Systematic review of cash plus or bundled interventions targeting adolescents in Africa to reduce HIV risk

The primary outcomes of interest in our review are HIV and STI infection, and as secondary outcomes we examine pathways and risk factors (mediators), which are laid out in the Conceptual Framework (Fig. 1). This framework illustrates the complex interplay of structural, community, household, relationship, and individual drivers influencing HIV and STI infection among adolescents (see Appendix 2 for details). By integrating bundled or “cash plus” approaches, programmes and interventions can tackle gendered vulnerabilities, provide economic empowerment, and address multiple other health and social risks.

Before describing our search process, we first provide some key definitions. Adolescence comprises a life phase of both biological growth and transitions in major social roles and is often defined as the period from 10 to 19 years [35]. Nevertheless, some have argued that 10–24 years corresponds more accurately to adolescent growth and reflects social aspects of the transition to adulthood [35]. In the current study, we focus on adolescents and young people as reflective of definitions for the age groups identified by the United Nations (ages 10–24 years). This age range also encompasses the target group (i.e., girls and young women aged 15–24 years) for increased investments identified by the Determined, Resilient, Empowered, AIDS-free, Mentored and Safe (DREAMS) partnership, based on increased risk of infection among this age group, particularly girls.

We conducted a systematic review informed by the PI[C]O framework. The interventions we sought to review comprised bundled interventions which include both 1) an economic (in-kind donations, livelihoods training, cash, voucher, or asset transfer); and 2) a health or life skills component (health information or training, psychosocial support, behavior change communication, self-efficacy training, voucher for services, safe spaces, information on gender-based violence prevention, parenting skills) and were implemented in Africa. In-kind economic strengthening included school fees and school kit donations. These could include standalone interventions targeted to adolescents or interventions targeted to adolescents but implemented as part of larger integrated social protection programmes (sometimes referred to as “cash plus”). Programmes with household-targeted cash transfers which provide additional programming targeted to adolescents were considered. Only studies with interventions implemented in Africa that adhered to the following criteria were included: 1) targeted young people ages 10–24 and 2) examined impacts of an intervention with both a health/life skills and economic component and 3) examined impacts on our primary or secondary outcomes.

The studies we included spanned various study designs, including experimental (randomized or cluster randomized control trials) and quasi-experimental (interrupted time series, matching, etc.) studies which include some sort of comparison (pre/post- test, comparison/control group) group, as well as qualitative studies. Qualitative studies were not restricted to those with a comparison group.

To select outcomes of interest, we first conducted a scoping exercise, comprised of a rapid assessment of the known literature, hand searched articles, and articles obtained from search engines (Google Scholar, PubMed). Prior to finalizing the search protocol, we convened a workshop of experts with prior experience on bundled interventions and adolescents (including academics from Eastern Africa, as well as United Nations staff from the United Nations Children’s Fund (UNICEF), the World Health Organization (WHO), the World Food Programme (WFP), and the Joint United Nations Programme on HIV/AIDS (UNAIDS)) in March 2022. The workshop helped to identify key literature, select mediators of interest, and narrowed the focus to HIV prevention (instead of prevention and adherence to treatment). Thus, any studies that focused on HIV-positive adolescents and youth and examined outcomes related to adherence were excluded from this review.

After finalizing the primary and secondary outcomes of interest (described in more detail below), we registered the search protocol with the International prospective register of systematic reviews (PROSPERO; registration ID: CRD42022325270).

We sought to include studies from 2005 through 2023. Studies before 2005 were not included, as the type of bundled capabilities strengthening interventions we aim to summarize were largely not conducted before that time. We did not conduct a meta-analysis because the interventions under review were not adequately homogeneous in terms of intervention components nor outcomes examined to do so.

Our primary outcomes of interest were HIV incidence/prevalence and STI incidence/prevalence. Secondary outcomes included risk factors (mediators) and were determined based on findings from previous research and reviews [13, 27], as well as consultation with an advisory board of experts in the field of HIV and social protection. These included: educational attainment, school enrolment and attendance; food security; health visits; HIV testing; STI testing; pregnancy testing; access to condoms; mental health (depressive symptoms, anxiety); self-esteem and self-efficacy; future expectations for work, livelihood activities, savings, and wealth creation-related knowledge, schooling, and marriage; physical, sexual violence, or emotional violence; transactional sex; age disparate sex (10 years or more difference in age between partners); early marriage (defined as before 18 years); early motherhood or pregnancy (defined as before 18 years); substance use (illicit drugs and alcohol); sexual debut; condom use at last sex; number of sexual partners (past 12 months); concurrent sexual relationships; comprehensive knowledge on HIV prevention; comprehensive knowledge on modern contraceptives and access to sexual and reproductive health services; engagement in sex work; and gender attitudes. (See Appendix 3 for details.)

The search was first conducted on 15 April, 2022 and then again on 25 September, 2023 using a key word search based on findings from the inception workshop and recommendations from experts in the field (see Appendix 1). The search was conducted using search engines PubMed, EMBASE, and Web of Science. Titles and abstracts from selected articles (based on criteria above) were imported from each database and screened by two members of the research team. Conflicts for inclusion/exclusion of articles were resolved through a third-party researcher, using the inclusion criteria listed above. The remaining articles were reviewed in full-text by two reviewers. Two members of the research team and two research assistants then extracted pertinent information from all studies, including study population, timeline, location, type of intervention, research design, analysis, and results (Table 1, and Appendices 4 and 5).

Aggregate qualitative and quantitative data were used in analysis. Following the PRISMA guidelines, we analyzed recurring patterns of impacts on outcomes across studies. We organized findings by intervention type and according to 12 categories of outcomes (Table 4). We rated the strength of causal identification of each quantitative study as low (observational, pre/post), medium (uses quasi-experimental methods to construct a counterfactual; RCT but low number of clusters), and high (RCTs with adequate number of units of randomization). Qualitative studies were not rated in terms of quality of causal identification, as this is not in line with the qualitative research approach.

In the summary charts and results below, counts refer to quantitative studies, and qualitative studies are described separately. One study (Berry, et al. [36]) was framed as a quantitative study, but we analyzed it together with qualitative studies because the sample size was small (N = 40), and the authors did not conduct any statistical tests. In quantitative studies that included a control or comparison group, we counted protective effects only as those that were statistically significantly different between treatment and control/comparison groups and not significant changes over time within the treatment group that may have been reported (e.g., Jennings et al. [37]). Otherwise, if studies only examined pre/post comparisons and did not have a control group, we counted significant differences over time within the treatment group (e.g., Naledi and colleagues [38]). In our counts, if a study examined two sexual risk behaviours, for example, and on one a protective effect was found and on the other a null effect was found, we counted this as the study having a protective effect on sexual risk behaviours in our overall count. However, if a study examined only one outcome in a category and found positive effects in one sub-group and null effects in another sub-group, we counted this as a mixed effect (and vice versa for adverse effects). For the purposes of this paper, separate implementation sites, age groups, and sexes were considered sub-groups; various treatment arms were not. If an effect was positive in one sub-group and not significant in another, it was considered mixed (likewise for negative effects). In contrast, if an effect was positive in one intervention treatment arm, and not significant in another, it was considered a positive effect for the intervention (likewise for negative effects).

A total of 1758 records were identified through the database searches (see Fig. 2). Of these studies, 395 duplicates were removed, and the remaining 1363 abstracts were screened. During the screening process, 1162 studies were excluded based on eligibility criteria. In the full-text review, 201 articles were then screened for eligibility, and 145 were excluded. Expert guidance recommended one additional paper and a backwards search of the included literature added one additional article. A total of 58 peer-reviewed studies were included in the final review, with several studies (those examining DREAMS specifically) including results in multiple countries. A majority of the studies reviewed came from Eastern and Southern Africa. Sixteen studies were conducted in Uganda; 12 each in Kenya and Tanzania; 9 in South Africa; five in Zambia; three each in Lesotho, Zimbabwe, Malawi, and Mozambique; and one each in Ethiopia and Liberia. Of the 58 studies, 43 were quantitative: 27 (46.5%) were randomized or cluster randomized control trials, seven (12%) were quasi-experimental, nine (15.5%) were observational; and 15 (26%) were qualitative. Thirty-eight studies included only adolescent girls and young women, 17 included both adolescent boys and girls, and three involved both AGYW as well as members of their community (e.g., caretakers, community leaders).

The 58 studies included in this review examined various outcomes from 26 interventions: Adolescent Girls Empowerment Programme (AGEP) in Zambia; Adolescent Girls Initiative – Kenya (AGI-K), also in Kenya; Empowerment and Livelihood for Adolescents (ELA) in both Uganda and Tanzania; Girls Empowerment in Lesotho; DREAMS, as implemented in Kenya, Lesotho, Malawi, South Africa, Zambia, and Zimbabwe; the Suubi, Suubi-Maka, and Suubi4Her studies in Uganda; Women First and Go Girls! in Mozambique; Ujana Salama: Cash Plus Model for Safe Transitions to a Healthy and Productive Adulthood in Tanzania; SHAZ! (Shaping the Health of Adolescents in Zimbabwe) in Zimbabwe; Biruh Tesfa (Bright Future) in Ethiopia; the Sauti Project in Tanzania; Bridges to the Future and BridgesPLUS in Uganda; Research Initiative to Support the Empowerment of Girls (RISE) trial in Zambia; Women of Worth in South Africa; Girl Empower in Liberia; and seven unnamed interventions. As such, each study does not represent a unique intervention; rather, the studies are grouped by outcomes and protective factors as relevant to this review’s conceptual framework. In the results that follow, we indicate how many studies found protective associations/effects in each outcome category, and then in parenthesis we list the unique number of interventions with protective effects. When listing percentages, we refer to the percentage of studies which found protective effects on at least one indicator in the outcome category, among the total number of studies which examined that outcome category.

The health/life skills and economic components were not homogenous across studies. Health information included sexual and reproductive health (SRH) knowledge incorporating menstruation, fertility, family planning, and condom use, STI/HIV knowledge and testing, mental health, and/or general knowledge of health resources. Life skills included gender attitudes, GBV-awareness education, conflict management skills, decision-making skills, and/or empowerment. Mentoring was either from a peer or other community member, and vocational training incorporated business planning, skill building workshops, entrepreneurial training, and/or income-generating activities. Fiscal literacy included wealth and savings education, economic development training, and/or financial management skills.

Combinations of intervention components were also not homogenous across studies. As such, we separated out the most frequent combinations, with the understanding that many studies had multiple overlapping combinations (e.g., a study included in the cash plus health information combination may also have a vocational training plus health information component): 23 studies had cash plus health information and 22 had cash plus some life skills component; 20 had cash plus mentoring and 20 studies included vocational training plus life skills; 18 included vocational training plus health information; 14 studies incorporated cash plus a savings account, fiscal literacy, and health information; 6 studies included vocational training plus a mentorship component, while 4 studies incorporated vocational training with microcredit, health, and life skills training; one study had fiscal literacy plus health information plus life skills; and one had a health voucher plus a savings account, fiscal literacy, and life skills. These categories are not mutually exclusive, and the totals sum to more than the 58 studies and 26 interventions reviewed.

Lastly, 13 studies evaluated DREAMS programming, where a majority (n = 11) did not specify what intervention components they were evaluating. DREAMS interventions generally combine core packages to empower girls and young women (condoms, PrEP, violence prevention and post-violence care, HIV testing and counselling, increasing contraceptive method mix, social asset building), reduce the risk of sexual partners (provision of ART to male partners), strengthen families (parenting and caregiver programmes, cash transfers, educational subsidies, socio-economic approaches), and mobilize communities for change (school-based HIV prevention and community mobilization/norms change) [39]. However, intervention components and implementation varied greatly – many of the sites that implemented DREAMS used only a few components of the overall DREAMS programming, but the studies that examined or evaluated DREAMS often did not specify which components were implemented in their study sites. This made it difficult to understand what most of the DREAMS studies we reviewed were actually evaluating.

Table 1 describes the programme components, target population, and location for each of the studies included in our review, both quantitative and qualitative. Tables 2 and 3 summarize the findings, based on the outcomes and mediators of interest, for quantitative and qualitative studies, respectively. Detailed findings, including sample sizes and effect estimates, from all included studies can be found in Appendix 4 for quantitative studies and Appendix 5 for qualitative studies.

Among the 43 quantitative studies, in terms of the study design’s ability to estimate causal impacts, 16 studies were rated as high, 12 as medium, 4 as low-medium, and 11 were rated as low quality (see Appendix 4). Study quality was determined by the rigor of the study design’s ability to causally identify impacts (e.g., RCTs with sufficient cluster size were rated as high, RCTs with low numbers of clusters and quasi-experimental designs were rated as medium, and observational studies without a causal identification strategy were rated as low).

We grouped outcomes into 12 categories across the 58 studies as follows: HIV incidence/prevalence; HIV testing or knowledge of HIV status; HIV prevention knowledge; STI incidence, prevalence, and/or testing (separate from HIV); sexual risk behaviour; early sexual debut; sexual and reproductive health; gender attitudes; gender-based violence; psycho-social well-being and mental health; education; and economic outcomes (see Table 4 for effectiveness of interventions by outcome category). Forty-three studies examined these outcomes quantitatively, and 15 qualitatively.

Two out of seven quantitative studies examining impacts on HIV incidence/prevalence found a protective association. (Protective effects were found in one of the three interventions that included HIV testing/prevalence). Two of these studies found a protection association with the time trend over the period examined in Lesotho and South Africa (but not Kenya) [40, 41], while two studies in Zimbabwe, one study in Kenya and South Africa, one in Tanzania, and another in South Africa found no protective effects [42‐46]. Four of the studies examining HIV incidence had low quality with respect to study design and identification of causal impacts [40‐42, 44], two were rated as low-medium quality [43, 46], and one was high quality [45].

Four out of eight quantitative studies examining outcomes related to HIV testing and/or knowledge of HIV status found protective effects, while one found adverse effects. (Protective effects were found in three of the four interventions that included HIV testing and/or related knowledge.) In Ethiopia, Erulkar et al. [47] found that girls in treatment areas were more likely to want voluntary HIV testing and counselling than those in the comparison group. In Tanzania, Waidler and colleagues [48] found a significant increase in HIV testing. Floyd and colleagues [49] found positive impact on knowledge of HIV status for all DREAMS participants in both Kenya and South Africa. In another DREAMS-related study, Govender and colleagues [44] found that participants were more likely to have been HIV tested. In adverse effects, in South Africa, Naledi and colleagues [38] found that, compared to baseline, the intervention group had significantly lower odds of self-reporting HIV testing in the last 6 months, as well as reduced reporting of perceived HIV risk. At a second, post-intervention follow-up, impacts on testing and risk perceptions were no longer significant [38].

Two studies found mixed protective effects. Birdthistle and colleagues [50] found that knowledge of HIV status was higher among DREAMS beneficiaries compared to non-beneficiaries in Kenya, while in South Africa associations were significant in a younger age group (13–17 years) but not in an older group (18–22 years). Mathur and colleagues [51] examined HIV risk-related behaviours among DREAMS participants in Kenya, Malawi, and Zambia and found mixed protective effects. In Zambia and Malawi, HIV testing increased among all participants, and was significant for both adolescent girls (15–19 years) and young women (20–24 years). In Kenya, HIV testing also increased among all participants, and was significant among adolescent girl participants (15–19 years), but not those 20–24 years old. One study in Zimbabwe found no association with knowledge of HIV status [42].

Nine out of 10 studies (nine out of nine interventions) quantitatively examining HIV prevention knowledge found protective effects. In Uganda, Austrian and Muthengi [52] found that AGYW in their SavingsPLUS arm were more likely to understand HIV transmission mechanisms and HIV prevention methods. Bandiera and colleagues [53] found that the HIV knowledge index increased among intervention participants (although this effect was not sustained at endline). Erulkar and colleagues [47] in Ethiopia found that girls in treatment areas had higher HIV knowledge than those in comparison areas. In Zimbabwe, Dunbar and colleagues [54] found that intervention participants had significantly increased HIV prevention knowledge. In Uganda, Jennings et al. [37] found that adolescents in the intervention increased HIV prevention attitudinal scores and higher odds of a maximum HIV-prevention score. Özler et al. [55] found that participants in both study arms of Girl Empower (GE and GE+) had increased HIV knowledge. Tozan and colleagues [56] found in Uganda that participants in both treatment arms (Bridges and Bridges PLUS) had significantly increased HIV knowledge over the control arm, but there were no impacts on HIV prevention attitudes. In Tanzania, Waidler et al. [48] found the intervention increased HIV-related knowledge. Govender and colleagues found that among participants in DREAMS-like interventions, exposure to an increasing number of interventions was associated with higher HIV prevention knowledge [44]. In Zambia, Austrian and colleagues [57] did not find any impacts on participants’ knowledge of HIV.

One out of seven quantitative studies (one out of five interventions) examining STI testing, symptoms and incidence outcomes found protective effects. Naledi et al. [38] found that the intervention increased STI testing in South Africa immediately following the intervention, but impacts were no longer significant at the post-intervention follow-up. Two studies found mixed protective effects. In Kenya, Kangwana and colleagues [58] found significant reductions in HSV-2 prevalence and incidence, but only among a younger sub-sample (13–14 years at baseline) and not in the full sample. Mathur et al. [51] found that STI symptoms were reduced in DREAMS participants 20–24 years old (but not among those 15–19 years) in Malawi; in Kenya, they found no association between DREAMS participation and STI symptoms. In another study, Govender and colleagues [44] found no association between the DREAMS-like interventions and STI prevalence, and in yet another DREAMS study, Mthiyane et al. [46] found no association between the intervention and HSV-2 incidence. In Zimbabwe, Dunbar and colleagues [43] found no impacts of the intervention on HSV-2, nor did Kuringe and colleagues [45] find association between the intervention and HSV-2 incidence in Tanzania.

Eight out of 20 studies (seven out of 11 interventions) that quantitatively examined outcomes related to sexual risk behaviours found protective effects on at least some outcomes. In Zambia, the intervention evaluated by Austrian et al. [57] reduced engagement in transactional sex. In another study in Kenya, Austrian and colleagues [59] found that the intervention increased condom use self-efficacy scores. Bandiera and colleagues [53] found that the intervention increased the probability of always using a condom. Chabata and colleagues [42] found that residence in a DREAMS area was associated with increased ability to negotiate condom use and decreased odds of condom-less sex with a regular partner. In Liberia, Özler et al. [55] found that participants in both intervention arms of Girl Empower had fewer sexual partners. In Lesotho, Van Heerdan et al. [60] found that DREAMS beneficiaries reported lower levels of sexual risk-taking than non-DREAMS participating peers. In Kenya and South Africa, Govender and colleagues [44] found that exposure to two or more DREAMS-like interventions increased the likelihood of condom use (although they found no association between the interventions and age-disparate sex). In Zambia, Hegdahl and colleagues [61] found that both intervention arms reduced the risk of adolescent girls having been sexually active in the past month, and the combined cash plus community support arm also reduced the risk of unprotected sex among participants.

Two studies found mixed protective effects. In Uganda, Ssewamala and colleagues [62] found that the intervention was protective against sexual risk-taking attitudes among boys, but not among girls. Floyd and colleagues [49] found that, in Kenya, DREAMS reduced condomless sex among 18–22-year-old girls and reduced the odds of having more than two lifetime partners among all DREAMS participants in Kenya, but there were no effects on these outcomes in South Africa. However, Floyd and colleagues [49] found no impact of DREAMS on transactional sex in Kenya and South Africa.

Three studies found adverse programme effects. In Kenya, Mathur and colleagues [51] found DREAMS participants were less likely to use condoms consistently and were more likely to engage in transactional sex (authors found no association with DREAMS and these outcomes in Malawi). In South Africa, Naledi and colleagues [38] also found that participants were less likely to use a condom at last sex. In another study in Uganda, Ssewamala and colleagues [63] found mixed adverse effects: while there appeared to be no effect on sexual risk or attitudes toward condom use at the 24-month follow-up overall, the combined intervention arm reported reduced favourable attitudes toward condoms at 12 months, and the savings-only arm increased sexual risk behaviour at 24 months.

Seven studies found no impacts. Buehren et al. [64] found no programmatic impacts on knowledge of safe sexual behaviour among participants in Tanzania. In Zimbabwe, Dunbar and colleagues [54] also found no impacts in regards to condom use, sexual activity, transactional sex, life preferences, or power in sexual relationship among participants over time. In a separate study in Zimbabwe, Dunbar and colleagues [43] found no intervention effects on condom use, transactional sex, or other sexual activity. In Tanzania, Palermo et al. [65] found no effects of Ujana Salama on age-disparate partnerships. In another study of the same intervention in Tanzania, Ranganathan and colleagues [66] found no impacts on engagement in transactional sex. In Uganda, Tozan and colleagues [56] found no programmatic impacts on sexual risk taking. Last, in Tanzania, Kuringe and colleagues [45] did not find protective effects for the DREAMS Sauti Project on any of the following behaviours: transactional sex, intergenerational sex, condom use, and number of sexual partners.

Eight studies (among five unique interventions) evaluated sexual risk behaviour outcomes qualitatively. Both Banda et al. [67] and Gangaramany and colleagues [68] found that economic support helped to decrease adolescent girls’ reliance on and engagement in transactional sex. Mason and colleagues [69] also found that girls who received both cash and cash plus menstrual education reported feeling empowered to refuse male sexual advances as well as a reduced need to engage in transactional sex due to the extra money. In Zambia, Burke et al. [70] found that the intervention supported less engagement in both transactional sex and age-disparate (intergenerational) partnerships, most likely due to the cash that participants received during the intervention period. In Zambia, Milimo et al. [71] found that adolescent girls had a decreased desire to pursue relationships with boys in exchange for money or gifts. Gichane et al. [72], Pettifor et al. [73], and Wamoyi and colleagues [74] all explored sexual risk behaviour within the Sauti Project Worth+ intervention. Gichane and colleagues [72] found similar results among their study participants in Tanzania, with participants less likely to engage in transactional sex. Pettifor and colleagues [73] found two primary mechanisms to reduce dependence on male sex partners through transactional sex in Tanzania: the first was that the cash provided for more basic needs, such as food or toiletries. The second was that the financial education component of the intervention appeared to empower participants to reject sexual partners. In Malawi, Chimwaza-Manda et al. [75] found that DREAMS participants who were also in Girls Only Clubs consulted others on decision-making and information on sexual relationships, used condoms, and quit sexual relationships more than their non-club counterparts Additionally, they corrected sexual misinformation among their peers with information they learned from the club. Last, Wamoyi and colleagues [74] found that cash transfers among participants in Tanzania empowered adolescent girls to reduce their participation in transactional sex, along with other risky sexual behaviour.

None of the five quantitative studies (across four interventions) examining sexual debut found fully protective effects. However, mixed protective effects were found in one of the four studies (in one intervention). Kangwana and colleagues [58] found in Kenya significant reductions in the percentage of AGYW ever having sex only among those intervention participants who were 13–14 years old at baseline. In Tanzania, Waidler and colleagues [48] and Palermo and colleagues [65] (both examining the same programme) found mixed results, where treatment females sexually debuted earlier than controls, but there were no effects among males. In Zambia, Austrian and colleagues [57] found adverse effects, where intervention participants had an increased probability of ever having sex. Lastly, in Zimbabwe, Dunbar et al. [43] found no effects on sexual debut.

Eight out of 13 quantitative studies (eight out of 10 interventions) found protective effects on SRH outcomes. In Uganda, Austrian and Muthengi [52] found that participants had increased knowledge of contraception. In Zambia, Austrian et al. [57] found the intervention increased SRH knowledge at both Rounds 3 and 5. In Ethiopia, Erulkar et al. [47] found that intervention participants were significantly more likely to know where to obtain voluntary health counselling and testing. In Kenya, two years post-intervention, Kangwana and colleagues [58] found that participants had increased SRH knowledge. Naledi and colleagues [38] found increased contraceptive use among participants following the intervention. Özler et al. [55] found positive effects on SRH knowledge in both treatment arms of the Girl Empower intervention in Liberia. In Tanzania, Waidler and colleagues [48] found the Ujana Salama intervention increased contraceptive knowledge and knowledge of where to seek condoms; however, they also found mixed effects in that the intervention increased health seeking for SRH among males, but not females. Also in Tanzania, Kuringe and colleagues [45] found increased use of biomedical services among DREAMS Sauti participants.

In Kenya, Austrian and colleagues [59] found mixed protective effects. The intervention increased SRH knowledge and knowledge of modern contraceptives in Kibera; however, in Wajir, the intervention reduced contraceptive knowledge.

Four studies found no significant impacts on SRH. In Kenya, Austrian et al. [76] did not find an increase in SRH knowledge nor fertility outcomes in the full sample; however, among those out of school at baseline, the intervention reduced the probability of pregnancy and the fertility summary z-score. Buehren and colleagues [64] did not find any programme impact on knowledge of safe sexual practice or reproductive health for adolescents in Tanzania, and in Uganda, Bandiera and colleagues [53] found no impact on contraceptive use. In Zambia, Hegdahl and colleagues [61] found no impacts on contraceptive use, knowledge of contraceptives, or norms surrounding contraceptive use for either treatment arm.

Two qualitative studies also examined sexual and reproductive health outcomes. In Malawi, Manda and colleagues [77] found that girls’ only clubs (as part of the DREAMS network) were influential for SRH knowledge acquisition among young adolescent girls. In Kenya, Mason et al. [69] found that girls who received menstrual, puberty, and hygiene education reported feelings of empowerment as a result of this knowledge (although only those who also received cash reported behaviour change).

One out of five studies (across five unique interventions) that included outcomes related to gender attitudes found fully protective effects, and one study found mixed protective effects. Özler and colleagues [55] found an increase in gender equitable attitudes regarding IPV among study participants in both arms of Girl Empower in Liberia. However, neither treatment arm had effects on the physical or sexual violence experiences index. Chzhen and colleagues [78] found mixed protective effects from the Ujana Salama intervention in Tanzania. In the pooled sample of males and females, gender equitable attitudes increased at midline, but the effect was not sustained at endline. In the sex-stratified sample, they found that while gender equitable attitudes increased for males at both midline and endline, there were no statistically significant impacts on gender equitable attitudes for females.

In Kenya, Austrian et al. [59] observed adverse effects, finding reduced gender equitable attitudes among intervention participants in Wajir.

Two studies had no significant findings related to gender attitudes. In Kenya, Austrian and colleagues [57] found no significant programme impacts on gender norms. Buehren et al. [64] found no impacts in ITT models on perceived gender role index in Tanzania.

Lenzi and colleagues [79] found in their qualitative evaluation that participants believed the intervention had reinforced or taught harmful gender stereotypes encouraging girls to be “good” (e.g., deferential and obedient, subservient to their families, and sexually chaste), despite objectives to the contrary.

Four out of the 13 quantitative studies (three out of 10 interventions) that examined gender-based violence-related outcomes found fully protective effects. Two out of 10 studies (two out of 10 interventions) found mixed protective effects. In Kenya, Austrian et al. [59] found reduced experience of violence among participants in Kibera. Austrian and colleagues [76] also found an increase in the violence prevention outcomes summary index among study participants in Kenya. Bandiera and colleagues [53] found reduced probability of “unwilling sex” among intervention participants at midline (although this effect was not sustained at endline). Naledi and colleagues [38] found that over time, participants reported lower rates of GBV threats and forced sex, but these effects were not sustained at the later follow-up round.

Mathur and colleagues [51] found mixed protective effects for DREAMS beneficiaries who reported reduced violence over time: less sexual IPV was reported among girls 15–19 years and 20–24 years in Kenya and Malawi (but not Zambia); less non-partner sexual violence was reported among women 20–24 years in Kenya and Zambia, and less physical IPV was reported among women 20–24 years in Malawi. In Tanzania, Palermo and colleagues [65] found mixed protective effects in the Ujana Salama intervention. They found reduced sexual violence experiences among the pooled and female samples, but not among males, and reduced physical violence perpetration among males, but not females.

Two studies observed adverse effects. Austrian and Muthengi [52] found that girls in the savings only arm of the intervention had increased odds of being sexually touched or teased by men. In Liberia, Özler and colleagues [55] found that the GE+ arm of the intervention (and the pooled treatment) increased non-consensual touching.

In Mozambique, Burke and colleagues [80] did not observe any programme impacts on participants’ knowledge related to gender-based violence. In Zimbabwe, Dunbar and colleagues [43] also found no effects on experiences of physical or sexual violence, and in Tanzania, Kuringe and colleagues [45] also did not find any impacts on sexual violence. Last, Govender and colleagues [44] found no association between exposure to DREAMS-like interventions and intimate partner violence in either Kenya or South Africa, nor did Wambiya and colleagues find any impact of DREAMS [81] on emotional, physical, or sexual violence in those same countries.

Two studies qualitatively examined GBV-related outcomes from two separate interventions. In Mozambique, Burke et al. [70] found that study participants reported increased knowledge of what types of behaviour constitute gender-based violence, as well as decreased perpetration of sexual or physical violence against women and girls within the community. In Tanzania, Wamoyi and colleagues [74] found that participants reported reduction in experiences of intimate partner violence in connection with an empowerment programme.

Nine out of 14 studies (five out of eight interventions) that quantitatively examined psycho-social well-being and mental health found protective effects. Austrian and colleagues [59] found that their intervention in Kenya increased the general self-efficacy z-score among participants. In Uganda, Karimli and Ssewamala [82] found that an intervention in Uganda reduced hopelessness and increased self-concept. In Uganda, Kivumbi and colleagues [83] found Bridges to the Future reduced depressive symptoms at Waves 2 and 3 and increased self-concept at Wave 2 (but not Wave 3). They did not find impacts for hopelessness at either wave. In Uganda, Ssewamala and colleagues [84] found lower levels of hopelessness and higher levels of self-concept resulting from the Suubi-Maka Project. In a separate study examining the impact of the Suubi intervention in Uganda, Ssewamala and colleagues [63] found that 24 months post-intervention, the savings-only arm reduced depression and increased self-concept, while the combination savings and family strengthening arm reduced hopelessness and depression and increased self-concept. Also in Uganda, Tozan and colleagues [56] found the Bridges to the Future Plus intervention increased self-concept and self-efficacy, and decreased hopelessness, but the Bridges (no plus component) intervention arm did not have these effects. Van Heerden et al. [60] found that adolescents exposed to DREAMS in Lesotho reported higher levels of self-efficacy as compared with their peers not exposed to DREAMS. In Tanzania, Palermo and colleagues [65] found that the Ujana Salama intervention increased self-esteem. In a separate study of the same intervention in Tanzania, Prencipe and colleagues [85] found the intervention reduced odds of depressive symptoms and increased self-esteem among participants after two years.

Two studies found mixed protective effects. Gourlay and colleagues [86] observed that AGYW reporting having been invited to participate in DREAMS reported higher self-efficacy than AGYW not invited to participate in Kenya (not significant in 2018; in 2019, association was significant among those 18–22 years in Nairobi and 15–17 years in Gem) and South Africa (in 2018 and 2019). In the Suubi4Her intervention in Uganda, Filiatreau and colleagues [87] found reduced hopelessness and increased self-esteem for the treatment arm that combined youth development accounts with a family group intervention, but found no effects for the youth development account only arm.

Özler et al. [55] observed no programmatic impact on the psycho-social index for either intervention arm of Girl Empower in Liberia. While Ssewamala et al. [88] reported a reduction in depression among the treatment group, the difference in slopes over time between treatment and control were not significant, indicating no significant programme impacts. In Uganda, Tutlam et al. [89] found no effects of the Suubi-Maka intervention on outcomes related to prosocial behaviour or emotional and behavioural difficulties.

In South Africa, Sitienei and Pillay [90] qualitatively evaluated participants who received psycho-social support from mentors and peer groups. They found that peer groups especially provided participants with an opportunity to share with and receive support from peers who had had similar experiences.

Five out of 11 quantitative studies (three out of eight interventions) examining educational outcomes found protective effects. In Kenya, Austrian et al. [59] found that Adolescent Girls Initiative-Kenya (AGI-K) increased completion of primary school and transition to secondary school (in Kibera), enrolment and grade attainment (in Wajir), and an education outcomes summary z-score (in Kibera and Wajir) after two years. After four years, Austrian and colleagues [76] found that AGI-K intervention increased the education outcomes summary z-score in Wajir. In Uganda, Curley and colleagues [91] found the intervention increased confidence in achieving education plans. In Kenya, Kangwana et al. [58] found that the intervention increased the schooling outcomes index. In Uganda, Ssewamala and colleagues [84] found that the Suubi Maka Project increased the probability of taking the primary leaving exam, confidence in achieving educational plans, and test scores.

One study found mixed adverse effects. In Tanzania, Palermo et al. [65] found that the Ujana Salama intervention reduced school attendance among girls, but not boys. However, when Prencipe et al. [85] examined a combined indicator of school attendance or vocational training, there were no impacts, and forthcoming findings from the larger study (not published at the time of our review and thus not included), indicated that impacts on school attendance were not sustained in later rounds of data collection, and there were no impacts on schooling attainment, suggesting that these adverse effects were temporary in nature and concentrated among a small sub-sample of intervention participants (older females) [92].

Five studies did not observe programmatic impacts for education outcomes. In Zambia, Austrian and colleagues [57] found that the Adolescent Girls Empowerment Programme (AGEP) intervention had no significant impacts on grade completion. In Mozambique, Burke et al. [80] also did not observe any programmatic impacts related to school attendance. Mulwa and colleagues [93] found no association between having been invited to participate in DREAMS and school attendance in Kenya. In Liberia, Özler and colleagues [55] found Girl Empower had no effect on grades attained or the schooling summary index. Last, as mentioned above, in Tanzania, Prencipe et al. [85] found no effects of the Ujana Salama intervention on school attendance and vocational training (combined indicator).

Four studies examined education outcomes qualitatively from four unique interventions. In Zambia, Banda and colleagues [67] observed an increase in school attendance for AGYW who participated in the intervention. In Malawi, Manda and colleagues [77] found that out-of-school female participants reported that when their guardians were economically empowered (e.g., through cash transfers), this facilitated their return to school. Milimo et al. [71] found that economic support increased the motivation to attend and/or remain in school among study participants. Finally, in Kenya, Mason and colleagues [69] found that girls who received menstrual cups reported less school absenteeism due to menstrual concerns, although this did not appear to impact overall school dropout rates.

Fifteen out of 16 studies (10 out of 10 interventions) that quantitatively examined economic outcomes found protective effects. In Uganda, Austrian and Muthengi [52] found that both the savings and savings plus arms of the intervention increased the odds of having a budget and saving money in the last six months. However, the savings-only arm also lowered the odds of participants having knowledge of reasons for saving. In Zambia, Austrian and colleagues [57] found that at round 3, the intervention increased financial literacy and the probability that girls had saved money in the last six months. At round 5, the intervention was found to have increased the probability of saving money. In Kenya, Austrian and colleagues [59] found that AGI-K increased financial literacy, the likelihood of savings, and the wealth outcomes summary z-score in Kibera and increased the probability of having savings and the wealth outcomes summary z-score in Wajir. In Uganda, Bandiera and colleagues [53] found that the ELA intervention increased self-perceived entrepreneurial abilities (significant at endline but not midline), the probability of engaging in any income-generating activities (significant at both rounds), and the probability of being self-employed (significant at endline but not midline). In Tanzania, Buehren and colleagues [64] found no programmatic effects for any key economic indicators, with the exception of the treatment arm including club + microfinance increasing the probability of having savings at a rotating savings and credit schemes (ROSCA). In Zimbabwe, Dunbar and colleagues [54] found that the intervention increased participants having their own income and savings. Qualitatively, however, participants indicated barriers to loan repayment including economic shocks and insufficient financing. In a later study, also in Zimbabwe, Dunbar and colleagues [43] found that Shaz! increased participants’ own income and reduced food insecurity. In Uganda, Jennings et al. [37] found that the intervention increased cash savings and savings attitudes. Kangwana and colleagues [58] found in Kenya that AGI-K increased participants’ wealth creation summary score. In Uganda, Karimli and Ssewamala [82] found that the intervention increased the likelihood of savings and amount saved, at both 12- and 24-months post-intervention. Naledi and colleagues [38] found in South Africa that the intervention increased the probability of being employed. In Tanzania, Palermo and colleagues [65] found that Ujana Salama increased participation in livestock tending. Also in Tanzania, Prencipe et al. [85] found positive impacts on paid work as a result of the Ujana Salama intervention. In Lesotho, Van Heerden et al. [60] found that AGYW who were exposed to DREAMS were more likely to have savings, as well as a plan for how to spend the money they had earned, as compared to peers not exposed to DREAMS. Last, in Tanzania, Kuringe and colleagues [45] found that the DREAMS Sauti intervention increased savings among participants.

Austrian et al. [76] did not find any programmatic impacts on economic outcomes in the full sample (however the intervention did increase the wealth summary score among AGYW who were out of school at baseline).

Qualitatively, five studies examined economic outcomes from five unique interventions. Burke and colleagues [70] examined economic outcomes related to the business component of their intervention in Mozambique. At round 1, they found that nearly all participants earned money and were satisfied with the amount earned. At round 2, roughly half of participants were earning money, largely due to the fact that they were repaying the intervention for loaned items related to their businesses. Berry and colleagues [36] found that the intervention increased income-generating activity knowledge among participants in Lesotho.