Background

Worldwide, more than one million cases of sexually transmitted infections (STIs) occur daily and 500 million people live with curable STIs including Chlamydia, Gonorrhoea, Syphilis and Trichomoniasis [1]. The burden of STIs is high in sub-Saharan Africa (sSA) with an incidence rate of 241 per 1000 among adults age 15–49, one of the highest in the world [2]. Approximately 70 % of those living with Human Immunodeficiency Virus (HIV) worldwide, and 80 % of infected women aged 15–24, live in sSA where one in six adolescent deaths is attributed to HIV [3, 4]. Approximately half of new HIV infections occur in individuals aged 15–24 [5] and more than 90 % are sexually transmitted. Thus, sexually active young people in sSA, including young women, are at high risk of HIV infection.

Several types of interventions have been employed to reduce the vulnerability of adolescents to STIs, including HIV. These include: preventive education in schools; services delivered in youth centres, including condom distribution; adolescent-friendly health centres that encourage utilisation of prevention services; school-based health services; conditional cash transfers to encourage young people to remain in school or to avoid risky sexual behaviours; various community-based interventions; and unconditional cash transfers [6]. School-based sexual health education (SBSHE) is arguably the most inclusive and potentially comprehensive of these approaches and has the potential to effectively promote population-level sexual health among adolescents and young adults [7], so reducing the spread of STIs, including HIV [8]. Mavedzenge et al. [6] found high quality evidence for effectiveness of SBSHE in relation to a number of STI-related outcomes from evaluations worldwide and recommended such interventions be implemented widely. However, evaluations of SBSHE interventions in sSA have provided mixed findings in relation to reduction of self-reported unprotected sexual intercourse and surprisingly, none has provided evidence of reduction of STI incidence or prevalence [810]. In this review, SBSHE was defined as any intervention delivered wholly or partially in a school setting aiming to reduce risky sexual behaviours, STIs and unplanned pregnancy, and promote positive sexual health.

Four reviews of evaluations of SBSHE interventions in Africa were found [810]. None assessed effects of interventions on STI incidence or prevalence. A review by Kaaya et al. [9] included 11 interventions and concluded that most studies reported effects on knowledge, attitudes and communication but fewer reported effects on self-reported onset of sexual intercourse, frequency of sexual intercourse and number of sexual partners. Similarly, a review by Gallant and Maticka-Tyndale [8] also including 11 interventions and concluded that knowledge and attitudes are easier to change than behaviours among African youths. These reviewers recommended that intervention design should be grounded in theorized accounts of behaviour change mechanisms and be directly relevant to local culture. A third review by Paul-Ebhohimhen et al. [10], including 10 evaluations, also concluded that SBSHE interventions had stronger effects on sexual health knowledge and attitudes than on behaviour patterns. More optimistically, a review of seven interventions for Nigerian students found changes in self-reported sexual behaviour patterns including delaying sexual debut, increasing condom and other contraceptives use and reducing frequency of sexual activity [11]. Other reviews include SBSHE interventions [5, 1226] but draw no conclusions about SBSHE in sSA.

The present study

This review extends previous reviews of effectiveness of SBSHE interventions in sSA [810] in four key respects. First, we include an updated and more comprehensive set of experimental evaluations. Second, we assess intervention effects on reduction of STIs indexed by biological markers. Third, we explore intervention characteristics that may differentiate between effective and ineffective interventions. Fourth, we examine assessment of implementation fidelity.

The review addressed three research questions:

  1. 1.

    How effective are school-based sexual health education interventions in sub-Saharan Africa in promoting condom use and preventing sexually transmitted infections?

  2. 2.

    What characterizes effective school-based sexual health education interventions implemented in sub-Saharan Africa?

  3. 3.

    Are school-based sexual health education interventions implemented with fidelity in sub-Saharan Africa?

Methods

This review was conducted according to a protocol [see Additional file 1, for the review protocol], and reported in accordance to PRISMA statement [27].

Inclusion criteria

Studies were included if they met the following criteria:

  • i. Published in English before March 2016.

  • ii. The study was a randomised control trial (RCT) or quasi-experimental (non-randomised trials and before-and-after studies with comparison groups).

  • iii. More than 80 % of participants were below the age of 25 years. A 25 year cut-off was applied because age of school enrolment varies considerably across sub-Saharan Africa, particularly in the rural areas. Hence, it is not uncommon to find older students in primary or secondary schools [28].

  • iv. The study evaluated a school-based sexual health education intervention delivered in sub-Saharan African schools. The intervention could be delivered completely in school or include components delivered to school students outside school and/or outside school hours.

  • v. The dependent measure was self-reported condom use and/or levels of STIs.

Exclusion criteria

Studies were excluded for the following reasons:

  • i. They employed no comparison or control group.

  • ii. They employed a comparison group that received sexual health education other than the usual curriculum.

  • iii. They were delivered in universities.

  • iv. Twenty percent or more of the participants were aged 25 years and above.

  • v. Knowledge, attitudes and behavioural intentions were the only outcomes reported.

Search strategy

Ten electronic databases including Medline, PsycInfo, EMBASE, CINAHL, Web of Knowledge, The Cochrane Library, British Education Index/EBSCOhost, Australian Education Index/ProQuest, Education Research Complete/EBSCOhost and ERIC/ProQuest were searched in February 2015 (see Additional file 2: Table S2, which contains search strategy for Medline that was modified and used in other databases). One new inclusion was identified in an updated search run in March 2016. We also searched the table of contents of Journal of AIDs, AIDs Care, AIDs and Behaviour, AIDs Education and Prevention, Journal of Adolescent Health, and Journal of Youth and Adolescence for relevant studies. Reference lists of similar reviews and included studies were also searched in an iterative fashion until no new article was found.

Study selection

Titles and abstracts of the 21,634 identified studies were screened by the primary reviewer (SAS) with a random selection (n = 500, 2.3 %) screened by a second reviewer (SD). Full texts of articles that passed the title-abstract stage were obtained for full text screening. All the full text articles were screened by SAS and 20 % (n = 53) randomly selected were screened independently by SD. Gwet’s [29] AC1 statistic was used to assess the inter-rater reliability at each stage of the screening and any disagreement was resolved through discussion. The opinion of a third reviewer (CA) was sought when, exceptionally, two reviewers (SAS and SD) were unsure how to resolve disagreements.

Data extraction

We extracted data relevant to the review questions, including study design, descriptions of the interventions, theories informing intervention design, features of effective interventions, descriptions of implementation and outcomes categorised by length of follow-up. The data extraction form is available as a Additional file 3: Table S3. Where more than one article described the same intervention, data were extracted from all papers. Data was extracted by the primary reviewer (SAS) and independently by a second reviewer (SD) to check accuracy. Furthermore, a statistician (SB) also extracted quantitative outcomes data included in meta-analysis independently.

Quality assessment of included studies

The quality of the included studies was assessed using four main dimensions based on recommendations in the Cochrane Collaboration Tool for Assessing Risk of Bias [30], namely selection bias, performance bias, detection bias, and attrition bias of the included studies. The Cochrane Collaboration Tool was used to assess the quality of included interventions because it is a domain-based evaluation that gives critical assessment of each domain (dimension) in which bias may arise [30]. It has the advantage of encouraging users to tailor how they assess studies and so adds to transparency unlike some other methodology assessment checklists (e.g., Jadad [31]). Selection bias for non-RCTs was assessed by determining selection bias due to confounding as recommended in the Cochrane Collaboration Handbook [30]. Random sequence generation or allocation concealment (or bias due to confounding for non-RCTs) and incomplete outcome data were considered critical for assessing the quality of studies in this review. The critical dimensions were used to score the overall risk of bias of the included studies. Random sequence generation and allocation concealment were scored as one dimension assessing selection bias. A score of two was given for ‘high’, one for ‘unclear’ and zero for ‘low’ risk of bias. Therefore, an intervention can have an overall score ranging from zero to four. An article with a total score of 3 or 4 was assessed as ‘high’, 2 as ‘moderate’ and 0 or 1 as ‘low’ risk of bias.

Data analysis

Review Manager 5.3 software [32] was used to undertake meta-analyses identifying intervention effectiveness in relation to increased condom use and reduced HSV2 infections using outcome measures closest to the median follow-up period. Separate analyses for condom use were conducted dividing evaluations into those with short (less than 6 months), intermediate (6–10 months, based on a median of 8 months) and long-term follow-up (more than 10 months). Random-effects method of meta-analysis that is based on inverse-variance technique that adjusts for varying study weights and heterogeneity was employed [30] because of variability in trial size and intervention content, intensity and duration. Whenever available, adjusted (for baseline) rather than crude odds ratios (OR) were used in the analyses. Heterogeneity across estimates was quantified using the I-squared statistic (I2) and the p-value from the chi-squared test for heterogeneity was used to quantify evidence against homogeneity [30]. We did not include studies in meta-analysis if heterogeneity was high (I2 of 75 % and above). Those studies that provided insufficient data to include in the meta-analyses were reported descriptively.

We also conducted sensitivity, or sub group, analyses to assess the effects of two study characteristics on effectiveness, namely, (i) the measure of condom use employed (condom use at last sex versus other measures) and (ii) use of crude versus adjusted odds ratios.

Quality of evidence

We used “Grading of Recommendations Assessment, Development and Evaluation” (GRADE) [33] to assess the overall quality of evidence reported in studies included in meta-analyses. This approach provides a structured and transparent way of developing and presenting results summaries that are easy to understand [33]. Five criteria were used in grading the evidence including limitations of design (risk of bias), inconsistency (heterogeneity), indirectness, imprecision, and reporting or publication bias. For limitations of design (risk of bias), the quality was downgraded if most of the studies were of high risk of bias as assessed with the Cochrane Collaboration Tool. For inconsistency, unexplained heterogeneity indicated by I2 of more than 75 % was used to downgrade the quality of evidence for this criterion. Indirectness was assessed by determining how closely the interventions, participants and measures of outcome of interest were similar across studies. A relative risk reduction or increase of greater than 25 % (±0.25) as suggested by GRADE was used to downgrade the quality of evidence for imprecision. Finally, visual inspection of asymmetry of funnel plots was used to detect the possibility of publication bias, and quality was downgraded if asymmetry was observed. These assessments were undertaken using GRADEpro software [34] and a summary of findings table (SoF) generated. The overall quality of each outcome was graded as ‘high’, ‘moderate’, ‘low’ and ‘very low’. These are interpreted as ‘very confident’, ‘moderately confident’, ‘limited confidence’ and ‘very low confidence’ that the true effect lies close to the estimated effect respectively [33].

Features of effective interventions

Intervention design and implementation characteristics associated with effectiveness have been identified in previous reviews. Design related features include: needs assessment with the intended participants and involving key stakeholders in the design or development process of the intervention [5, 10]; adapting the intervention or curriculum from other evaluated interventions [5]; basing the intervention on behavioural change theory [9]; and providing the participants with skills that help reduce risky sexual behaviours [10). Implementation characteristics include: provision of adolescent-friendly health services [5]; distribution of condoms [5]; extending activities to the community outside the school environment [5]; training of facilitators; and fidelity of delivery [10]. Intervention descriptions in the current review were coded for inclusion of these features. We classified interventions as “interventions with evidence of benefit” or “interventions without evidence of benefit”. “Evidence of benefit” was identified as a statistically significant increase condom use or less prevalence/incidence of STI at any follow-up among any sub-group of the participants (e.g., among sexually active participants). The frequency of occurrence of potentially effectiveness enhancing features among the interventions with evidence of benefit and those without evidence of benefit was then determined.

An intervention was regarded to have been delivered with fidelity if the authors reported that the intervention was delivered as intended. This includes how “faithful” components, content, and activities of the intervention were delivered as designed. It also includes frequency and duration of exposure (intensity) of the intervention.

Results

Selection and description of studies

We obtained 21,634 papers after removing duplicates (Fig. 1), out of which 271 were selected after screening through titles and abstracts (AC1 = 0.98). After full-text screening two reviewers (SAS and SD) independently screened and agreed (100 % agreement, AC1 = 1.0) on inclusion of 51 papers, reporting on 31 interventions. The Mema Kwa Vijana (MkV) intervention was reported in six included papers [3540], however, Ross et al. [35] is cited henceforth when referring to this intervention because most data were extracted from that report. Similarly, other interventions reported in more than one paper include: (i) Primary School Action for Better Health (PSABH) [41, 42]. (ii) HealthWise South Africa [4346]. (iii) Promoting Sexual and Reproductive Health, School-based HIV/AIDS Intervention in Sub-Saharan Africa (SATZ) [4750]. (iv) HIV Prevention Intervention for Rural Youth (HP4RY) [5154]. (v)‘Let Us Protect Our Future’ South Africa [5557]. (vi) The Regai Dzive Shiri Project [5861]. Subsequently, key papers ([41, 43, 47, 51, 55, 58] respectively) are cited when referring to these interventions.

Fig. 1
figure 1

Studies selection flow diagram

Seventeen of the evaluations were RCTs and 14 used quasi-experimental designs. Twenty-six of the interventions were delivered in secondary or high schools (84 %), four (13 %) in primary or elementary schools and SATZ intervention in both primary and secondary schools (3 %). Four interventions [35, 51, 58, 62] included out-of-school and community activities, some involving health centres. The MkV intervention [35] had four components namely, a teacher- and peer-led in-school programme; provision of youth-friendly health services; condom promotion and distribution; and community mobilisation activities. The HP4RY intervention [51] had two components, a teacher delivered school programme and a community programme delivered by young people. The Regai Dzive Shiri Project [58], had three components, teacher-delivered school activities, community activities targeting parents and other community members, and provision of reproductive health services. The intervention by Brieger et al. [62] was a peer-led activity involving in-school activities as well as visits to, clinics and community centres activities. All interventions included in this review had both female and male participants, and participants were 9–30 years old [55, 63]. The number of participants varied from 24 [64] to 13814 [35], with a total number 70201 across all included evaluations. The median follow-up period for condom use was 8 months and 54 months for the biological outcomes. See Additional file 4: Table S4, which provides lists of excluded studies with reasons for the exclusion and Table 1, which provides the characteristics of the included studies. Intervention descriptions included in Table 1 are summaries of those provided in the papers describing included interventions. All studies reported on ethical approval and/or received informed consent from the participants.

Table 1 Characteristics of Included Studies

Methodological quality of included studies

Methodological quality was generally high; 20 of the included studies were assessed as having “low”, 8 as “moderate” and 3 “high” risk of bias (see Additional file 5: Table S5, which contains the quality assessment process). Two of the high risks of bias studies [43, 65] were found to be at risk of attrition bias due to loss to follow up of more than 30 % and ‘unclear’ selection bias. The other high-risk study [66] was assessed to be at risk of selection bias because the baseline characteristics of confounders differed between the two arms of the intervention, which were not adjusted for in the analysis, and ‘unclear’ attrition bias. See Figs. 2 and 3 for risk of bias graph and risk of bias summary for each study respectively.

Fig. 2
figure 2

Risk of bias graph

Fig. 3
figure 3

Risk of bias summary

Description of interventions

All included interventions delivered comprehensive sexual health education in classroom settings (see Table 1 for interventions description). Comprehensive SBSHE provides participants with information on transmission of sexual infections, safer sex practices and prevention of STIs and unwanted pregnancies [67], in contrast to abstinence-only interventions. Various intervention delivery methods were employed in one or more combinations some of which include lectures or presentations (n = 6), group discussions (n = 14), role-plays or dramas (n = 14), and distribution of information, education and communication (IEC) materials (n = 6). Films shows or audio vignettes (n = 7), songs (n = 4), counselling (n = 6), quiz and essay competitions (n = 7) were also used. Condoms were distributed in three interventions [35, 62, 66]. Two were interactive computer-based programmes [68, 69] and one only involved provision of a printed material [65].

The dose and duration of the interventions varied widely and ranged from a single 1 h and 45 min [70] session to multiple sessions delivered over 36 months [35]. However, in general, the interventions employed one session per week of 30–60 min duration over a period of 6–12 weeks [64, 68, 7176]. Fourteen theories were said to inform the design of 16 of the interventions with Social Learning Theory (n = 6), Social Cognitive Theory (n = 3) and Theory of Planned Behaviour (n = 3) being most frequently used.

Implementation details

Seven of the interventions were delivered by both teachers and peer educators [35, 41, 51, 66, 69, 73, 75]; similarly, seven by peer educators [62, 70, 7781]; and three by teachers only [47, 72, 82]. Health educators, community physicians, volunteer adults or youths, nurses or other health personnel were involved in delivering six of the interventions [43, 58, 63, 71, 74, 83] and one was delivered by the researchers that developed it [84]. Twenty-one of the 31 interventions reported that the facilitators received some form of training (see Additional file 6: Table S6, which contains implementation details of the included interventions).

Only twelve (of 31) studies reported monitoring of implementation and only seven of these [35, 47, 58, 68, 69, 75, 82] reported on fidelity of implementation. Just two studies [35, 75] reported that fidelity had been achieved and explained how fidelity had been assessed. In other cases lack of compatibility with local circumstances undermined fidelity of delivery. For example, in the intervention evaluated by Mathews et al. [47] some teachers did not implement condom demonstrations and other skilled-based activities due to overwhelming large number of students per class. Similarly, in the intervention evaluated by Rijsdijk et al. [69] poor availability of computers meant that the intervention had to be modified to delivery through print materials. Sub-sample analyses in this evaluation showed that schools with ‘complete’ implementation had most of the significant positive effects compared to those with ‘partial’ implementation [69]. Complete implementation schools are those where the teachers fully implemented more than 50 % of the 14 lessons in the programme.

Outcomes

Three studies reported STI outcomes [35, 55, 58]. Two [35, 58], measured HIV infections close to the median follow-up period of 54 months. Cowan et al. [58] found no evidence of an effect on HIV infections among males or females (adjusted odds ratio (aOR) = 1.20, 95 % CI = 0.66–2.18 and aOR = 1.15, 95 % CI = 0.81–1.64 respectively). Ross et al. [35] reported incidence rate per 1000 person-years and the intervention also did not significantly reduce HIV infection risk for both short (adjusted rate ratio (aRR) = 0.75, CI = 0.34–1.66 for young women) and long-term follow-up periods (adjusted prevalence rate (aPR) = 0.91, CI = 0.50–1.65 for men and aPR = 1.07, CI = 0.68–1.67 for women).

Three studies [35, 55, 58], also measured HSV2 infections for median follow-up period of 54 months. SBSHE showed no statistically significant effect in reducing the risk of this infection (OR = 1.07, 95 % CI = 0.94–1.23, p = 0.31) (Fig. 4 Panel a). Ross et al. [35] also did not find any significant effect at long-term follow-up (aPR = 0.94, CI = 0.77–1.15 for males and aPR = 0.96, CI = 0.87–1.06 for females) and similarly, the intervention by Jemmott III et al. [55] did not find a significant effect at 42-month follow-up period.

Fig. 4
figure 4

Forest plots for meta-analysis. a HSV-2 infections. b Condom use for less than six months follow-up period (RCTs). c Condom use for less than six months follow-up period (non-RCTs). d Condom use for six to ten months follow-up period (RCTs). e Condom use for six to ten months follow-up period (non-RCTs). f Condom use for more than 10 months follow-up period (RCTs). g Condom use for more than 10 months follow up period (non-RCTs)

Ross et al. [35] measured other STIs including Syphilis, Chlamydia, Gonorrhoea and Trichomonas and found no statistically significant difference between the intervention and control group in the prevalence of these infections for both short and long-term follow-up periods. However, the intervention by Jemmott III et al. [55] significantly reduced curable STIs (Chlamydia, Gonorrhoea and Trichomonas) at 42-month follow-up period (OR = 0.71, 95 % CI = 0.54–0.95), but not at 54-months follow-up (OR = 1.15, 95 % CI = 0.84–1.57).

All 31 studies assessed self-reported condom use. Fifteen of the interventions [35, 41, 43, 51, 55, 62, 64, 71, 72, 75, 77, 79, 82, 83, 85] resulted in statistically-significant increases in condom use while 16 showed no statistically-significant increases in condom use [47, 58, 63, 65, 66, 6870, 73, 74, 76, 78, 80, 81, 84, 86]. No intervention resulted in statistically significant reductions in self-reported condom use.

Twenty of the studies that measured condom use provided adequate data to enable inclusion in meta-analyses [35, 41, 47, 51, 55, 58, 63, 66, 68, 70, 73, 74, 76, 77, 7981, 83, 85, 86]. Measures of condom use at last sex [35, 41, 47, 51, 58, 63, 66, 70, 73, 74, 7981, 83, 85, 86], consistent condom use in the last 12 months [77], condom use [76], condom use in the past three months [55] and 100 % condom use in the last three months [68] were in the meta-analysis. All of these measures were coded by the original authors as dichotomous use/non-use scores. For short-term follow-up of less than 6 months, intervention participants were more likely to report condom use in both RCTs (OR = 1.62, 95 % CI = 1.03–2.55, p = 0.04) (Fig. 4 Panel b) and non-RCTs (OR = 2.88, 95 % CI = 1.41–5.90, p = 0.004) (Fig. 4 Panel c). Similarly, intervention participants were more likely to use condoms at intermediate follow-up of 6–10 months with the effect being statistically significant for the RCTs (OR = 1.40, 95 % CI = 1.16–1.68, p = 0.0004) (Fig. 4 Panel d) but not for non-RCTs (OR = 1.05, 95 % CI = 0.65–1.71), p = 0.84) (Fig. 4 Panel e). At follow-up of more than 10 months, the effect was marginally significant for RCTs (OR = 1.22, 95 % CI = 0.99–1.50, p = 0.06) (Fig. 4 Panel f) and non-significant for non-RCTs (OR = 1.18, 95 % CI = 0.92–1.52, p = 0.20) (Fig. 4 Panel g).

Sub group analyses showed no effect of type of condom use measure on condom use but greater condom use effects when crude, compared to adjusted ORs, were employed (see Additional file 7: Table S7, Panel A-H).

All authors were emailed to acquire missing data. Nonetheless, three studies [43, 71, 82] were excluded from the meta-analysis because they reported only ORs without CI, standard error of mean or p-value, precluding further analysis. Another study [78] was also removed because the follow-up period was unclear. Another seven studies [62, 64, 65, 69, 72, 75, 84] measured condom use using continuous measures composed of differing items and could not be included. James et al. [82] measured consistent use of condom in the preceding 6 months by assessing whether a condom was used during all intercourse instances, sometimes or not at all. Mba et al. [84] assessed whether participants practised or intent to practise condom use, as a STI-prevention technique. Esere [64] used a 4-point Likert At-Risk Sexual Behaviour Scale which includes ‘do not use condoms while having sex’ as a component of the scale. The Ugandan study ‘The World Starts with Me’ used a 4-item condom use performance behaviour scale to measure condom use [69] and Burnett et al. [72] measured condom use using a 14-item scale. Frequency of condom use was measured on a scale of one (never) to six (always) in the study by Karnell et al. [75]. Finally, Brieger et al. [62] measured contraceptive information including condom use, pills and foaming tablets. This variability of outcome measures prevented the inclusion of these studies in our meta-analysis. Seven of the studies [43, 62, 64, 71, 72, 75, 82] not in the meta-analysis found statistically significant results in increased condom use in the intervention group compared to the control group (see Table 2 which contains results and scales used to measure condom use for studies not included in the meta-analysis).

Table 2 Studies That Reported Condom Use Not Included In Meta-Analysis

Quality of evidence and summary of findings

Table 3 shows the summary of findings and quality of evidence for outcomes included in meta-analyses. The quality of evidence for HSV-2 infection, condom use for 6–10 months and more than 10 months follow-up among RCTs is ‘high’, which means we are very confident that the true effect lies close to the estimate. We are moderately confident in the evidence for self-reported condom use for less than 6 months follow-up among the RCTs. For the remaining outcome categories, we have limited to very low confidence in the proximity of the estimates to the true effects.

Table 3 Quality of evidence and summary of findings table

Features of effective interventions

Small study samples mean that interpretation of the distribution of characteristics across interventions that did or did not result in increased condom use can only be tentative (see Table 4). Nonetheless, we can observe that effective interventions were more often adapted from other programmes, were theory-based, included provision of health services, included activities outside school and were implemented with fidelity.

Table 4 Frequencies of occurrence of features associated with effectiveness

Discussion

We conducted a comprehensive review of school-based sexual health education interventions in sub-Saharan Africa evaluated using experimental or quasi-experimental methods. Given the need for public health interventions to reduce sexually transmitted infections, including HIV, and the potential effectiveness of school-based sexual health interventions, the most striking finding is paucity of published evaluations. Across 31 interventions meeting our inclusion criteria, we found no evidence of effectiveness in reducing STIs, including HIV, although one study [55] reported a reduction in curable STIs (Chlamydia, Gonorrhoea and Trichomonas) at one follow-up period. We also found no evidence of harm. This mirrors the findings of previous, comprehensive reviews [22, 25, 87, 88]. More rigorous evaluations of best practice, sustainable, school-based sexual health programmes in sub-Saharan Africa are needed.

Interventions were effective in increasing self-reported condom use and, encouragingly, the positive effect on condom use was stronger among evaluations employing more robust experimental methods (RCTs) for intermediate and long-term follow-up periods. Previous reviews [9, 11, 26] have also found methodically stronger studies to be associated with stronger effects. Perhaps, unsurprisingly, short and intermediate (versus, long follow-up periods) yielded greater condom use gains, suggesting that further intervention may be needed to sustain behaviour change [11]. These findings contrast with previous suggestions that SBSHE in sSA has a poor record of changing sexual behaviours including condom use [810] and supports further investment in SBSHE to promote condom use in low-and middle-income countries [5, 25].

Tentative consideration of characteristics found in interventions that did or did not result in statistically-significant increases in condom use recommends that intervention designers should consider adapting interventions from pre-existing effective programmes, base their interventions on theory-based logic models of mechanism and link them to health service provision including condom distribution. Finally, intervention designers need to ensure that they assess fidelity and take steps to ensure that interventions are delivered as designed. In this review, just two studies [35, 75] reported on fidelity of delivery. It was impossible, therefore, to determine whether or not the interventions were delivered as intended and whether this determined effectiveness. This review highlights the need for further rigorous evaluations of SBSHE to assess impact on incidence or prevalence of STIs including HIV. In addition, future evaluations need to assess and report on implementation processes including fidelity. This will provide better guidance on how and why interventions ‘work’ or ‘do not work’.

Limitations in the available data recommend caution in interpretation of our findings. For example, condom use reporting is subject to social desirability bias and recall bias, although guidance is available on measures that may minimise such bias [89]. Greater consistency in use of best measures of condom use would assist data synthesis, although sensitivity analysis did not reveal differences in effectiveness as a result of the self-report measures used. Ideally, a larger sample of studies would have been available and further moderators of effectiveness could have been considered. In particular, we would have liked to report on whether interventions in primary or secondary schools were more or less effective and whether studies with greater or lesser risk of bias tended to result in greater increases in condom use. However, for both these sub-group analyses multiple cells included just one study across follow up points. The four evaluations of primary school interventions [35, 41, 55, 71] suggest that these can be just as effective as secondary school interventions and this may indicate that early school-based intervention is likely to be more effective. Similarly, we would have liked to assess whether effects were greater or lesser for young men and women but only four studies [35, 41, 58, 79] included in the meta-analysis presented separate gender analyses. It is worth noting too that because blinding is impossible in relation to school-based sexual health education, we were only able to employ two critical, of four dimensions of the Cochrane Collaboration Tool assessed to score the overall risk of bias of the studies in this review. Reviewing studies published in English may have limited our sample. In addition, although double screening of 500 randomly selected title and abstract entries showed near perfect agreement between two reviewers (generating an ACI score of 0.98) further double screening at this stage could have been conducted.

Conclusion

We conducted a review of school-based sexual health education interventions in sub-Saharan Africa. Interventions to safeguard adolescents from sexually transmitted infections, including HIV are especially needed in sub-Saharan Africa. School-based interventions have the potential to be inclusive and to provide comprehensive preventive education and training. We assessed the impact of such interventions on incidence or prevalence of STIs and self-reported condom use. We also identified characteristics associated with effective interventions. We found no effect of the interventions on STIs, however, some positive effect on condom use was observed. Certain features present among interventions effective in promoting condom use were observed. Despite limitations, our review indicates that school-based sexual health education may be an effective strategy to promote condom use among sub-Saharan African adolescents over periods of up to 10 months. Interventions may be optimised by including features found in previous effective programmes. Above all, this review highlights the need for further rigorous evaluations of school-based sexual health education interventions in sub-Saharan Africa including assessment of infection prevalence and fidelity of delivery. Guidance on reporting implementation processes including fidelity would be helpful to intervention designers.