Achieving equity in HIV-treatment outcomes: can social protection improve adolescent ART-adherence in South Africa?

Low ART-adherence amongst adolescents is associated with morbidity, mortality and onward HIV transmission. Reviews find no effective adolescent adherence-promoting interventions. Social protection has demonstrated benefits for adolescents, and could potentially improve ART-adherence. This study examines associations of 10 social protection provisions with adherence in a large community-based sample of HIV-positive adolescents. All 10–19-year-olds ever ART-initiated in 53 government healthcare facilities in a health district of South Africa’s Eastern Cape were traced and interviewed in 2014–2015 (n = 1175 eligible). About 90% of the eligible sample was included (n = 1059). Social protection provisions were “cash/cash in kind”: government cash transfers, food security, school fees/materials, school feeding, clothing; and “care”: HIV support group, sports groups, choir/art groups, positive parenting and parental supervision/monitoring. Analyses used multivariate regression, interaction and marginal effects models in SPSS and STATA, controlling for socio-demographic, HIV and healthcare-related covariates. Findings showed 36% self-reported past-week ART non-adherence (<95%). Non-adherence was associated with increased opportunistic infections (p =.005, B.269, SD.09), and increased likelihood of detectable viral load at last test (>75 copies/ml) (aOR 1.98, CI 1.1–3.45). Independent of covariates, three social protection provisions were associated with reduced non-adherence: food provision (aOR.57, CI.42–.76, p <.001); HIV support group attendance (aOR.60, CI.40–.91, p <.02), and high parental/caregiver supervision (aOR.56, CI.43–.73, p <.001). Combination social protection showed additive benefits. With no social protection, non-adherence was 54%, with any one protection 39–41%, with any two social protections, 27–28% and with all three social protections, 18%. These results demonstrate that social protection provisions, particularly combinations of “cash plus care”, may improve adolescent adherence. Through this they have potential to improve survival and wellbeing, to prevent HIV transmission, and to advance treatment equity for HIV-positive adolescents. One thousand fifty-nine ART-initiated adolescents were interviewed using clinic sampling with community tracing in a mixed urban, peri-urban and rural health district of the Eastern Cape, South Africa. From 2014–2015, all public health facilities that provided ART to >4 adolescents were identified (n = 53). Within each facility, all adolescents aged 10–19 who had ever initiated ART were identified through paper and computerised records. All adolescents were followed up in their homes or met at clinics, to ensure inclusion regardless of clinic attendance rates or being lost to follow-up. About 90.1% of the eligible sample was interviewed. Of the remainder, 4.1% refused participation (either adolescent or caregiver), 0.9% had such severe cognitive disability that they were unable to participate, 1.2% were unable to be interviewed for safety reasons and 3.7% were unable to be traced. Voluntary informed consent was obtained from caregivers and adolescents for a 90-minute interview. No incentives were provided, but all adolescents were given a certificate, snack, toothbrush and toothpaste. To prevent identification or stigmatisation through HIV-related research, the study was presented locally as focusing on general needs of adolescents using social and health services. Also with this aim, 467 additional adolescents who were co-resident, or who lived in neighbouring homes, were also interviewed with a version of the questionnaire that did not include items on HIV-medication or HIV-illness (not included in these analyses). Questionnaires, interview schedules and consent forms were translated and back-translated between English and Xhosa, and used tablets with youth-friendly graphics and interactive games. Adolescents participated in the language of their choice. Interviewers were trained in working with HIV-affected adolescents. Confidentiality was upheld, except in cases of significant harm or when participants requested assistance. Where participants reported recent abuse, rape or risk of significant harm, referrals were made to child protection and health services, with follow-up support. Ethics protocols were approved by the Universities of Cape Town (CSSR 2013/14) and Oxford (SSD/CUREC2/12-21), the Provincial Departments of Health and Education and ethics review committees of participating hospitals. The study design was developed in collaboration with the South African National Departments of Health, Social Development and Education, UNICEF South Africa, Regional and New York Pediatric HIV teams, PEPFAR-USAID, and NGOs including Pediatric AIDS Treatment for Africa (PATA) and the Regional Psychosocial Support Initiative (REPSSI). Research tools were informed by in-depth qualitative research, and pre-piloted with 25 HIV+ adolescents in the Eastern Cape. Questionnaires, accompanying vignettes, pictures and games were developed in consultation with two Teen Advisory Groups of HIV-infected and affected adolescents from urban and rural areas of the Eastern Cape (n = 20) and Western Cape (n = 18). ART adherence was measured by adolescent self-report (Evans et al., 2015), using the standardised Patient Medication Adherence Questionnaire (Duong et al., 2001), combined with adolescent adherence measures developed in Botswana (Lowenthal, Haruna, et al., 2014). After piloting, and in order to reduce social desirability bias, vignettes were added, for example, “Even if Andiwe tries his best sometimes unexpected things get in the way and prevent him from taking his pills … this is not his fault”. Past-week and past-year non-adherence were measured using a 95% adherence cut-off, based on the number of prescribed daily doses (Paterson et al., 2000), but past-week adherence was used for all analyses due to evidence of increased reliability for more recent recall. Two validation measures of self-reported adherence were included. Opportunistic infections were measured as sores on the body or face, tuberculosis symptoms (e.g., coughing blood and night sweats), shingles and mouth ulcers in the past six months, using a verbal symptom checklist (Lopman et al., 2006), validated in previous studies of adults in South Africa. Additionally, for a 25% subset of adolescents from randomly selected clinics, viral load measures were collected from clinic files. Social protection provisions included economic “cash” and psychosocial “care” provisions. Within “cash”, cash transfers were any government welfare grant provided to the household (child support, foster child, care dependency, pension or disability grant); food security was measured using items from the National Food Consumption Survey and defined as two meals daily for the past week; school access was capacity to pay for or free access to school, textbooks and uniform. School feeding was measured as daily free provision of a meal at school. Access to sufficient clothing was measured using items from the SA Social Attitudes Survey (Pillay, Roberts, & Rule, 2006). Within “care”: access to an HIV-support group was past-month attendance at either a youth-focused or general HIV-support group; access to sports, choir or arts groups was attending past-month extra-curricular organised activities. Positive parenting (i.e., praise and positive reinforcement from any primary caregiver) and parental supervision/monitoring (i.e., primary caregiver’s monitoring of adolescent activities, rules about going out) were measured using adolescent-reported subscales of the Alabama Parenting Questionnaire (Elgar, Waschbusch, Dadds, & Sigvaldason, 2007). “Parenting’ referred to any biological or non-biological primary caregiver. Potential covariates that were controlled for in analyses were socio-economic factors of adolescent age, gender, language, formal/informal (shack) housing, urban/rural location and education level (highest school grade passed) measured using items adapted from the South African census (SSA, 2011). Maternal and paternal death were asked using items from a South African national survey of AIDS-affected children (Cluver et al., 2013). HIV and medication factors included perinatal/horizontal infection, using modelling data from Southern Africa (Ferrand et al., 2009), whether the adolescent lived with a caregiver who was AIDS-symptomatic or on ART, whether the adolescent was aware of their own HIV-positive status (using clinic file data and adolescent report) and duration of time on treatment using patient file data, supported by caregiver report and cross-checked with adolescent self-report. Healthcare factors included general past-month self-reported health and time of travel to clinic, and whether the participant had received care in hospital for illness in the past year. Analyses were conducted in four stages in SPSS 21.0 and STATA 13.1. First, known characteristics (age, gender, urban/rural location) of excluded participants were compared to those included, to check for potential differences, and subsequently descriptive statistics for outcomes, social protection variables, and covariates were calculated, and social protection provisions were excluded from analyses where a comparison group was too small for reliable analysis (Table 1). Second (Table 2), linear and logistic regressions tested associations of self-reported non-adherence, number of opportunistic infections and detectable viral load, controlling for all potential covariates. ***p < .001, **p < .005, *p < .05. aAll variables shown are entered simultaneously. Third (Table 3), associations between specific social protection provisions and past-week ART non-adherence were assessed, following the sequential approach recommended by Hosmer and Lemeshow (1989). Three logistic regression models were run: (a) with all potential covariates and potential social protection factors to control for potential confounding from non-randomised allocation of social protection provisions, (b) with all covariates and all potential social protection factors significant at .1 or below and (c) with only those covariates and social protection factors significant at .05 or below. Note: All variables entered simultaneously in each stage. ***p < .001, **p < .005, *p < .05, a P < .10. Fourth, we tested for potential interactive or additive effects on adolescent ART-adherence of combinations of social protections. To test for interactive effects, a logistic regression included all covariates, significant social protection provisions (using only those significant in Stage 3 above), and all possible two-way and three-way interactions of significant social protections. To identify potential additive effects, all potential combinations of the statistically significant social protection variables were entered into a marginal effects analysis using binary logistic regression, with covariates held at their mean values. This analysis indicated how the predicted probability of the outcome changed when different interventions (and combinations of interventions) were present (Figure 1). Marginal effects model testing for additive effects of combination social protections on adolescent ART-adherence.