HIV‐sensitive social protection for vulnerable young women in East and Southern Africa: a systematic review

Abstract Introduction Social protection programmes are considered HIV‐sensitive when addressing risk, vulnerability or impact of HIV infection. Socio‐economic interventions, like livelihood and employability programmes, address HIV vulnerabilities like poverty and gender inequality. We explored the HIV‐sensitivity of socio‐economic interventions for unemployed and out‐of‐school young women aged 15 to 30 years, in East and Southern Africa, a key population for HIV infection. Methods We conducted a systematic review using a narrative synthesis method and the Mixed Methods Appraisal Tool for quality appraisal. Interventions of interest were work skills training, microfinance, and employment support. Outcomes of interest were socio‐economic outcomes (income, assets, savings, skills, (self‐) employment) and HIV‐related outcomes (behavioural and biological). We searched published and grey literature (January 2005 to November 2019; English/French) in MEDLINE, Scopus, Web of Science and websites of relevant international organizations. Results We screened 3870 titles and abstracts and 188 full‐text papers to retain 18 papers, representing 12 projects. Projects offered different combinations of HIV‐sensitive social protection programmes, complemented with mentors, safe space and training (HIV, reproductive health and gender training). All 12 projects offered work skills training to improve life and business skills. Six offered formal (n = 2) or informal (n = 5) livelihood training. Eleven projects offered microfinance, including microgrants (n = 7), microcredit (n = 6) and savings (n = 4). One project offered employment support in the form of apprenticeships. In general, microgrants, savings, business and life skills contributed improved socio‐economic and HIV‐related outcomes. Most livelihood training contributed positive socio‐economic outcomes, but only two projects showed improved HIV‐related outcomes. Microcredit contributed little to either outcome. Programmes were effective when (i) sensitive to beneficiaries' age, needs, interests and economic vulnerability; (ii) adapted to local implementation contexts; and (iii) included life skills. Programme delivery through mentorship and safe space increased social capital and may be critical to improve the HIV‐sensitivity of socio‐economic programmes. Conclusions A wide variety of livelihood and employability programmes were leveraged to achieve improved socio‐economic and HIV‐related outcomes among unemployed and out‐of‐school young women. To be HIV‐sensitive, programmes should be designed around their interests, needs and vulnerability, adapted to local implementation contexts, and include life skills. Employment support received little attention in this literature.


I N T R O D U C T I O N
In 2018, East and Southern Africa represented nearly one half of global human immunodeficiency virus (HIV) incident cases [1]. Adolescent girls and young women (AGYW) aged 15 to 25 years accounted for 26%, despite making up 10% of the population [1]. With 6000 new infections per week, their HIV risk is 60% higher than for same-aged males [1]. Vulnerable young women-defined as unemployed and outof-school, aged 15 to 30 years-are at especially high risk of HIV infection [2,3]. They may know about this risk [4] but structural drivers of HIV vulnerability like poverty and gender inequality can reduce their ability to act on HIV prevention choices [5]. Absolute poverty is linked with unprotected and transactional sex [6], and unemployment predicts young women's disproportionate HIV burden [2]. Economic vulnerability constrains their ability to negotiate safe sex and makes it harder to leave abusive relationships [7]. Gender inequality at individual level can translate into women's low relationship power; at societal level, harmful hegemonic masculine norms can result in sexual risk taking and violence against women [8]. Out-of-school girls do not benefit from the protection implicit in educational attainment [9,10] or even the lower risk associated with school attendance [11]. HIV infection among female school dropouts is triple that of schoolgirls [3].
In 2005, UNAIDS advanced consensus on combining programmes reducing HIV risk, vulnerability and impact, formalized as 'combination HIV prevention' in 2009 [12,13]. Socio-economic interventions addressing HIV vulnerabilities like poverty and gender inequality have since been fully endorsed as part of combination HIV prevention [13,14]. Socio-economic interventions could improve young women's power to negotiate contraception and pregnancy, delay sexual debut [15], reduce fertility [16], hence influence lifetime earnings and HIV risk. In the context of social protection, socioeconomic interventions aim to enhance income and employability through livelihood and skills development programmes [17]. Such programmes are considered HIV-sensitive when they also help reduce HIV risk and vulnerability, or mitigate social and economic impacts of the infection [18].
The United Nations Fast-Track Strategy recommends leveraging HIV-sensitive social protection to end AIDS by 2030 [19]. Commitment 6 prescribes that 75% of people at risk of, living with, or affected by, HIV benefit from HIV-sensitive social protection by 2020; Commitment 3 recognizes young women in high-prevalence countries as key beneficiaries; Commitment 5 states 90% of youth should have the skills, knowledge and capacity to protect themselves from HIV in order to reduce new infections among young women [20]. Beyond income transfers that aim to prevent extreme poverty, like welfare or child grants, the 2018 UNAIDS Guidance Note also encourages using socio-economic approaches to address structural drivers of HIV vulnerability [21].
Existing systematic reviews on HIV prevention have summarized combined structural interventions [22,23], income generating [24], microenterprise [25], microcredit [23], and household economic strengthening interventions [26]. No published systematic review has examined HIV-sensitive social protection interventions for unemployed and out-of-school young women, and how they were leveraged for HIV prevention. Most existing reviews included men and women of all ages [22][23][24]26]. Some focused on female sex workers [25,26] or included quantitative studies only [22][23][24]. Additionally, despite their premise that socio-economic empowerment could reduce HIV risk, none assessed socio-economic outcomes when reporting HIV outcomes.
In the context of HIV prevention, we reviewed published and grey literature on HIV-sensitive social protection interventions that aim to enhance livelihood and employability among vulnerable young women in East and Southern Africa. We aimed to collate their documented effects on socioeconomic and HIV-related outcomes and how programmes achieved them.

M E T H O D S
We conducted a systematic review using the narrative synthesis method by Popay et al. (2006), which supports synthesis of complex interventions with considerable heterogeneity [27]. The method relies on text to synthesize findings from studies using different methods. It involves four steps: (i) developing a theory of change or conceptual framework; (ii) a preliminary synthesis; (iii) exploring of relationships within and across studies; (iv) assessing the robustness of the synthesis [27].

Conceptual framework for HIV-sensitive social protection
Our theory of change is as follows: socio-economic and gender inequality increase HIV risk among vulnerable young women, defined as unemployed and out-of-school, aged 15 to 30 years, in East and Southern Africa. 'Cash' social protection reduced sexual risk behaviours among adolescents in South Africa [28]. HIV-sensitive social protection interventions that improve livelihood and employability could enhance income and capabilities and similarly enable young women to act on HIV prevention choices. This could reduce sexual risk behaviours and intimate partner violence (IPV) [29], which in turn may reduce incidence of HIV infection.
Interventions of interest are work skills training, microfinance, and employment support. Work skills training include life skills and professional skills training, like business or livelihood training. Livelihood training can be formal (vocational) or informal (income-generating activity, IGA). Microfinance includes microcredit, savings and microgrants in the form of transfers in cash, in-kind or productive assets. Employment support can be offered in the form of income transfers for public works, work-integrated learning like apprenticeships, or job matching services like job placement or career counselling support. Box 1 provides detailed definitions.
We consider these interventions HIV-sensitive when they address both socio-economic and HIV-related outcomes. Socio-economic outcomes include (self-) employment, income, assets, savings and skills (professional and life skills). HIVrelated outcomes are behavioural (sexual risk behaviour and IPV) and biological: HIV infection, measured as HIV incidence or prevalence, or sexually transmitted infections ( Figure 1).

Income-generating activity (IGA) training
Informal professional skills training for low-skill self-employment. Vocational training Formal professional skills training at nationally accredited institutions for wage employment. Microfinance

Microfinance (MFI)
Financial services for the poor who are unable to access formal banking services. It encompasses a range of services including microgrants, microcredit and savings.

Microcredit
Small business loans given to credit groups who use social pressure for loan repayment.
Group collateral often consists of mandatory savings. Upon repayment, groups can request larger loans. These small business loans are characterized by short repayment periods and high interest rates.

MFI in-kind
Material contributions to provide investment capital like kits with products to sell, waiving of training fees or subsidies of materials to support training and IGA.

MFI savings
Services or support that encourage saving to absorb economic shocks or invest in future expenditure: adolescent-friendly savings accounts; providing a safe place to save; informal revolving group saving schemes.
Productive asset transfers Transfer of material as investment capital to generate sustainable income. Examples are tools, sewing machines, or agricultural inputs like seed, fertilizer or livestock.

Job matching
Services that link individuals with public or private sector employment opportunities, career counselling, job searching and placement support, including support for producing and sharing of curriculum vitae.
Public works Infrastructure and development projects to transfer income to the poor through (temporary) low-skill employment. Wages are kept low to target the poorest through self-selection. Work-integrated learning Occupational opportunities to apply professional training in the real world through observation (internships) or mentoring (apprenticeships).

Mentorship
Provision of (health) information and (psychosocial) support, training and coaching by often slightly older female mentors who model positive behaviour. Safe space (social and physical) Social safe space: regular group meetings that serve as venues for training, information dissemination, critical dialogue, but also for sharing of personal experiences and peer and mentor support. Physical safe space: girls-only or girl-friendly clubs where girls benefit from social safe space (meetings) or merely hangout with peers; often with social activities.

Search strategy
A specialized librarian supported the search strategy based on population, interventions and outcomes of interest described in the theory of change. Studies reporting both socioeconomic and HIV-related outcomes were included. We used text words and indexing terms to identify published studies in three health and social science databases (MEDLINE, Scopus, and Web of Science Core Collection) and grey literature from websites of the World Bank, International Labour Organization, Centre for Social Protection (IDS), UNAIDS and socialprotection.org. We conducted the search on 28 October 2019 with start date January 2005, when socio-economic interventions were acknowledged as part of combination HIV prevention [13]. The search was limited by language (English and French) and place (countries in East and Southern Africa with an adult HIV prevalence higher than 2.5%). Study designs included qualitative, quantitative and mixed methods. We checked references of included papers with backward and forward citation tracking. See Box 2 for inclusion and exclusion criteria and Additional file 1 for the search string.

Study selection
We removed duplicates with EndNote and screened records in Rayyan QCRI. A two-stage process involved screening of titles and abstracts, followed by full-text screening. For review efficiency, we double-screened a random sample of records until reaching a good interrater agreement [30,31]. Two reviewers (RW and DL) independently screened a random sample of 10% of titles and abstracts. They resolved disagreements through discussion, which helped clarify selection criteria. As the interrater agreement was good (k = 0.85), the first author (RW) screened remaining records [32]. We followed the same process for full-text screening. During title and abstract screening, we excluded six protocols pertaining to our review topic. In April 2020, we performed forward citation tracking of these protocols, identified associated published papers, and screened them against eligibility criteria. Two reviewers (RW and DL) reviewed all selected papers to confirm the final sample of included studies.

Data extraction, appraisal and synthesis
Following a convergent data-based synthesis design, we processed included papers with the same synthesis method [33]. One reviewer (RW) extracted data from included papers in two stages. For step 2 of the narrative synthesis (the preliminary synthesis), data extraction followed the population, intervention, context, outcome, study design (PICOS) framework, reported by paper [34]. Several papers reported results for the same project at different stages (pilot and trial) or for different aspects (qualitative and quantitative results). Hence, the second data-extraction stage involved extraction of detailed implementation data per project (Additional file 2) and programme delivery data (mentorship and safe space) (Additional file 3). For synthesis step 3, the exploration within and across studies [27,35], we shifted our focus from projects to intervention components for which we developed two additional tables: (i) the Synthesis To assess the robustness of included studies (synthesis step 4), two reviewers (RW and DL) independently appraised included papers with the Mixed Methods Appraisal Tool [36]. We rated papers as high, moderate or low quality and contacted authors when missing information. No papers were excluded but ratings were taken into account during the interpretation of findings.

Study selection
The PRISMA flow diagram presents results of the search and selection process ( Figure 2) [34]. After removal of duplicate records, we reviewed 3870 titles and abstracts, excluding 3682 in accordance with eligibility criteria (Box 2). Fulltext screening of 188 papers identified 16 papers. Forward citation tracking of relevant protocols identified two additional papers. The resulting 18 papers represented 12 projects. Additional file 4 presents excluded full-text papers with reasons for exclusion. Table 1 shows descriptive characteristics of the 18 included papers. Five papers used qualitative methods, two used mixed methods, six were cluster-randomized controlled trials (CRCT), one was a randomized controlled trial (RCT) and four were observational, of which one was analytical crosssectional; one a clustered non-equivalent two-stage cohort trial; one a longitudinal pre-post intervention with matched controls, and one a shortened interrupted time series. The 18 papers represented 12 different projects that included 22,288 participants from eight countries in East and Southern Africa. All had been implemented by nongovernmental organizations. The average intervention duration was 2.8 years, ranging from 18 months to five years. Four projects focused solely on adolescent girls (13 to 19 years) [37][38][39][40][41][42][43][44]; three on young women 18 years and above [45][46][47][48]; and five on both AGYW [49][50][51][52][53][54].

Quality assessment
We rated nine papers as high, six as moderate and three as low quality. Additional file 5 shows the full appraisal of each paper.

HIV-sensitive social protection interventions and socio-economic and HIV-related outcomes
All projects included work skills training, nine offered microfinance, one offered employment support in the form of apprenticeships. None leveraged employment support in the form of public works or job matching. The Synthesis Table of HIV-sensitive social protection interventions ( Table 2) shows intervention components with associated socio-economic and HIV-related outcomes and additional implementation information.

Work skills training
All projects offered work skills training. Life and business skills contributed improved socio-economic and HIV-related outcomes, which were often sustained after interven-tions ended. Livelihood training produced mixed results: IGA training improved self-employment and income, but failed to reduce HIV-risk behaviours with one exception [37]; standalone vocational training was less suitable for vulnerable young women than more comprehensive interventions.
Life skills training. All projects offered life skills training, but few described content and only five reported outcomes of interest [37,38,[45][46][47][48][49]. Life skills ranged from skills in communication, negotiation, leadership and conflict-resolution to higher order skills like problem-solving and critical thinking. Life skills training increased self-esteem, self-confidence, selfefficacy and aspirations, which helped negotiate condom use, resist transactional sex [47,49], mediate economic empowerment and unwanted sex [37,38]. Psychosocial and sexual risk behaviour outcomes were sustained two years after projects ended [37,38,49]. ZOE Orphan Empowerment (ZOE) in Kenya showed mixed results. Self-efficacy was significantly associated with reduced odds of unprotected sex, sexual initiation and concurrency. Increased resilience, however, was associated with small increases of sexual initiation and concurrency [48].  As they can ask savings group for money, they are less tempted to engage in transactional sex + + Savings groups were not described

Employment support
Public works

None
The lack of research on public works may be due to socio-cultural norms that view public works as appropriate for men. Public works projects may need to pay more attention to gender and consider additional support to make them sensitive to young women Work-integrated learning SHAZ!-I Distrust between mentors and students; 60% of the girls were satisfied with their mentor. More income -likely due to loan No significant change in condom use, sexual activity +?
− Work-integrated learning may require more attention to training of mentors, compensation of students, access to and involvement with professional networks and hiring opportunities for employment after the apprentice period

None
The lack of research on employment support may be due to incompatibility of low-skilled vulnerable young women and limited (higher skills) wage jobs available in contexts with generalized poverty Note: Stepping Stones and Creating Futures (SS&CF) in South Africa took a critical participatory approach to life skills training. Vulnerable young women reflected on skills and resources they could leverage for livelihood and income. Both pilot and full trial reported statistically significant increases in earnings by 278% and 47%, respectively. In the pilot, IPV reduced [46] but the trial showed no effect on women's experience of IPV, although self-reported male IPV-perpetration significantly decreased [45]. Neither pilot nor the trial found changes in sexual risk behaviour. The pilot saw young women's drinking problem significantly increased by 33% but quarrelling about alcohol reduced by half. Authors suggested improved communication skills may have de-escalated conflicts. A similar trend in the trial mid-way was not sustained at two years [45]. Business and financial literacy training. Nearly all projects offered some business or financial training without describing content, duration or level of training. The three projects reporting outcomes offered financial education or general business skills like budgeting and accounting [37,38,47,49]. Financial literacy, self-efficacy and self-reported entrepreneurial skills increased. Business skills significantly increased self-employment [37] and helped young women save, plan and spend responsibly [47,49]. Projects reported reduced sexual risk behaviour [37,47,49]. Improved entrepreneurial skills were sustained two years later [37]. Livelihood training (vocational and IGA-training). Six projects offered livelihoods training, of which four offered IGA training [37][38][39]45,46,48]. Shaping the Health of Adolescents in Zimbabwe (SHAZ!-I and II) offered both IGA and vocational training [42,43] and the Asset project in Kenya compared the two types of training [47].
Formal vocational training took place at nationally accredited institutions. Asset found vulnerable young women with vocational training at increased socio-economic and HIV risk compared with peers engaged in IGA or in a comprehensive programme [47]. SHAZ-II combined vocational training with microgrants, mentors and health services. It found statistically significant results for increased income, food security, condom use, reduced transactional sex and unintended pregnancies [42]. While incidence of HIV (2.3/100 years) and herpes simplex virus-2 (HSV-2) infection (4.7/100 years) were high, SHAZ!-II was not powered to detect statistically significant changes. Only 60% of intervention girls completed vocational training, as they struggled with instruction in English and competing family responsibilities [42,44].
Informal IGA training ranged from candle or soap making, tailoring, hair dressing to small-scale agriculture or animal rearing. Whereas, income increased in all projects but one, IGA training failed to show impact on sexual risk behaviours. The exception was Empowerment and Livelihood for Adolescents (ELA) in Uganda, which reported increased selfemployment, sustained after two years, and significant reductions of teenage pregnancy, unwanted sex and delayed marriage/cohabitation [37,38]. The ELA replication trial in Tanzania failed to demonstrate any statistically significant outcome. Resource constraints negatively affected implementation fidelity. The process evaluation identified girls would have preferred supplementary tutoring. Authors suggest this could be linked with school enrolment being higher in Tanzania than in Uganda [39].
Some studies reported unintended outcomes. In Zimbabwe's collapsing economy, some orphan girls started crossborder trading and faced physical and sexual harm that increased their HIV risk [43]. Increased food consumption in ZOE, Kenya, was associated with increased transactional sex. The authors suggested reverse causality, whereby transactional sex might have increased access to food [48].

Microfinance
All projects offered some form of microfinance, except for SS&CF that encouraged leveraging available resources through capabilities development [45,46]. Microgrants contributed positive socio-economic outcomes like increased earnings and savings, but did not always reduce IPV [52] or sexual risk behaviour [48], and impacted the poorest and most vulnerable differently [40,41,53]. The single microcredit project showing positive effects judged it suitable for 'older and bolder' young women only [50]. Projects offering savings reported improved socio-economic and HIV-related outcomes.
Microgrants: Cash, in-kind and productive assets. Seven projects offered microgrants, of which five offered cash grants [42,47,48,52,53]; two offered productive assets [37,48]. Strengthening Communities through Integrated Programming (SCIP) in Mozambique offered in-kind grants in the form of business kits [40,41]. All projects reported improved socio-economic outcomes like increased earnings [41,48,52], food security [42,52], savings [47] and self-employment [37]. Results were mixed for HIV-related outcomes. Five projects reported reduced sexual risk behaviours [37,41,42,47,53]. When earnings from business kits halted, some SCIP girls married or re-engaged in transactional sex out of financial need. SCIP also explored perceptions of heads of households, influential males and community leaders. Many credited the intervention for perceived reductions in early marriage and pregnancy, and more 'respectful' behaviour in girls, which could reflect prevailing gender norms. Respondents believed gender training had increased community awareness, reducing intergenerational sex and gender-based violence (GBV) [41]. Productive assets in ZOE were not associated with sexual behaviour change [48]. In Northern Uganda, microgrants in Women's Income Generating Support (WINGS) had no effect on IPV except for a small but significant increase in marital control. A one-day gender training session for women and their partners, added in a second phase, had no effect on IPV and economic outcomes, but found significant results for improved communications, quality of relationships and male implication in household chores [52]. Out-of-school AGYW in WORTH+ received three-monthly grants for 18 months. They perceived increased self-esteem, agency and aspirations. They internalized the goal to develop IGA to reduce transactional sex. Linked to basic needs, only the poorest girls reported reducing transactional sex, whereas the better off developed or expanded businesses. The young women also reported cash grants reduced tensions with family and boyfriends and potential IPV [53]. Microcredit. Six projects offered microcredit [37][38][39][40][41]43,50,51,53,54]. Only Tap and Reposition Youth (TRY) in Kenya reported both positive socio-economic and HIV-related outcomes, but only 53% of young women took up the offer of microcredit and half had difficulties to repay. The inflexible lending system led to high dropout rates, but young women appreciated the club's safe space and mentors and leveraged their newfound social networks to start informal rotating saving schemes. Those 20 years and older had significantly more assets, income and savings than adolescent girls, and authors concluded that microcredit was appropriate for 'older and bolder' young women only [50,51]. In ELA-Tanzania, savings similarly increased. Despite low uptake (4%), the offer of microcredit triggered interest in club participation, offering opportunities for informal saving schemes [39].
The Intervention with Microfinance for AIDS and Gender Equity (IMAGE) in our review [54] concerns the subgroup of young women (n = 262) from the CRCT in South Africa, which had been ineligible due to women's mean age (41 years) [29]. It combined microcredit with gender training and reported significant results for reduced sexual risk behaviour, improved communications about sex and having gone for testing. Qualitative findings suggested that discussing sex and testing increased young women's selfconfidence and facilitated negotiating safe sex. With eight new HIV infections, the event rate was too low to measure impact on HIV incidence [54]. Savings. Of five projects that mentioned savings [40,41,47,[49][50][51]53], SCIP did not report savings outcomes [40,41]. Four projects reported improved socio-economic outcomes with increased savings [47,[49][50][51], saving at safer places [50,51], and increased future orientation, as young women saved to buy land or productive assets [53]. Savings enabled young women to refuse sex, insist on condom use [50,51] and resist transactional sex [47,49,53]. In the Adolescent Girls Empowerment Program (AGEP), outcomes did not impact fertility two years after the intervention ended and the most vulnerable girls were more likely married, pregnant or had given birth [49].

Employment support
Only one project, SHAZ!-I, offered employment support in the form of work-integrated learning. SHAZ!-I identified mentors for apprenticeships through community outreach. Hampered by trust issues due to perceived exploitation when mentors lacked time for on-the-job training, and perceived laziness of mentees not showing up for work when lacking transport money, SHAZ!-I changed to mentees choosing their own mentors in SHAZ!-II. Increased income likely resulted from loans and sexual risk behaviours did not change [42][43][44].

Mentorship and safe spaces
All projects instrumentalized mentorship and/or safe space to deliver interventions. Ten projects used mentors who were slightly older young women from the same community [37][38][39][45][46][47]49] or adults [42][43][44]50,51]. They were positive role models [38,47] delivering health, gender or life skills training [37,[48][49][50][51], or offering business support [42][43][44][49][50][51][52]. Most mentors received remuneration and mentor training. Mentors helped create social cohesion, boost attendance [50] and were generally appreciated by girls and young women. The lack of a structured framework in SHAZ!-I led to mistrust between mentors and mentees [43] and inadequate mentor training in ELA Tanzania contributed to null results [39]. Safe space was social space, in the form of regular group meetings, or physical space, as girls-only clubs. Except for WINGS [52], all projects offered regular group meetings, although only three referred to it as safe space [37][38][39]49]. Meetings were venues for peer or mentor support, critical dialogue and sharing of experiences. Many offered socialization free from pressures from (older) men and several offered recreational activities. In TRY, these meetings were the only source of social contact and support for girls [50]. Binti Pamoja Centre in Kenya and ELA clubs in Uganda and Tanzania were physical safe spaces [37][38][39]47]. Girls and young women formed new social networks in social and physical safe spaces and leveraged them to start informal rotating saving schemes [39,50,51]. They relied on these social networks in times of need, reducing their reliance on transactional sex [47,53]. In ELA-Uganda, sustained reductions in sexual risk behaviours at four-year follow-up were attributed to mentors and physical safe spaces, as girls continued attending clubs after training activities halted at two years [37]. In contrast, donated club spaces ELA-Tanzania used were not safe, contributing nonsignificant outcomes [39].

D I S C U S S I O N
Our systematic review identified 12 HIV-sensitive social protection projects that aimed to improve socio-economic and HIV-related outcomes among unemployed and out-of-school young women in East and Southern Africa. All projects offered work skills training, with a majority also offering some type of microfinance. Most projects leveraged mentorship and safe space for programme delivery. Impact on socio-economic outcomes was mostly positive, albeit modest, but impact on HIVrelated outcomes was less consistent. Employment support was under-researched. Our review found insufficient tailoring to participants and local implementation contexts in several interventions. This offers three transferable lessons. First, sensitivity to needs, age, interests, and socio-economic vulnerability of target populations is essential. Of all interventions, microcredit seemed least responsive to vulnerable young women's needs. Low uptake, as little as 4%, indicates little interest in microcredit among adolescent girls [39,43,50,51]. With few assets and high mobility they are considered credit risks [55]. Loan repayment among microcredit users was low indeed  [ 43,50,51]. A recent study found that constraints in savings rather than credit contributed to the inability to sustain increased income after receiving microgrants [56]. Our review shows that young women were eager to save, even starting informal saving schemes in their newfound social networks [39,50,51]. These informal saving schemes can help smooth consumption and guard against negative income shocks, but savings will not overcome poverty if all members are poor [57]. Participants' socio-economic vulnerability also requires attention in programme design. Although microgrants contributed positive socio-economic and HIV-related outcomes, grants only reduced transactional sex among the poorest who used it for basic needs, whereas the financially better off managed to develop or expand IGA [53]. The poor are often reluctant to go into debt and lack time and resources to invest in credit groups [58]. In TRY, authors recommended microcredit, but also work-integrated learning and vocational training for 'older and bolder' young women only [50]. Livelihood training should be adapted to young women's social realities. For example, offered at flexible hours with free childcare to account for competing care responsibilities [59]. Vulnerable young women may also need psychosocial support to benefit from interventions. Mentorship and safe space were key to programme delivery but their spillover effect on social capital may indicate another change mechanism. Frequent socialization and sharing of personal experiences created social networks of trust and reciprocity on which young women relied for psychosocial and economic support, enabling some to reduce transactional sex. IMAGE found social networks increased self-confidence and self-esteem [60], which facilitated acting on HIV-prevention choices. Another study found young women belonging to voluntary savings clubs more likely to drink alcohol and engage in casual sex, however [61]. Safe space may therefore require supportive mentors who model positive behaviour. Second, interventions need to be comprehensive, adapted to local contexts and rely on enabling environments. Although structural, interventions in our review mostly relied on individual behaviour change mechanisms to reduce HIV risk, whereas social and economic environments need to change to address drivers of HIV vulnerability. The Asset study described a context of overwhelming unemployment, sexual harassment while job seeking, and young women lacking professional networks [47]. Zimbabwe's collapsing economy drove girls to risky livelihoods [43], and business kits in SCIP were insufficiently adapted to local context [41]. IGA requires relatively inelastic demand. Vulnerable people prefer steady income flows as they value income most for its capacity to absorb shocks [62]. This requires evaluation of, and interaction with, local markets. In our review, only ELA-Uganda described demand-driven IGA with local entrepreneurs delivering training adapted to local markets. It led to high selfemployment and big reductions in sexual risk behaviour, sustained two years after programme end [38].
Vulnerable young women need linking interventions to facilitate their transition into productive livelihoods. The lack of literature on employment support, notably work-integrated learning and job matching, suggests a lack of 'linking social capital' , the deliberate connecting of young women with other networks [63]. Interventions could forge private sector links through apprenticeships and coaching like projects did in Latin America [64,65], Liberia [59] or Uganda [66].
More generally, interventions may require more time and work with other population groups to change gender norms. WINGS added a gender component for men, but the one-day workshop was insufficient to change gender norms [52]. Interventions could look at how community mobilization efforts in Botswana, South Africa and Uganda changed gender norms through engagement with other population groups [67][68][69][70][71]. Projects in our review were delivered by NGOs and lasted on average 2.8 years, which might be too short to detect significant improvements in socio-economic and HIV-related outcomes, let alone change gender norms. Livelihood and employability interventions may require government involvement and ownership to support a more prolonged, intersectoral approach to HIV-sensitive social protection and achieve more than the mostly modest outcomes we reported.
Third, the review highlights the pivotal role of life skills. Rarely described in detail although offered by all, few projects aimed to measure life skills outcomes. Life skills training improved self-efficacy, self-esteem, sexual negotiation [48,54], HIV testing [54] and reduced sexual risk behaviours [48,49]. Self-confidence and future aspirations facilitated investing in IGA and productive assets [37,38,53]. Communication skills may have de-escalated IPV [45,46,52]. SS&CF demonstrated that higher order life skills like critical thinking and dialogue can lead to economic empowerment without any material or financial support [45,46]. Even when increased income was not associated with reduced sexual risk behaviours, increased self-efficacy was [48]. Enhanced capabilities can sustain outcomes beyond interventions. Improved self-efficacy and selfesteem continued to reduce sexual risk behaviours two years after programmes ended [37,38,49], despite not sustaining increased earnings [37,38].

Updated conceptual framework for HIV-sensitive social protection
We updated the conceptual framework with findings of this review ( Figure 3). For livelihood and employability interventions, we included workforce training, microfinance and employment support. The lack of research on the latter indicates a research gap. As nearly all projects offered additional health and gender training, these have been added as supporting intervention components. We added mentorship and safe space as delivery components along causal pathways to intended outcomes. In addition to income and capabilities, we have added social capital as socio-economic outcome. Improved income, capabilities and social capital may contribute to reduced IPV and sexual risk behaviour and, ultimately, reduced HIV infection among vulnerable young women.
To our knowledge, this is the first systematic review on HIV-sensitive social protection interventions for unemployed and out-of-school young women reporting both socioeconomic and HIV-related outcomes. Our use of multiple databases, specialized librarian, two reviewers for quality assessment, detailed data extraction and conceptual grounding contribute to the strengths of this review. As with any comprehensive intervention with multiple outcomes, it was challenging to attribute specific results to different components. Lack of biomarkers in included studies was another limitation. Including quantitative, qualitative and mixed methods studies provided complementary information that improved understanding of phenomena under study. The narrative synthesis method helped draw out transferable lessons for both impact and change mechanisms.
We recognize a potential selection bias due to independent screening of a proportion of abstracts, titles and fulltext papers by the second reviewer. Other reviews took a similar approach [72,73] and our selection criteria were clear, reflected by a satisfactory kappa statistic [32].

C O N C L U S I O N S
Given intersecting structural drivers of HIV vulnerability, HIVsensitive social protection interventions need to be comprehensive and designed around young women's needs, interests and socio-economic vulnerability. They need to be sensitive to local implementation contexts, to leverage local demand and resources. Microgrants, savings and skills development seem to contribute positive socio-economic and HIV-related outcomes, of which life skills are most likely sustained. Microcredit may not be appropriate for unemployed and out-of-school girls and young women. The potential of leveraging employment support for HIV-sensitive socio-economic programming requires further research. Young women may need psychoso-cial and professional support to achieve and sustain socioeconomic outcomes from livelihood interventions. This could be instrumentalized in design and delivery through mentorship, safe space and the establishing of linking social capital. To also achieve HIV-related outcomes, interventions may benefit from government involvement, longer implementation durations and simultaneously work towards an enabling environment in support of more gender-equal norms.

A C K N O W L E D G E M E N T S
We thank peer reviewers for their constructive feedback and Genevieve Gore, family medicine librarian at McGill University for her help with the search strategy.

F U N D I N G
R.W. is supported by CIHR Vanier Canada Graduate Scholarship; Q.N.H. was supported by a FRQS postdoctoral fellowship. The authors thank the Quebec Population Health Research Network (QPHRN) for its contribution to the financing of this publication.

D I S C L A I M E R
Funding agencies had no role in the study design, data collection and analysis.