Introduction
Globally, adolescent girls and young women (AGYW, aged 15–24 years) constitute 75% of annual seroconversions among all 15-to 24-year-olds, with an estimated 7000 AGYW acquiring HIV weekly. In sub-Saharan Africa, AGYW account for 20% of the region's seroconversions despite representing only 10% of the population [1]. Although HIV rates have been declining over time, a persistent gender gap remains [2]. In Kenya, adolescent girls’ (15–19 years) and young women's (20–24 years) HIV prevalence (1.2 and 3.4%, respectively) are double that of their male counterparts (0.5 and 0.6%, respectively) [3]. AGYW's heightened HIV risk is driven by intersecting individual, social, and structural factors [1,2,4]. As such, there is growing recognition that ending the HIV pandemic requires attention beyond biomedical and facility-level interventions, necessitating interventions that embrace complexity [5] and address the social, economic, and cultural contexts perpetuating risk for marginalized groups [6–8].
In response to this charge, the ambitious DREAMS (Determined, Resilient, Empowered, AIDS-free, Mentored, and Safe) partnership was launched by the US President's Emergency Program for AIDS Relief (PEPFAR), country governments, and private sector partners to comprehensively address AGYW's HIV vulnerability in select communities with the highest burden of the HIV across 10 countries in sub-Saharan Africa [9]. The DREAMS programming approach aims to empower AGYW and reduce their risk by increasing access to HIV and health services and building their social assets (e.g. enhancing life skills and social networks). Additionally, it aims to strengthen families by increasing access to social protection interventions (e.g. educational support) and parenting programs, engage, and mobilize communities to address gender-based violence and social/gender norms, and reduce risk among male sex partners by linking them to HIV services [9]. The central premise of the DREAMS approach is to provide AGYW with ‘layered’ programming, defined as multiple interventions or services delivered to benefit DREAMS participants directly or indirectly. Specifically, interventions delivered to AGYW provide health information, health services, asset-building activities, and social protections, whereas interventions offered to families, male partners, and community members indirectly benefit DREAMS participants by targeting social and gender-inequitable norms and behaviours. Over time, DREAMS has expanded its outreach within the initially targeted countries/communities to new geographies [10], and similar multisectoral efforts are being considered by other donors and policymakers [7].
Questions remain about which combination of interventions lead to the desired outcomes. Early lessons from DREAMS implementation highlight the challenges of implementing this complex, multicomponent program, leading to variability in program delivery and exposure [11]. In Kenya and South Africa, for instance, over half of DREAMS’ AGYW had participated in social asset building interventions but few had accessed all the interventions available for their age group; their exposure to community-based programs was limited, as well [12]. At the same time, PEPFAR data shows reductions in new HIV diagnoses among AGYW in all DREAMS intervention sites, when compared with other PEPFAR intervention sites [13].
As interest in this multicomponent HIV prevention approach proliferates, research is needed to understand how this programming approach – and specifically, which combinations of programs – can influence HIV-related outcomes. A particular challenge with program evaluations of multicomponent approaches has been assessing the attribution of each component on the desired outcomes because of the high degree of interdependency between components (e.g. health services provided through safe space platforms) [14,15]. Questions around synergies – when a combination of components yields a greater effect than each individual component – also remain [15]. Given this complexity and interdependency, traditional regression approaches are often inappropriate, necessitating the use of novel analytical techniques. Classification and regression tree (CART) analysis is a robust, nonparametric recursive partitioning technique [16,17] that – while allowing for interactions between predictor variables, nonnormal distributions, and missing data – identifies predictor pathways to maximize gains in outcomes. Utilizing data from two sites in Kisumu County, Kenya, we used CART analyses to assess which combinations of DREAMS interventions (i.e. the predictors) affected the likelihood of three HIV-related outcomes – engagement in transactional sex, consistent condom use, experience of sexual violence – among adolescent girls and young women enrolled in DREAMS. We hypothesized that different combinations of interventions would increase or decrease the likelihood of the desired outcomes; we also anticipated that there would be differences by age group and outcome. The results of this analysis may help program planners make informed decisions about the most important investments for HIV risk reduction among AGYW.
Methods
Study setting
The DREAMS program is implemented in sites purposively selected by stakeholders based on the high levels of HIV prevalence and incidence among AGYW. Two sites in Kisumu County, Kenya – one urban and one peri-urban – were selected for the current study in consultation with PEPFAR, implementing partners, and local stakeholders. These two sites are representative of key geographic characteristics and features of DREAMS sites in the county (including, high HIV prevalence, poverty levels, school drop-out rate, orphanhood, population density, and population growth).
Intervention
Details regarding the DREAMS program and its approach are provided elsewhere [9]. In brief, in Kisumu County, Kenya, program implementers identify and invite at-risk AGYW to join the DREAMS program. AGYW risk and vulnerability is defined based on a broad criteria, such as school drop-out, orphanhood, experience of violence, and living in informal settlements [18]. DREAMS interventions are delivered by mentors, trained female community members, in well tolerated and secure community-based locations (safe spaces). Additional professionals/staff are invited to the safe spaces to offer specialized services, such as HIV testing and oral preexposure prophylaxis (PrEP). The mentors and AGYW meet regularly (i.e. once or twice a week) to openly discuss health issues, challenges, and solutions, as well as social and economic empowerment issues affecting AGYW. Each DREAMS participant is offered a comprehensive package of interventions, including support for schooling, economic support, or connection to youth entrepreneurship schemes [19]. Violence prevention is prioritized for all age groups through SASA! (a community mobilization intervention to prevent violence and reduce HIV-risk behaviours) [20] and SHUGA (a television drama series addressing multiple issues covering interpersonal communication, navigating relationships, HIV, gender-based violence and referral for services) [21,22]. Additional social and behaviour change programs are offered for parents and caregivers of adolescent girls or young women who were parents/guardians themselves (Families Matter! Program) to promote positive parenting practices and enhance AGYW–parent communication on HIV and sexual and reproductive health [23,24]. Finally, characterization of male sexual partners is conducted with DREAMS participants to better understand the profiles of their sexual partners and to identify where AGYW meet their partners to inform targeted HIV service delivery.
Data
The study included two rounds of data collection with study participants, the second round of data collection is the analytical sample for the study. For the first round of data collection, study participants were identified using an age-stratified random sample from program rosters of DREAMS participants (n = 5997) [25]. The sample was powered (β=0.20) to detect a minimal 10% change in key outcome indicators. From October 2016 to February 2017, 914 AGYW were interviewed in the first round of data collection. Between April and June 2018, after 14–16 months of DREAMS program engagement, 736 AGYW were re-interviewed (analytical sample for this analysis) [26]. Despite repeated attempts to recontact participants, loss to follow-up occurred because of out-migrations of respondents from the study setting. Comparison of baseline characteristics among those who were followed-up and those who could not reveals no systematic patterns of drop out.
Study procedures
After obtaining informed consent from the respondent (in case of minor, parent/guardian consent, and minor assent), trained female interviewers administered a structured survey to DREAMS participants. Surveys were conducted in a language of the respondent's choosing (Kiswahili, Luo, or English) and in private and convenient locations (e.g. room in respondent's home, nearby field, or nearby community centre) to ensure privacy and confidentiality. The survey included questions on sociodemographic characteristics, HIV knowledge, sexual behaviours, condom use, partner characteristics, experience of violence, and exposure to/participation in DREAMS program interventions
Measures
To understand the impact of the DREAMS program, this study assessed exposure to and engagement with different interventions of the DREAMS program. For the predictor variables, we include respondents’ awareness of PrEP, engagement in livelihoods skills training, and receipt of schooling support or economic support. Additional predictors include the respondent's, family member's, or partner's exposure to the youth fund program, Families Matter! (FM) parenting program, gender-based violence (GBV) prevention program, or male sexual partner engagement program. This study examines occurrence of transactional sex, consistent condom use, or experience of sexual violence in last 12 months. All study measures are presented in Table 1.
Table 1 -
Measures included in the classification and regression tree analysis.
| Variable |
Level of measurement |
Measure |
| Predictors (n = 736) |
| Oral preexposure prophylaxis (PrEP) awareness |
Binary |
Single item question assessed if respondent knew about PrEP for HIV prevention with response categories yes/no. |
| Livelihoods skills |
Binary |
Respondents were asked if they participated in the financial capability and entrepreneurship sessions, learned about financial literacy or business skills. If the respondent said yes to any of these, then they were considered to have acquired livelihood skills. |
| Schooling support |
Binary |
Single item question asked if the respondent had received money for school fees, uniforms, transport, or other help with schooling expenses. Those receiving any of these benefits were considered to have received schooling support. |
| Economic support |
Binary |
Single item question asked if the respondents had received any cash support from the program with response categories yes/no. |
| Youth fund program |
Binary |
A question was asked to respondents if AGYW, their families, or partners had participated in the youth fund program. If any of them participated, it was considered as exposure to youth fund program. |
| Families matter! (FM) program |
Binary |
A question was asked to respondents if AGYW, their families, or partners had participated in the FM program. If any of them participated, it was considered as exposure to FM Program. |
| Gender-based violence (GBV) prevention program |
Binary |
A question was asked to respondents if AGYW, their families, or partners had participated in the gender-based violence prevention programming. If any of them participated, it was considered as exposure to GBV prevention program. |
| Male sexual partner program |
Binary |
A question was asked to respondents if their families or partners had participated in the male sexual partner program. If any of them participated, it was considered as exposure to male sexual partner program. |
| Outcomes |
| No transactional sex with a casual partner in last 12 months (n = 494) |
Binary |
Sexually active respondents who reported sex with a casual partner in the last 12 months in exchange of cash or money, place to stay or sleep at night, monetary and nonmonetary support to children/family, things that cannot be afforded by the respondent and/or payment of school fees or bills. If the respondent reported affirmative to any of these, she was considered to have engaged in transactional sex. |
| Consistent condom use in last 12 months (n = 472) |
Binary |
Sexually active respondents with a current/recent partner who reported ‘always’ using a condom with their main and other sexual partners were considered to have used condom consistently; else, considered to be inconsistent condom user. |
| No sexual violence in last 12 months (n = 736) |
Binary |
Respondents who reported sexual violence either from a partner or nonpartner in the last 12 months. Respondents were asked about physically forced to have sex, use of threats or intimidation to have sex with or forced to do something sexual, which respondent did not want to do. If the respondent reported affirmative to any of these forced sexual experiences, she was considered to have experienced sexual violence. |
AGYW, adolescent girls and young women.
Statistical analyses
Univariate and CART analyses were conducted. Univariate analysis examined the profile of the study participants and their engagement in DREAMS interventions. CART analysis identified the factors influencing the levels of the outcome variable using a nonparametric recursive partitioning technique [16,17,27]. This nonparametric method is advantageous as it requires fewer assumptions about the distribution of data and can reveal important programmatic insights from smaller sample sizes. At the first split, the CART model evaluates the predictors and splits the model on the predictor variable that maximizes intra-node homogeneity and inter-node heterogeneity. Although the statistical program automates the splitting, one may also specify a particular variable to be used for the first splitting. For subsequent splitting, it considers the remaining predictor variables and continues splitting until maximization is achieved, resulting in an inverted tree-like structure.
In this study, we fit separate CART models for each outcome, with age group (adolescent girls vs. young women) as the first preferred predictor for splitting as DREAMS services are specified by age group. This age-stratified approach will also aid in assessing how programmatic factors vary between age groups. For the CART analysis, all the outcomes were recoded where the value ‘1’ represents positive outcome and value ‘0’ represents negative outcome of the variable. For example, transactional sex was coded with value ‘1’ if a respondent did not engage in transactional sex and value ‘0’ if she engaged in transactional sex. In line with the study objective of identifying optimal intervention combinations, we examined all nodes that maximized the gain in each outcome. Analyses were conducted using PASW Statistics (Version 18) [28].
Ethics reviews
The institutional review boards of Population Council, Kenyatta National Hospital/University of Nairobi Ethics and Research Committee, and National Commission for Science Technology and Innovation reviewed and approved the study protocols. Written informed consent was obtained from the respondent, and parent or guardian assent and respondent consent in case of minor.
Results
Table 2 presents the sociodemographic profile for our study respondents and their exposure to DREAMS interventions. Nearly 9% of adolescent girls and 21% of young women had lost both parents, and about 75% of adolescent girls and 27% of young women were in school at the time of the survey. Most young women were engaged in paid work (49%), compared with adolescent girls (10%). Both adolescent girls and young women had similar experience of mobility outside the community. Nearly all young women were sexually active (93%) compared with 45% of adolescent girls.
Table 2 -
Sociodemographic characteristics and program exposure among adolescent girls and young women, 15–24 years.
|
Overall (N = 736) |
Adolescent girls, 15–19 years (N = 389) |
Young women, 20–24 years (N = 347) |
|
% |
% |
% |
| Sociodemographic characteristics |
| Orphan (both parents died) |
14.4 |
8.5 |
21.0 |
| Currently in school |
52.6 |
75.3 |
27.1 |
| Engaged in paid work |
28.6 |
10.3 |
49.3 |
| Ever travelled outside of this community |
88.2 |
86.9 |
89.6 |
| Sexually active |
67.1 |
44.5 |
92.5 |
| Program exposure |
| PrEP awareness |
81.8 |
74.8 |
89.6 |
| Received livelihoods skills |
32.1 |
14.7 |
51.6 |
| Received schooling support |
62.5 |
66.8 |
57.6 |
| Received economic support |
53.4 |
49.6 |
57.6 |
| Participation in youth fund program |
26.4 |
20.8 |
32.6 |
| Exposure to Families matter! (FM) program |
19.8 |
16.5 |
23.6 |
| Exposure to gender-based violence (GBV) prevention program |
31.4 |
28.5 |
34.6 |
| Exposure to male sexual partner program |
9.2 |
6.7 |
12.1 |
| Depth of program exposure |
| Less than 3 |
36.3 |
43.2 |
28.5 |
| 3 |
25.3 |
29.3 |
20.8 |
| 4 |
18.2 |
15.4 |
21.3 |
| 5 and above |
20.2 |
12.1 |
29.4 |
Per the DREAMS program design, there were differences in program exposure by age group. HIV education and awareness of PrEP was high among both adolescent girls and young women, and more young women had received livelihood skills (52%) than adolescent girls (15%). Most adolescent girls and young women had received schooling support (63% overall) and economic support (53% overall) but a higher percentage of adolescent girls received schooling support (67%), whereas a greater percentage of young women received economic support (58%). Overall, 26, 20, and 31% had, respectively, participated in the youth fund program, FM program, and GBV prevention program. Few respondents (9% overall) reported a family member's or partner's engagement with the male sexual partner program. Approximately 36% of the respondents had been exposed to three or less program components, 25% to three, 18% to four, and 20% to at least five, with some differences in the depth of exposure between adolescent girls and young women.
In our sample, 91% of all respondents and 92% of adolescent girls and 90% of young women reported no engagement in transactional sex (Supplementary Figure 1, https://links.lww.com/QAD/C483). For adolescent girls, the outcome of no transactional sex was maximized by 8 percentage points (from 92 to 100%) when respondents received schooling support but were not exposed to the FM parenting or GBV prevention programs. Among young women, transactional sex engagement was minimized through four pathways: exposure to the male sexual partner program but not the FM program (these were young women who were doing paid work and completed schooling at grade 9 or above); exposure to the youth fund program but neither the GBV, male sexual program nor FM programs (key characteristics of young women in this pathway include orphanhood, travel outside the community, engagement in paid work, and completed grade 5 or less); exposure to GBV prevention but not the male sexual partner program, livelihood skills, or FM programs (young women in this pathway were currently in school, not an orphan, engaged in paid work, and had completed grade 9 or above); and receipt of schooling support and exposure to FM program and the youth fund program but not the male sexual partner program (key characteristics of these young women included travel outside the community, engagement in paid work, and having completed grade 9 or above) (Supplementary Figure 1, https://links.lww.com/QAD/C483, each path that maximized the outcome is indicated in bold).
More adolescent girls reported consistent condom use than young women (Supplementary Figure 2, https://links.lww.com/QAD/C484). Among adolescent girls, receipt of schooling support, PrEP awareness, and GBV program exposure (but not the FM program) increased the likelihood of consistent condom use by 58% (29 percentage point increasing from 49 to 78%). Among young women, receipt of schooling support, exposure to the male sexual partner program but not livelihood skills maximized consistent condom use by 1.4 times, shifting 29 percentage points from 21 to 50%.
Of all AGYW, 79% reported no sexual violence from a partner or nonpartner in the last 12 months (Supplementary Figure 3, https://links.lww.com/QAD/C485). Receipt of schooling support and exposure to FM program but not the GBV prevention program minimized adolescent girls's sexual violence experience (no sexual violence shifting 21 percentage points from 79 to 100%). For young women, sexual violence experience was minimized when exposed to the male sexual partner and GBV prevention programs (no sexual violence shifting 17 percentage points from 79 to 96%).
Table 3 summarizes the predictor pathways from the CART analysis that yield maximum positive percentage in an outcome (for more details, see Supplementary Figures 1–3, https://links.lww.com/QAD/C483, https://links.lww.com/QAD/C484, https://links.lww.com/QAD/C485; in each figure predictor pathways that optimized outcomes for each age group are highlighted in bold).
Table 3 -
Summary of DREAMS program components that predict HIV-related outcomes among adolescent girls (15–19-year-olds) and young women (20–24-year-olds) using classification and regression tree models.
|
Program components |
|
|
| HIV-related outcomes |
Livelihood skills |
Schooling support |
Economic support |
Families Matter! Program |
Male sexual partner program |
GBV prevention program |
Youth fund program |
PrEP education/awareness |
Prevalence in the sample (%) |
Percentage improvement in prevalence attributed to pathway |
| No transactional sex |
| adolescent girls |
|
Yes |
|
No |
|
No |
|
NA |
91.9 |
8.8 |
| young women |
|
|
|
|
|
|
|
|
|
|
| Path 1 |
|
|
|
No |
Yes |
|
|
NA |
90.0 |
11.1 |
| Path 2 |
|
|
|
No |
No |
No |
Yes |
NA |
90.0 |
11.1 |
| Path 3 |
No |
|
|
No |
No |
Yes |
|
NA |
90.0 |
11.1 |
| Path 4 |
|
Yes |
|
Yes |
No |
|
Yes |
NA |
90.0 |
11.1 |
| Consistent condom use |
| Adolescent girls |
|
Yes |
|
No |
|
Yes |
|
Yes |
49.1 |
58.4 |
| young women |
No |
Yes |
|
|
Yes |
|
|
|
21.1 |
136.9 |
| No sexual violence |
| Adolescent girls |
|
Yes |
|
Yes |
|
No |
|
NA |
78.7 |
22.0 |
| Young women |
|
|
|
|
Yes |
Yes |
|
NA |
78.7 |
27.1 |
‘Yes’ suggests the exposure to the program predictors and ‘No’ indicates vice-versa. Blank cell suggests that the predictor was not significant enough to contribute in the classification and regression tree (CART) model.
Discussion
Our analysis shows that layering interventions can reduce the likelihood of engaging in sexual risk behaviours and experiencing sexual violence among adolescent girls and young women. The analytical approach used in our study assessed the relative effect of each component on the desired outcomes. Except for three instances, exposure to multiple interventions was necessary to reduce behavioural and experiential HIV risk factors, suggesting that singular, siloed interventions might not be an optimal or strategic choice for AGYW HIV prevention programming in high-risk communities. The layering combinations differed by age group and outcome. We found that even in a multicomponent layering program, microprogramming for each of the intended outcomes is required. Identification of optimal intervention combinations/packages through robust analytical approaches, like CART, can guide more efficient/targeted programming.
For adolescent girls in our study, schooling support was a key intervention affecting all three HIV-related sexual risk behaviours and experiences. Among AGYW, school dropout/being out-of-school has been associated with transactional relationships [29] and HIV acquisition [4,30]. Though prior research supports an equivocal relationship between education and transactional sex [31], a recent longitudinal study in Malawi found both being in school and educational attainment was protective against engaging in transactional sex [32]. In our analysis, schooling support decreased the likelihood of transactional sex among adolescent girls. When schooling support was combined with PrEP awareness/HIV education and exposure to the GBV prevention programming, we found an increased likelihood of consistent condom use among adolescent girls. Additionally, we found that for adolescent girls, likelihood of sexual violence reduced when they received schooling support and were exposed to the Families Matter! parenting program. Although primary education tuition is free in Kenya, secondary education has tuition and other administrative fees (e.g. admission fee, application fee, etc.). Plus, other school costs – regardless of education level – related to attendance (e.g. uniforms, transportation, supplies, etc.) are often cost-prohibitive and a key reason for school dropout [29,30,33]. The financial or material support for schooling provided through DREAMS might have fully mitigated the burden of schooling-related costs among adolescent girls. Although the impact of economic interventions on violence and HIV risk are mixed, a global review of literature found that when economic strengthening interventions were coupled with interventions that shifted gender norms, roles, and responsibilities, there were more positive outcomes [34]. These findings suggest that the combination of social protection interventions (e.g. schooling support), HIV education, and interventions that enhance family support or community-based gender transformative approaches – particularly regarding violence – can lead to improved HIV-related outcomes for adolescent girls.
For young women in our study, additional social protection interventions (like the youth fund program) – including exposure to additional social support through the parenting, male partner, and GBV prevention programs – were needed to minimize engagement in transactional sex, increase consistent condom use, and avoid sexual violence. In this group, the likelihood of transactional sex reduced with access to the Kenyan government's youth fund program. Likelihood of transactional sex also decreased when access to the youth fund program was layered with schooling support and exposure to the parenting program. Access to educational support and exposure to the youth fund program may have eased the financial burden on young women and their families, reducing young women's need for transactional sex. Additionally, exposure of the respondent or the respondents’ family to the parenting program may signal greater connectedness between respondents and their parents/guardians or other trusted adults in the household [24,35,36]. Interestingly, we also found that transactional sex reduced with exposure to either the GBV prevention, youth fund, or male sexual partner program. For instance, we found that for young women who were not orphans, were currently in secondary school and engaged in paid work, access to the GBV prevention program minimized their transactional sex engagement. These promising results suggest multiple mitigation opportunities for reducing transactional sex among young women. The results also underscore the need to engage male partners, families, and trusted adults, and shifting broader social and gender norms to reduce vulnerability for young women.
For young women, consistent condom use was maximized among respondents who received schooling support and were also exposed to the male sexual partner program. Additionally, sexual violence experience minimized with exposure to the GBV prevention and male sexual partner programs. Prior research in these study sites found an association between STI experience and sexual violence [37], confirming earlier research on the links between sexual violence and HIV [38]. Prior research in these study sites also supports the connection between AGYW's low relationship power and increased odds of inconsistent condom use and physical or sexual violence experience [39]. We found that young women or their family's exposure to social and behaviour change programs – such as male sexual partner and GBV prevention programs – attributed to risk reduction. These programs could have weakened the influence of harmful gender norms by facilitating dialogue, challenging assumptions, and educating community members (including male partners) about HIV prevention and condom use. Young women in our study may have benefited from both direct and indirect exposure to these programs. Access to schooling support training may have given young women access to greater resources and agency, whereas male partners’ exposure to services and HIV prevention or engagement in parenting programs may have shifted relationship dynamics to enhance communication, reduce conflict, and reduce violence. Thus, this combination interventions may have influenced both individual-level and environmental-level structural vulnerabilities for young women.
Limitations
Our analysis offers novel insights into the effects of layered interventions but it is not without limitations. First, we cannot infer causality from this cross-sectional analysis. However, the outcomes were measured post program implementation. Therefore, the maximization in outcomes demonstrated using CART models provide insights on how program components have worked once implemented. Second, this study was conducted in a high-transmission urban/peri-urban setting and might not be generalizable to other locations and/or cultural contexts. Third, other program intervention components, such as contraceptive education or counselling, could not be utilized because of lack of sufficient cell frequencies for at least one category of the variables (primarily because of nearly universal program coverage for some program components). Fourth, this analysis does not delve into intervention fidelity (content, intensity, etc.), an area that deserves additional exploration. Lastly, since all data were self-reported, we cannot overlook the potential effects of recall bias and social desirability bias.
In conclusion, innovative, yet complex, multicomponent programming implemented at-scale to reduce HIV risk among AGYW require innovative analytical approaches – like CART – to assess what layering combinations maximize the desired outcomes. Exposure to DREAMS interventions predicted outcomes that can reduce HIV risk among AGYW. To be most effective, layered interventions must address individual-level and structural-level risk for each targeted outcome and account for the heterogeneity of adolescent girls’ and young women's circumstances.
Acknowledgements
S.M. and J.O. conceptualized and implemented the research study, S.M. and B.M. conceptualized the analysis, R.M. and B.M. conducted the data analysis, C.J.H. supported the data management and manuscript preparation, S.M. took the lead in the manuscript development, all authors contributed to the drafting and review of the manuscript. The authors would like to acknowledge support from the program implementation team at Afya Ziwani (PATH) for their collaboration throughout the research study and to the study participants who gave their time generously during the data collection. This study has been made possible with funding support from the Bill & Melinda Gates Foundation (OPP1150068 and OPP1136778).
Funding support: the study was funded by the Bill & Melinda Gates Foundation, Grant Numbers OPP1150068 and OPP1136778.
The contents of this manuscript are the sole responsibility of authors and do not necessarily reflect the views of PEPFAR or the Bill & Melinda Gates Foundation.
Conflicts of interest
There are no conflicts of interest.
References
1. UNAIDS.
Women and HIV — A spotlight on adolescent girls and young women. Geneva, Switzerland: UNAIDS; 2019.
2. UNAIDS.
We’ve got the power — Women, adolescent girls and the HIV response. Geneva: Joint United Nations Programme on HIV/AIDS (UNAIDS); 2020.
3. National AIDS and STI Control Programme (NASCOP).
Preliminary KENPHIA 2018 Report. Nairobi, Kenya: NASCOP; 2020.
4. Santelli JS, Edelstein ZR, Mathur S, Wei Y, Zhang W, Orr MG, et al.
Behavioral, biological, and demographic risk and protective factors for new HIV infections among youth, Rakai, Uganda.
J Acquir Immune Defic Syndr 2013; 63:393–400.
5. Piot P, Bartos M, Larson H, Zewdie D, Mane P.
Coming to terms with complexity: a call to action for HIV prevention.
The Lancet 2008; 372:845–859.
6. Eisinger RW, Fauci AS.
Ending the HIV/AIDS Pandemic.
Emerg Infect Dis 2018; 24:413–416.
7. Dehne KL, Dallabetta G, Wilson D, Garnett GP, Laga M, Benomar E, et al. Global Prevention Focal Point Group.
HIV Prevention 2020: a framework for delivery and a call for action.
Lancet HIV 2016; 3:e323–e332.
8. Gupta GR, Parkhurst JO, Ogden JA, Aggleton P, Mahal A.
Structural approaches to HIV prevention.
The Lancet 2008; 372:764–775.
9. Saul J, Bachman G, Allen S, Toiv NF, Cooney C, Beamon T.
The DREAMS core package of interventions: a comprehensive approach to preventing HIV among adolescent girls and young women.
PLoS One 2018; 13:e0208167.
10. Reflections from AIDS 2020 - DREAMS: A Movement Still in the Making. U. S. Dep. State. Available at:
https://2017-2021.state.gov/reflections-from-aids-2020-dreams-a-movement-still-in-the-making/index.html. [Accessed 15 January 2021]
11. Chimbindi N, Birdthistle I, Shahmanesh M, Osindo J, Mushati P, Ondeng’e K, et al.
Translating DREAMS into practice: early lessons from implementation in six settings.
PLoS One 2018; 13:e0208243.
12. Gourlay A, Birdthistle I, Mthiyane NT, Orindi BO, Muuo S, Kwaro D, et al.
Awareness and uptake of layered HIV prevention programming for young women: analysis of population-based surveys in three DREAMS settings in Kenya and South Africa.
BMC Public Health 2019; 19:1417.
13. DREAMS Partnership. U.S. Dep. State. Available at:
https://www.state.gov/pepfar-dreams-partnership/. [Accessed 15 January 2021]
14. Gift TL, Aral SO.
Predicting the marginal impact of interventions - issues and challenges.
Sex Transm Dis 2016; 43:S8–10.
15. Pronyk P, Schaefer J, Somers M-A, Heise L. Sommer M, Parker RG.
Evaluating structural interventions in public health: challenges. Structural approaches in public health. Routledge,
Options and global best practice. Abingdon, Oxon, New York:2013.
16. Krzywinski M, Altman N.
Classification and regression trees.
Nat Methods 2017; 14:757–758.
17. Breiman L, Friedman JH, Olshen RA, Stone CJ. Classification and regression trees. Wadsworth & Brooks/Cole Advanced Books & Software; 1984. Available at:
https://cds.cern.ch/record/2253780. [accessed 6 February 2021]
18. USAID. The Power of DREAMS: HIV Prevention for AGYW in Western Kenya. Washington, DC, USA: USAID. Available at:
https://path.azureedge.net/media/documents/DREAMS_Kenya_Overview_branded_FINAL.pdf. [accessed 18 November 2021]
19. Youth Enterprise Development Fund. Available at:
http://www.youthfund.go.ke/. [Accessed 10 February 2021]
20. Abramsky T, Devries K, Kiss L, Nakuti J, Kyegombe N, Starmann E, et al.
Findings from the SASA! Study: a cluster randomized controlled trial to assess the impact of a community mobilization intervention to prevent violence against women and reduce HIV risk in Kampala, Uganda.
BMC Med 2014; 12:122.
21. MTV Shuga. MTV Staying Alive Foundation. Available at:
https://www.mtvstayingalive.org/content-creation/. [Accessed 6 February 2021]
22. Banerjee A, Ferrara EL, Orozco-Olvera V.
The Entertaining way to behavioral change: fighting HIV with MTV. Cambridge, MA: National Bureau of Economic Research; 2019.
23. Miller KS. The Families Matter! Program. PEPFAR and CDC. Available at:
https://www.cdc.gov/globalaids/publications/fmp-2-pager-final-jan-2014.pdf [Accessed: 10 February 2021]
24. Poulsen MN, Vandenhoudt H, Wyckoff SC, Obong’o CO, Ochura J, Njika G, et al.
Cultural adaptation of a U.S. evidence-based parenting intervention for rural Western Kenya: from parents matter! To families matter!.
AIDS Educ Prev 2010; 22:273–285.
25. Mathur S, Pilgrim N, Patel S, Okal J, Mwapasa V, Chipeta E, et al.
HIV vulnerability among adolescent girls and young women: a multicountry latent class analysis approach.
Int J Public Health 2020; 65:399–411.
26. Mathur S, Heck CJ, Patel SK, Okal J, Chipeta E, Mwapasa V, et al.
Temporal shifts in HIV-related risk factors among cohorts of adolescent girls and young women enrolled in DREAMS programming: evidence from Kenya, Malawi and Zambia.
BMJ Open 2022; 12:e047843.
27. Loh W-Y.
Fifty years of classification and regression trees.
Int Stat Rev 2014; 82:329–348.
28. SPSS, Inc. PASW Statistics. Chicago. Available at:
http://www.spss.com.hk/statistics/. [Accessed 5 February 2021]
29. Oruko K, Nyothach E, Zielinski-Gutierrez E, Mason L, Alexander K, Vulule J, et al.
He is the one who is providing you with everything so whatever he says is what you do’: a qualitative study on factors affecting secondary schoolgirls’ dropout in rural western Kenya.
PLoS One 2015; 10:e0144321.
30. Santelli JS, Mathur S, Song X, Huang TJ, Wei Y, Lutalo T, et al.
Rising school enrollment and declining HIV and pregnancy risk among adolescents in Rakai District, Uganda, 1994–2013.
Glob Soc Welf 2015; 2:87–103.
31. Krisch M, Averdijk M, Valdebenito S, Eisner M.
Sex trade among youth: a global review of the prevalence, contexts and correlates of transactional sex among the general population of youth.
Adolesc Res Rev 2019; 4:115–134.
32. Gichane MW, Moracco KE, Pettifor AE, Zimmer C, Maman S, Phanga T, et al.
Socioeconomic predictors of transactional sex in a cohort of adolescent girls and young women in Malawi: a longitudinal analysis.
AIDS Behav 2020; 24:3376–3384.
33. Glennerster R, Kremer M, Mbiti I, Takavarasha K. Access and quality in the Kenyan education system: a review of the progress, challenges and potential solutions: J-PAL, MIT, 2011.
34. Gibbs A, Jacobson J, Wilson AK.
A global comprehensive review of economic interventions to prevent intimate partner violence and HIV risk behaviours.
Glob Health Action 2017; 10:1290427.
35. Vandenhoudt H, Miller KS, Ochura J, Wyckoff SC, Obong’o CO, Otwoma NJ, et al.
Evaluation of a U.S. evidence-based parenting intervention in rural Western Kenya: from parents matter! To families matter!.
AIDS Educ Prev 2010; 22:328–343.
36. Kamala BA, Rosecrans KD, Shoo TA, Al-Alawy HZ, Berrier F, Bwogi DF, et al.
Evaluation of the Families Matter! Program in Tanzania: an intervention to promote effective parent-child communication about sex, sexuality, and sexual risk reduction.
AIDS Educ Prev 2017; 29:105–120.
37. Mathur S, Okal J, Musheke M, Pilgrim N, Kishor Patel S, Bhattacharya R, et al.
High rates of sexual violence by both intimate and nonintimate partners experienced by adolescent girls and young women in Kenya and Zambia: findings around violence and other negative health outcomes.
PLoS One 2018; 13:e0203929.
38. Li Y, Marshall CM, Rees HC, Nunez A, Ezeanolue EE, Ehiri JE.
Intimate partner violence and HIV infection among women: a systematic review and meta-analysis.
J Int AIDS Soc 2014; 17:18845.
39. Pulerwitz J, Mathur S, Woznica D.
How empowered are girls/young women in their sexual relationships? Relationship power, HIV risk, and partner violence in Kenya.
PLoS One 2018; 13:e0199733.
