To find studies presenting evidence on determinants of in-school and out-of-school adolescent sexual behavior in SSA, the following databases were consulted: Embase, PubMed, PsycInfo, SCOPUS, AJOL, and Google Scholar. Search equations were built using descriptors and keywords related to adolescent, sexual behavior, determinant, and sub-Saharan Africa, organized with Boolean terms. Results were imported into Rayyan, a web-based tool for systematic literature review. After eliminating duplicate references, two independent reviewers (R.K. and S.A.) screened articles based on predefined eligibility criteria in two rounds: first by titles and abstracts, then by full texts. Database searches were performed on April 24, 2023, and on Google Scholar only on April 28, 2023. Reference lists of included studies were scanned to identify additional studies and ensure literature saturation.

Studies were included in the systematic review if they: (1) were conducted in sub-Saharan countries; (2) focused on in-school or out-of-school adolescents aged 10–19 years (as defined by the United Nations (UN)), whereby the adolescent subpopulation had to be the majority (over 50%); (3) used quantitative methodology; (4) investigated determinants of various sexual behaviors; (5) were published between January 2010 and April 2023 (2010 being 1 year after the publication of the first edition of UNESCO’s International Technical Guide on Adolescent Sexuality Education, a crucial moment in the recognition of adolescents' right to sexuality education); (6) were published in English or French.

Studies were excluded when they: (1) failed to meet one or more of the above inclusion criteria; (2) lacked detail to verify majority of participants were 10–19 years old; (3) did not specify if participants were in-school or out-of-school; (4) did not explicitly address determinants of sexual behavior; (5) used only qualitative data. Conference abstracts, commentaries, and letters that did not contain sufficiently detailed information about the study were also excluded, as were studies for which the full text could not be retrieved, even contacting the corresponding author.

To assess the methodological quality and risk of bias of the studies retained for the review, two independent assessors (R.K. and S.A.) scored all articles using the Joanna Briggs Institute (JBI)’s critical appraisal tool for analytical cross-sectional studies [11]. The JBI checklist considers 8 quality criteria: (1) definition of inclusion criteria, (2) description of study subjects and the setting, (3) valid and reliable measurement of the exposure, (4) objective criteria for measuring the condition, (5) identification of confounding factors, (6) strategies for dealing with confounding, (7) valid and reliable measurement of outcomes, and (8) appropriate statistical analysis. To assess the quality of each study, an overall score was calculated expressed as a percentage, with a score of 70%–100% indicating high methodological quality, a score of 50%–69% moderate methodological quality, and a score below 50% poor methodological quality with a high risk of bias. The quality assessment was used for indicative purpose only, meaning that no studies were excluded based on this assessment.

Data extraction was performed by a single reviewer (R.K.) using a standardized and pretested extraction form developed collaboratively by the research team. This form clearly defined data fields to ensure consistency in recording key study characteristics and findings. In addition, any uncertainties were resolved through discussion with the co-authors.

Due to the heterogeneity of definitions and measurements of sexual behaviors and their determinants across the included studies, a meta-analysis was not feasible. Instead, we conducted a narrative synthesis using a qualitative content analysis approach [12]. To reduce potential subjectivity, key decisions such as grouping different terminologies referring to the same sexual behavior or category of determinant were reviewed and validated in consultation with a second author (SV). For each category of sexual behavior, we listed the identified determinants and retained those most frequently reported across studies. Bronfenbrenner’s ecological model was applied deductively: determinants were categorized according to predefined ecological levels based on their nature and the context in which they were described. When difficulties or overlaps in classification arose, decisions were discussed among co-authors to ensure clarity and coherence in the allocation. Given the strong interconnections between determinants classified under the exosystem and the macrosystem, we decided to merge these categories and report all such determinants under the macrosystem level.

The 132 studies included for this review covered 28 countries in Sub-Saharan Africa. Six studies contained data from several SSA countries, the other 126 studies had data from one country. Of these, a majority (over 70%) were concerned with one of three countries: Ethiopia (n = 60), Nigeria (n = 21), and South Africa (n = 17). Most studies involved adolescents from secondary or preparatory schools, while one study focused on university adolescent, two on out-of-school adolescents, and two on a mix of in- and out-of-school adolescents. All studies were cross-sectional, employing quantitative data collected via validated and non-validated questionnaires. Among them, 104 investigated determinants of risky sexual behaviors (RSB), and 28 investigated determinants of sexual violence (SV).

The overall quality of the included studies was medium to strong, according to the assessment using the JBI critical appraisal tool. More precisely, 45 studies scored above 70% for overall quality, 75 studies scored between 50% and 70%, and 12 studies scored below 50%. The main limitations to quality were a lack of clear information on participant inclusion, a lack of identification of confounding factors, a lack of an explicit strategy for dealing with the effect of confounding factors, and a lack of validation of the questionnaires that were used.

A first series of findings concerns the main determinants of sexual behaviors that can be considered as risky, due to their potential to increase the risk of STIs or unintended pregnancies among adolescents. Risky sexual behavior includes sexual intercourse, premarital sex, early sexual intercourse, having multiple sexual partners, not using contraceptives, and not using condoms. Because of the interconnectedness of these risk behaviors, most studies used measures that considered all these types of RSB together. Of the 104 included studies, 101 involved school-going adolescents, one involved out-of-school adolescents, and two involved both.

Determinants of RSB are factors that can either protect against RSB or increase the probability of RSB. Based on Bronfenbrenner’s ecological systems theory, they can be categorized as related to the adolescent’s ontosystem, microsystem, mesosystem, and macrosystem (Table 1).

At the ontosystem level, determinants include demographic characteristics, socio-cognitive factors, behaviors, and exposure to specific experiences. Findings for several sociodemographic characteristics are ambivalent. Being male [22, 31, 32, 41, 50, 51, 54–66] or female [13, 23–27, 33, 45, 46, 53, 67–69], age [16–18, 21, 22, 33–39, 42, 43, 45, 47, 51, 57, 58, 60, 61, 63–66, 70–78], ethnicity [39, 49, 50], grade level [14, 19, 22, 23, 31, 38, 53, 55, 78, 80–84], and urban residence [31, 43, 50, 53, 85] are identified both as factors increasing and reducing the probability of RSB. Being single [13–15], religious [15–20], and having parents with a high educational level [22–30] are protective factors against RSB. Conversely, rural residence [22, 36, 37, 46–49], living without families [23, 25, 41–43, 50–52], alone [51, 53], or with a single parent [54] increase the probability of RSB. Good knowledge of STIs, HIV, and safe sexual practices is mostly protective [17, 28, 42, 56, 64, 82, 86, 87], except in one study [73]. Self-efficacy in adopting safe sexual behaviors [13, 18, 53, 79], empowerment in safe sexual behaviors [88], a positive attitude toward safe sexual behaviors and abstinence [13, 86, 88], and a high perception of risks associated with RSB [18, 70] are protective factors. Perceived peer pressure generally increase the probability of RSB [16, 23, 26, 36, 38, 41, 43, 47, 48, 50, 55, 56, 58, 61, 62, 67, 82, 83, 89, 90], except in one study [18]. Perceived gender norms [58, 70], perception of social norms associated with the sexual behavior of peers and family [17, 47, 61, 70, 91], a low perception of social norms regarding RSB and abstinence [64, 73, 87, 92], a low perception of risks associated with RSB [21, 79], a low intention to adopt safe sexual behaviors [73], and a positive attitude towards RSB increase the probability of RSB [93]. Behaviors such as substance use [55, 57], [50, 58–61], [31, 32], [22, 65, 66], [13, 23–26], [46], [45, 67, 68], [16, 19, 36, 71, 74, 81, 83, 84], [28–30, 48], [52, 86, 89, 90, 92], [40, 44, 94–98], except in one study [34], pornography use [14, 16, 22, 27–29, 32, 36, 40, 41, 46, 48–51, 67, 68, 80, 82, 90, 98, 99], internet use [33, 89, 95], sexting [99], partying [37, 68, 89, 100], violent behavior [20, 101], and truancy [84] increase the probability of RSB. Experiences of sexual harassment [76, 79], sexual violence [76], domestic violence [23], parental neglect [52], and knowing someone who died of HIV [79] also increase the probability of RSB.

At the microsystems level, family, peers, relationships, and school factors influence RSB. Discussions on sexual and reproductive health (SRH) issues with family are mainly protective against RSB [26, 27, 49, 56, 80, 98], except in one study [47]. Discussions with peers can either increase or reduce the probability of RSB [18, 22, 51, 62, 70, 81]. Good family connectedness [19, 25, 44, 58, 84, 87, 91, 104], parental monitoring [26, 38, 94, 104], family support [44, 63, 65, 105], having living parents [35, 106], close friends [57, 84], being in a stable relationship [85], good school connectedness [19, 44, 66, 78], academic performance [34], and ambitious educational goals [107] are protective factors. Conversely, having sexually active friends [16, 47, 59, 74, 78, 94], polygamous families [42, 54, 75], large household size [83], and dropping out of school [108] increase the probability of RSB.

At the mesosystem level, socio-economic characteristics and social interactions influence RSB. Coming from a low socio-economic family increases the probability of RSB [72], while results for high socio-economic family are ambivalent [51, 102]. Receiving money from family [27, 32, 42, 53, 62, 81], a job [27, 76, 109], or a boyfriend [76] increases the probability of RSB, whereas not receiving money is protective.

Finally, at the macrosystem level, religious schools are protective against RSB [40, 67], while public [99] and mixed schools [35] increase the probability of RSB. The influence of access to information on SRH issues depends on the source [14, 42]. School-based SRH programs [79, 92, 110], health services [105], and awareness campaigns [18, 79] are protective factors against RSB.

The second series of findings concerns studies that focus on the determinants of sexual violence among adolescents in SSA, i.e., the risk and protective factors of either victimization or perpetration of SV. The WHO defines sexual violence as “any sexual act, attempt to obtain a sexual act, or other act directed against a person’s sexuality using coercion, by any person regardless of their relationship to the victim, in any setting” [111]. Due to the interrelationship between sexual violence and psychological or physical violence, some studies consider all these types of violence together.

Of the 28 included studies that dealt with sexual violence (Table 2), only one involved out-of-school adolescent girls. Eleven studies focused exclusively on girls who were victims of SV, one on boys who were victims, and three on boys who were perpetrators.

At the ontosystem level, determinants include demographic characteristics, socio-cognitive factors, behaviors, and exposure to specific experiences. The influence of gender on SV victimization is ambivalent [115–118, 140], but being male increases the probability of SV perpetration [115, 116, 125], while being female is protective [125]. Being older [115–117, 125] and in a romantic relationship [117–120, 126] increase the probability of both SV victimization and perpetration. Living with both parents is protective against SV victimization [114], while other living arrangements [39, 112, 113, 122–124] increase SV victimization. Poor knowledge of child rights among girls increases the probability of both SV victimization and perpetration [129]. Perceived gender norms [118, 127], perceived peer pressure to have sex [130], perceived risk of contracting HIV [115], positive attitudes toward male sexual entitlement [131], and tolerance towards SV [130] increase the probability of SV victimization, while generalized self-efficacy is protective. Substance use [107, 122, 126, 128, 130, 132, 135], pornography use [126, 130], previous SV perpetration [115, 125, 133], and experiencing any type of SV [115, 125, 127–129, 132, 136, 137] increase both SV victimization and perpetration. Witnessing domestic violence [119, 120, 128], being bullied [118, 132] and having experienced a breakup [93] increase only SV victimization.

At the microsystems level, family, peers, relationship, and school factors influence SV victimization and perpetration. Polygamous families [118], separated parents [114], parental conflict [118], and fearful parental attachment among girls [130] increase SV victimization. Discussions about SRH issues with family increase SV perpetration among boys [126], while lack of discussion increases SV victimization among girls [122, 123]. Peers’ SV experience [129], low school connectedness [125], low feelings of safety at school [125], and poor school engagement [132] increase both SV victimization and perpetration. Having an older partner [130, 134], schoolwork problems [118], and having repeated a school year [125] increase SV victimization, while never attending school is protective for girls [138].

At the mesosystem level, socio-economic characteristics and social interactions influence SV. Low socio-economic status family [116, 123] and food deprivation [129, 139] increase both SV victimization and perpetration. An unemployed head of household [121] and receiving money from external sources [117] increase SV victimization, while a father working, or retired is protective for girls [112, 114].

Finally, at the macrosystem level, educational system, health system, and societal values and norms influence SV. Attending public school increases SV victimization among girls [120], while private school is protective [113]. Boarding school increases SV perpetration [115]. Financial access to health services is protective against SV victimization [128]. Unequal and sexist gender norms increase both SV victimization and perpetration [129].

This study is not without limitations. Most included studies relied on self-reported data, introducing potential social desirability and recall biases that could lead to under- or over-reporting of certain sexual behaviors, especially given the sensitivity of the topic. All studies used cross-sectional designs, limiting the ability to establish causal relationships. A major limitation of the literature is the strong imbalance between studies conducted among in-school and out-of-school adolescents affecting generalizability. Similarly, studies on SV victimization predominantly focused on girls, contributing to gender gaps in the evidence base. Although this review includes studies from various SSA countries, a large proportion of the data comes from a limited number of nations, notably Ethiopia, Nigeria and South Africa. This geographical concentration may limit the regional representativeness of the findings, as sociocultural, educational, and policy contexts vary widely across the continent. Finally, heterogeneity in definitions and measures of sexual behaviors and their determinants precluded the possibility of conducting a meta-analysis.