses
NMA findings (Table 3) suggested that single-component interventions may be useful for reducing short-term and long-term DRV victimization and perpetration, and also short-term GBV victimization (no single-component interventions were tested in long-term GBV victimization). Multilevel interventions also showed effectiveness for long-term DRV victimization. For GBV outcomes, there was strongest evidence for curriculum interventions, which were more successful than other intervention types at short-term follow-up of victimization and short-term and long-term perpetration. Consistency tests for short-term DRV outcomes yielded no evidence of inconsistency (victimization: χ2 = 0.29, df = 3, P = .96; perpetration: χ2 = 0.16, df = 3, P = .98). However, inconsistency tests were significant for all GBV analyses except for short-term GBV victimization (χ2 = 7.24, df = 3, P = .06), driven primarily by 1 trial.
TABLE 3—
Network Meta-Analyses of School-Based Interventions for Dating and Relationship Violence (DRV) and Gender-Based Violence (GBV), by Intervention Typology
| Short-Term Follow-Up, OR (95% CI) | Long-Term Follow-Up, OR (95% CI) | |||||||
| Single | Curriculum | Multicomponent | Multilevel | Single | Curriculum | Multicomponent | Multilevel | |
| DRV victimization | ||||||||
| Control | 0.88 (0.75, 1.03) | 0.97 (0.70, 1.34) | 1.01 (0.84, 1.22) | 0.89 (0.76, 1.05) | 0.60 (0.41, 0.86) | 0.92 (0.61, 1.40) | 0.94 (0.73, 1.20) | 0.83 (0.69, 1.00) |
| Single | 1.10 (0.77, 1.57) | 1.15 (0.90, 1.47) | 1.02 (0.81, 1.27) | 1.54 (0.88, 2.69) | 1.57 (1.01, 2.45) | 1.39 (0.92, 2.10) | ||
| Curriculum | 1.04 (0.72, 1.51) | 0.92 (0.69, 1.24) | 1.02 (0.63, 1.66) | 0.90 (0.62, 1.31) | ||||
| Multicomponent | 0.88 (0.70, 1.11) | 0.88 (0.65, 1.21) | ||||||
| DRV perpetration | ||||||||
| Control | 0.81 (0.65, 1.02) | 0.90 (0.68, 1.19) | 0.99 (0.79, 1.24) | 0.87 (0.70, 1.08) | 0.57 (0.40, 0.83) | 0.97 (0.63, 1.50) | 0.83 (0.65, 1.06) | 0.86 (0.69, 1.06) |
| Single | 1.11 (0.78, 1.58) | 1.22 (0.89, 1.68) | 1.07 (0.79, 1.45) | 1.69 (0.94, 3.02) | 1.44 (0.90, 2.29) | 1.49 (0.96, 2.32) | ||
| Curriculum | 1.10 (0.78, 1.56) | 0.96 (0.72, 1.29) | 0.85 (0.53, 1.37) | 0.88 (0.61, 1.29) | ||||
| Multicomponent | 0.87 (0.66, 1.17) | 1.04 (0.77, 1.39) | ||||||
| GBV victimization | ||||||||
| Control | 0.87 (0.66, 1.14) | 0.72 (0.54, 0.95) | 0.95 (0.77, 1.16) | 0.90 (0.75, 1.09) | . . . | 0.93 (0.66, 1.32) | 0.89 (0.75, 1.07) | 0.95 (0.77, 1.18) |
| Single | 0.83 (0.57, 1.21) | 1.09 (0.78, 1.53) | 1.04 (0.74, 1.46) | . . . | . . . | . . . | ||
| Curriculum | 1.32 (0.94, 1.85) | 1.26 (0.94, 1.69) | 0.96 (0.67, 1.37) | 1.02 (0.69, 1.53) | ||||
| Multicomponent | 0.95 (0.74, 1.23) | 1.07 (0.80, 1.44) | ||||||
| GBV perpetration | ||||||||
| Control | 1.00 (0.90, 1.11) | 0.88 (0.70, 1.10) | 0.95 (0.85, 1.06) | 0.89 (0.78, 1.02) | . . . | 0.82 (0.54, 1.26) | 0.89 (0.70, 1.14) | 0.95 (0.72, 1.25) |
| Single | 0.88 (0.69, 1.11) | 0.95 (0.81, 1.11) | 0.89 (0.75, 1.07) | . . . | . . . | . . . | ||
| Curriculum | 1.08 (0.83, 1.41) | 1.02 (0.83, 1.25) | 1.08 (0.66, 1.76) | 1.15 (0.79, 1.68) | ||||
| Multicomponent | 0.94 (0.81, 1.09) | 1.07 (0.74, 1.54) | ||||||
Note. CI = confidence interval; OR = odds ratio. Ellipses denote “not applicable.”
Assessment of transitivity suggested that interventions were more similar within node than between node, but effect modifiers (specifically, country context and sex) were explored to evaluate the impact of imbalances (Appendix C). These analyses had minimal effect on DRV outcomes. However, accounting for the percentage of girls in the trial sample led to comparable effectiveness for short-term GBV victimization across intervention types (although no effect at long-term). Controlling for country context did not affect short-term GBV victimization or perpetration, although in long-term analyses curriculum interventions became more effective.
Rank data are presented in full in Appendix C for consistent NMAs. Overall, single-component interventions were most likely to be top-ranked for DRV victimization and perpetration (SUCRA = 0.8–1.0). Curriculum interventions were most likely to be top-ranked for GBV victimization in the short term (SUCRA = 0.9).
Publication Bias Analyses
Funnel plots (Appendix C) showed evidence of publication bias in short-term DRV victimization, DRV perpetration, and GBV victimization, and in long-term DRV perpetration and GBV victimization. In most cases, bias was toward publication of positive intervention effects by smaller trials, although for GBV victimization, smaller trials were more likely to report negative effects.
DISCUSSION
The results of this comprehensive systematic review and meta-analysis of school-based interventions for DRV and GBV suggest that evidence for the effectiveness of school-based interventions is stronger for DRV than for GBV. However, effects may not be immediate and may require more than 1 school year to become apparent. Effects are evident in the aggregate rather than for any specific type of DRV or GBV. This is an advantage of our analysis strategy, which used an innovative statistical method to integrate all relevant evidence. Interventions are also linked with primarily short-term effects on knowledge and attitudinal mediators. Our consideration of mediators is the most exhaustive to date. It is possible that, whereas effects on mediators may have faded after short-term measurement, longer-term behavior change occurred via changes in school social systems, practices, and norms that may be less amenable to measurement in terms of knowledge and attitudes.
However, there are some caveats in this body of evidence. First, there were clear differences in the sufficiency of evidence for different types of violence. For example, homophobic GBV was evaluated in very few trials, despite clear evidence of DRV and GBV inequalities in sexual-minority groups.18 Moreover, most GBV analyses relied on omnibus measures that did not distinguish between types of violence. Very few studies reported data for groups at higher risk for DRV or GBV—for example, those with experience of violence, or sexual minorities.19 In addition, publication bias was assessed as a serious risk for several of the victimization and perpetration outcomes.
NMAs for most GBV outcomes were inconsistent, limiting interpretation of this evidence. Specifically, inconsistency in NMAs derived from conflicts between trials comparing different intervention types directly and trials comparing each intervention type against control. This again suggests that evidence supporting the effectiveness of interventions was stronger for DRV than for GBV. GBV victimization and perpetration effectiveness was also moderated by country context. Although this did suggest that interventions were effective for GBV in high-income contexts, these analyses relied on relatively few studies and meta-regressions are not causal. Future evaluations should also consider our findings related to the proportion of girls in trial samples and its relationship to victimization outcomes. Interventions addressed to mixed-sex audiences could only show effectiveness because of the positive effects on girls’ victimization, rather than on adolescents’ perpetration. This is important because decreases in victimization do not suggest specifically that primary prevention of violence is occurring, only that the violence being committed by anyone in the sample is (postintervention) more often directed outside of the sample.
Strengths and Limitations
Compared with prior reviews, our analysis has several strengths. First, the extensive and wide-ranging search permitted a clearer perspective as to the effectiveness of interventions on a range of mediators. Moreover, we were able to include a number of RCTs, including in the gray literature, that previous reviews have not included. Second, our joint consideration of DRV and GBV highlights an important gap in the evidence that requires further consideration; specifically, why intervention impacts appear stronger for DRV than for GBV.
However, our analysis has several limitations. First, we cannot exclude the possibility that relevant trials were missed either because of database indexing or the “file drawer problem,” and indeed, our analysis of publication bias indicates some risk of this. Second, we did not analyze broader gender norms and related constructs (e.g., homophobia generally), given the need to identify clear inclusion and exclusion criteria. Third, substantial heterogeneity in intervention effects reflected that variation between interventions could not be explained by our intervention typology, or by variation in potential effect modifiers such as sample demographics, outcome measurement, or trial design. This suggests the need for careful consideration of fit between interventions and local contexts before implementing, and the possibility that explanations for heterogeneity arise from configurations of conditions and components rather than individual predictors. Finally, we adapted the Cochrane risk of bias tool to avoid a floor effect in quality-appraisal ratings across trials. This decision allowed greater comparison of quality across included studies, but the lack of blinding is nevertheless a significant risk of bias in the evidence base.
Implications for Policy and Practice
This is the first published systematic review in this area to compare different intervention types via NMA. Our classification strategy, led by stakeholder consultation suggesting the importance of understanding intervention breadth and difficulty of implementation, led to a surprising finding: that more extensive interventions targeting a broader range of system levels, stakeholders, and change mechanisms were not necessarily more effective than single-component (and frequently technologically mediated) interventions. A possible reason for this relates to school capacity to implement complex interventions,20 such that the effectiveness of single-component interventions may be related to the relative ease of rigorous implementation. This finding may be significant for schools seeking to deliver an intervention for DRV or GBV to students, but with limited resources for complex, multilevel interventions.
Our analysis raises important questions about why interventions might be more effective—and more consistently effective—for DRV. A possible reason for this is that, whereas GBV is immanent and ingrained in school cultures and practices, DRV is a more private behavior3 and potentially more open to change via addressing individual knowledge and attitudes. Similarly, given rapid turnover in adolescent dating relationships, young people may have more opportunities to alter relationship dynamics in ways not present for GBV, given that peer relationships may be less amenable to change.
Future trials and reviews should incorporate outcomes beyond individual behaviors, knowledge, or attitudes. Although these are useful at gauging intervention impacts on individuals, they do not capture the broader system and community effects of an intervention,21 which were not evidenced in our review. In addition, our findings suggest that interventions may require several years of implementation to show meaningful impacts for DRV and GBV. This may be a barrier for many schools given short-term improvement targets. Schools should consider preintervention implementation work to integrate delivery and maintenance of an intervention into existing school practices, and to maximize the public health benefits of implemented interventions.
ACKNOWLEDGMENTS
This study is funded by the National Institute for Health Research (NIHR) Public Health Research Programme (NIHR130144). In addition, V. Berry and G. J. Melendez-Torres are partly supported by the NIHR Applied Research Collaboration South West Peninsula (NIHR PenARC) and Chris Bonell is partly funded by an NIHR senior investigator award.
We gratefully acknowledge the assistance of Fraizer Kiff with study appraisal.
Note. The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care. The funders had no role in the design and conduct of the study.
CONFLICTS OF INTEREST
C. Bonell was the principal investigator, and H. Young and G. J. Melendez-Torres co-investigators, of one of the trials included in this meta-analysis.
HUMAN PARTICIPANT PROTECTION
This research did not require ethics approval as it was based on publicly available data. However, ethics approval from the University of Exeter (ID 488499) was received to access and generate summary descriptive statistics from 2 data sets (ICPSR 22660, ICPSR 36355).
REFERENCES
- 1.Young H, Long SJ, Melendez-Torres GJ, et al. Dating and relationship violence victimization and perpetration among 11–16 year olds in Wales: a cross-sectional analysis of the School Health Research Network (SHRN) survey J Public Health (Oxf )2021431111–122. 10.1093/pubmed/fdz084 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Taquette SR, Monteiro DLM. Causes and consequences of adolescent dating violence: a systematic review. J Inj Violence Res. 2019;11(2):137–147. doi: 10.5249/jivr.v11i2.1061. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Office for Standards in Education, Children’s Services and Skills (Ofsted) 2022. https://www.gov.uk/government/publications/review-of-sexual-abuse-in-schools-and-colleges/review-of-sexual-abuse-in-schools-and-colleges
- 4.Exner-Cortens D, Eckenrode J, Bunge J, Rothman E. Revictimization after adolescent dating violence in a matched, national sample of youth. J Adolesc Health. 2017;60(2):176–183. doi: 10.1016/j.jadohealth.2016.09.015. [DOI] [PubMed] [Google Scholar]
- 5.Exner-Cortens D, Eckenrode J, Rothman E. Longitudinal associations between teen dating violence victimization and adverse health outcomes. Pediatrics. 2013;131(1):71–78. doi: 10.1542/peds.2012-1029. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Earnest AA, Brady SS. Dating violence victimization among high school students in Minnesota: associations with family violence, unsafe schools, and resources for support. J Interpers Violence. 2014;31(3):383–406. doi: 10.1177/0886260514555863. [DOI] [PubMed] [Google Scholar]
- 7.Bendixen M, Daveronis J, Kennair LEO. The effects of non-physical peer sexual harassment on high school students’ psychological well-being in Norway: consistent and stable findings across studies. Int J Public Health. 2018;63(1):3–11. doi: 10.1007/s00038-017-1049-3. [DOI] [PubMed] [Google Scholar]
- 8.Chronister KM, Marsiglio MC, Linville D, Lantrip KR. The influence of dating violence on adolescent girls’ educational experiences. Couns Psychol. 2013;42(3):374–405. doi: 10.1177/0011000012470569. [DOI] [Google Scholar]
- 9.Aragon SR, Poteat VP, Espelage DL, Koenig BW. The influence of peer victimization on educational outcomes for LGBTQ and non-LGBTQ high school students. J LGBT Youth. 2014;11(1):1–19. doi: 10.1080/19361653.2014.840761. [DOI] [Google Scholar]
- 10.Fellmeth GLT, Heffernan C, Nurse J, Habibula S, Sethi D. Educational and skills‐based interventions for preventing relationship and dating violence in adolescents and young adults. Cochrane Database Syst Rev. 2013;(6):CD004534. doi: 10.1002/14651858.CD004534.pub3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Kettrey HH, Marx RA, Tanner-Smith EE. Effects of bystander programs on the prevention of sexual assault among adolescents and college students: a systematic review. Campbell Syst Rev. 2019;15(1–2):e1013. doi: 10.4073/csr.2019.1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Piolanti A, Foran HM. Efficacy of interventions to prevent physical and sexual dating violence among adolescents: a systematic review and meta-analysis. JAMA Pediatr. 2022;176(2):142–149. doi: 10.1001/jamapediatrics.2021.4829. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.De La Rue L, Polanin JR, Espelage DL, Pigott TD. A meta-analysis of school-based interventions aimed to prevent or reduce violence in teen dating relationships. Rev Educ Res. 2016;87(1):7–34. doi: 10.3102/0034654316632061. [DOI] [Google Scholar]
- 14.Stanley N, Ellis J, Farrelly N, Hollinghurst S, Downe S. Preventing domestic abuse for children and young people: a review of school-based interventions. Child Youth Serv Rev. 2015;59:120–131. doi: 10.1016/j.childyouth.2015.10.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Tanner-Smith EE, Tipton E. Robust variance estimation with dependent effect sizes: practical considerations including a software tutorial in Stata and SPSS. Res Synth Methods. 2014;5(1):13–30. doi: 10.1002/jrsm.1091. [DOI] [PubMed] [Google Scholar]
- 16.Higgins JPT, Thomas J, Chandler J, et al. Cochrane Handbook for Systematic Reviews of Interventions. 2nd ed. Oxford, UK: John Wiley & Sons; 2019. [DOI] [Google Scholar]
- 17.Meiksin R, Crichton J, Dodd M, et al. A school intervention for 13- to 15-year-olds to prevent dating and relationship violence: the Project Respect pilot cluster RCT. Public Health Res. 2020;8(5):1–338. doi: 10.3310/phr08050. [DOI] [PubMed] [Google Scholar]
- 18.Olsen EOM, Vivolo-Kantor A, Kann L. Physical and sexual teen dating violence victimization and sexual identity among US high school students, 2015. J Interpers Violence. 2020;35(17–18):3581–3600. doi: 10.1177/0886260517708757. [DOI] [PubMed] [Google Scholar]
- 19.Blondeel K, de Vasconcelos S, Garcia-Moreno C, Stephenson R, Temmerman M, Toskin I. Violence motivated by perception of sexual orientation and gender identity: a systematic review. Bull World Health Organ. 2018;96(1):29L–41L. doi: 10.2471/BLT.17.197251. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Moore GF, Evans RE, Hawkins J, et al. From complex social interventions to interventions in complex social systems: future directions and unresolved questions for intervention development and evaluation. Evaluation. 2019;25(1):23–45. doi: 10.1177/1356389018803219. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Burnham J, Banyard V, Ast RS, Edwards KM. Case study of community-level domestic and sexual violence prevention: using concept mapping to evaluate community narratives over time J Fam Violence. 202237143–57. 10.1007/s10896-021-00296-z35342224 [DOI] [Google Scholar]