WPS7760 Policy Research Working Paper 7760 A Randomized Control Trial of a Peer Adherence and Nutritional Support Program for Public Sector Antiretroviral Patients Frederik Booysen Damien de Walque Mead Over Satoko Hashimoto Chantell de Reuck Development Research Group Human Development and Public Services Team July 2016 Policy Research Working Paper 7760 Abstract Access to antiretroviral treatment has expanded rapidly self-reported adherence, timeliness of clinic and hospital in South Africa, making it the country in the world with visits, and immunologic response to antiretroviral treat- the largest treatment program. As antiretroviral treatment ment. Peer adherence and nutritional support improved coverage continues to rise in resource-constrained settings, the timeliness of adults´ clinic and hospital visits for effective community-based adherence support interven- routine follow-up while on antiretroviral treatment. Peer tions are of central importance in ensuring the long-term adherence support impacted positively on immunologic sustainability of treatment. This paper reports the find- response to antiretroviral treatment. Scale-up of effective ings from a randomized control trial of a peer adherence and sustainable community-based, peer-driven adherence and nutritional support program implemented in a public and nutritional support interventions should form part of health care setting in South Africa’s antiretroviral treat- the United Nations AIDS Treatment 2.0 strategy’s com- ment program. The analysis assesses the impact of these munity mobilization and health system strengthening pillar. peer adherence and nutritional support interventions on This paper is a product of the Human Development and Public Services Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at booysenfrikkie@gmail.com and ddewalque@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team A Randomized Control Trial of a Peer Adherence and Nutritional Support Program for Public Sector Antiretroviral Patients Frederik Booysen (University of the Free State, South Africa) Damien de Walque (The World Bank) Mead Over (Center for Global Development) Satoko Hashimoto (The World Bank) Chantell de Reuck (University of the Free State, South Africa). JEL Codes: I15, O12, O15. Keywords: HIV and AIDS, Health, Nutritional Programs. Acknowledgements: We thank the study participants and the fieldwork staff as well as the Free State Department of Health (FSDOH) and National Health Laboratory Service (NHLS). The study was funded by the following institutions: Research Committee of the World Bank; The Bank Netherlands Program Partnership; WB-DfiD ‘Evaluation of the Community Response to HIV and AIDS’; Program to Support Pro-Poor Policy Development (PSPPD); a partnership between the Presidency, Republic of South Africa and the European Union; Health Economics and Aids Research Division (HEARD) at the University of Kwazulu-Natal; University of the Free State (UFS); and South Africa’s National Research Foundation (NRF). 1. Introduction In view of the reduction in the cost of triple-drug therapy and the accumulating evidence of the feasibility and effectiveness of ART in slowing the progression of AIDS in poor countries (Havlir, 1998, Lange et al, 2004; Zewdie, Lange and Kuritzkes, 2004), the South African government in 2003 announced a “roll-out” plan for antiretroviral treatment (ART) via the public health sector. South Africa faces the largest HIV burden in the world and currently has the largest ARV treatment program in the world. At present, the program provides treatment to three million people living with HIV and AIDS (PLWHA) (Nene, 2015). The potential benefits of ARV treatment to the individual patient are enormous (Lange et al, 2004; Sterne, et al, 2005) and improvements in the health status of infected individuals are likely to benefit their families’ and their children’s well-being (Bhargava, 2005). The patient’s better health will enable him or her to contribute more to household production and wage-earning activity and to enjoy more leisure activities and a healthy life. As a consequence of the patient’s health, other household members whose time would have gone to care for the AIDS patient or to substitute for the patient’s work can instead pursue higher gain activities, including schooling for the children, labor market participation for adults and leisure activities. However, in order to ensure that these benefits are realized, adherence to ARV treatment must be strict (Nachega et al, 2010). As ART is extended to populations far larger than have ever been reached by ARV treatment programs in developed countries, adherence is likely to suffer and information on the socioeconomic and policy determinants of adherence in poor countries will be increasingly needed. The UNAIDS Treatment 2.0 strategy, moreover, envisages an expanded role for the mobilization of communities in treatment programs (pillar 5), including the use of lay health workers or treatment supporters to provide adherence support to ART patients (UNAIDS, 2011), which furthermore may impact on two other pillars of the Treatment 2.0 strategy, namely reduction in costs and the strengthening of delivery systems (Torpey et al., 2008). Research synthesis of empirical studies on ARV adherence however paints a relatively gloomy picture. An early synthesis of the effects of adherence interventions for ART concludes that more widely targeted interventions have relatively small effects, while only interventions targeting 2    patients with known or anticipated adherence problems show medium effects (Amico et al., 2006). A more recently completed systematic review, which compares the adherence benefits of directly observed versus self-administered ARV treatment moreover finds that directly observed therapy does not outperform self-administered treatment on viral suppression and a range of other adherence measures (Ford et al., 2009), thus bearing out Liechty and Bangsberg’s (2003) pessimism regarding the benefits of directly observed therapy programs for ARV treatment. Yet, another systematic review, which expands study selection criteria to include nonrandomized study designs, claims that directly observed therapy does outperform self-administered treatment on adherence, immunologic and virologic outcomes (Hart et al., 2010). Two recently published studies not included in the former syntheses in turn do report statistically significant, positive impacts on adherence of directly observed therapy interventions (Babudieri et al., 2011; Berg et al., 2011), the latter of which employs a randomized study design. To date, therefore, there appears to be no consensus that community-based adherence support provided through the likes of lay workers impacts positively on adherence and clinical response to ARV treatment. This paper employs longitudinal, panel data from the Effective Aids Treatment and Support in the Free State (FEATS) study to determine how a randomized peer-adherence and nutritional support intervention impact on various subjective and more objective measures of ARV treatment adherence. Section 2 describes the study design and data, while section 3 summarizes the methods employed in the statistical and econometric analyses. Section 4 presents the results, while section 5 concludes. 2. Methodology The Effective Aids Treatment and Support in the Free State (FEATS) study, a prospective cohort study and experimental study with a combined group time-series, quasi- or field experiment and Zelen-type double randomized consent design,1 was approved by the Ethics committee of the Faculty of Health Sciences (UFS) [ETOVS 145/07]. The experimental component of the FEATS                                                              1 In the Zelen design subjects are only offered and provided with information regarding the treatment that they are assigned. The Zelen design is deemed appropriate given the fact that: blinding is not practicable or possible; the use of classical randomization and informed consent procedures significantly threatens internal validity; the interventions are highly attractive; the control group receives standard care; the study focuses on a clinically relevant objective(s) and offers important new insights (Kaptchuk, 2001; MacLehose et al, 2001; Rains & Penzien, 2005). 3    study comprised a peer adherence and nutritional support intervention. The trial is registered in South Africa [DOH-27-0907-2025] and with the United States’ National Institutes of Health (NHI) [NCT00821366]. 2.1 Experiment The open enrollment into this prospective, experimental study was managed by ARV nurses employed at each of the 12 larger phase-I ART clinics in Free State province. Eligible ARV clients had to be adult (18+ years), had to have initiated ART in the past month, and had to reside in the local community where the particular clinic is based. ARV patient and patient households recruited into the study were randomly assigned to one of three groups: a control group receiving treatment and support provided in the existing program, a group that received bi-weekly visits by trained peer adherence supporters (PAS) and a group that received the bi weekly visits by the PAS adherence supporters and nutritional supplements (refer Figure 4). The nutritional support consisted of two 400g cans of meatballs and spaghetti in tomato sauce per week. Each can contains (per 100 grams) 420 kJ of energy, 6.6 grams of protein, 9.9 grams of carbohydrates, 4.1 grams of fat, 29.9 mg of magnesium, 116.6 mg of phosphorous, and 1.6 mg of iron. In addition, the tomatoes are a good source of vitamin C. Prior to follow-up, the peer adherence and nutritional interventions were implemented. Approximately 60 peer adherence supporters were recruited in consultation with ART staff at the relevant clinic, in proportion to the number of FEATS study participants recruited at each facility. To be considered, individuals had to have been on ARV treatment for ≥ 12 months, had at least a grade-10 certificate, and live within walking distance from the relevant clinic. Peer adherence reporters received 5-days of basic training in ART and adherence support from staff at the School of Nursing at the University of the Free State. Training focused on seven main themes: facts about HIV/AIDS, anti-retroviral therapy (ART), adherence supported needed by an ART client, nutrition, infection control at home, and using a health care team approach. On the 5th day of training, peer adherence supporters’ knowledge and practical skills was assessed by the trainers using an oral test and practical exercise. 4    Following the completion of training and assessment, peer adherence supporters were assigned eight FEATS study participants each, four in each intervention arm. An original randomization list was employed for this purpose, with an additional, rank-ordered list of reserve cases, per intervention group. During recruitment, PAS explained the intervention to each client and obtained written, informed consent from FEATS study participants who volunteered to partake in the experiment. As illustrated in Figure 5, a relatively large proportion of patients randomized to the two treatment arms of the study (almost one-third) could not be traced and contacted to be offered the relevant intervention. As a result, one in four study participants recruited into the two experiments represented so-called reserve cases originally included in the control group. A small proportion only (6.1% or 22 cases) of contacted study participants refused the offered adherence and nutritional support, while due to logistical issues a small number of participants assigned to a treatment crossed over to the other treatment arm at implementation (13 cases or 3.9%). (For a detailed discussion of Figure 5 refer to the CONSORT statement.) During the study, a brief survey instrument was used to collect key information from the 52 peer adherence supporters (PAS) working in the project at the time. (Initially, a total of 57 peer adherence supporters was recruited, trained and employed. By this time, five PAS had resigned due to not being able to fulfil their responsibilities, with patients being re-assigned to other PAS.) The peer adherence supporters can be described as follows: • Predominantly female (98%) – only one male PAS remains in the study • Mainly holds higher secondary education levels: grade 10 (20%), 11 (34%), 12 (38%) • Aged mid to higher 30’s: mean age 35.8 years, [median 34.5 years, IQR 32.0 – 39.5] • Majority walk when visiting clients: 56% walk, 38% taxi/bus, 6% own transport • Single trip to closest client on average takes 39.1 minutes [median 30 min, IQR 15–60] • Single trip to furthest client on average takes 94.5 minutes [median 90 min, IQR 45–120] • Majority (98%) meets their client at the client’s home 5    By completion of the experiment, the majority of PAS (55.6%) had already been on ARV treatment for 5 years. During the experimental phase of the study, these peer adherence supporters visited each client at least twice per week. At each visit, the peer adherence supporter had to complete a one-page check list. Peer adherence supporters were paid a monthly stipend of $100 (ZAR 800), conditional on performance. Supervision was performed by two trained peer adherence support coordinators, whom visited each clinic once a month to meet with the relevant peer adherence supporters. During this visit, peer adherence supporters received re-training on selected topics included in the original main training to refresh and maintain knowledge transfer. In addition, supervisors debriefed PAS, dealt with any logistic problems and collected completed checklists from PAS. Supervisors completed a standard, short-report for each site visit. Following each monthly visit, supervisors met with the PAS coordinator at the Faculty of Medicine, who collated information from their reports and further discussions into a monthly progress report. Information in the monthly report was used to determine PAS payments for the subsequent month. For the study participants allocated to group C, the PAS were in addition delivering two cans of the food support at each visit. By completion of the FEATS intervention at 18-months, a substantial proportion of FEATS study participants (39.8%) had dropped out of the treatment arms, probably due to a combination of fatigue among both participants and peer adherence supporters as well as the potential discontinuation of support in anticipation of the completion of the study. On aggregate, patients spent 13.8 months on the FEATS treatment. Patients in the peer adherence support only arm on average were in the intervention for longer that patients in the adherence and nutritional support arm (14.4 versus 13.1 months), a difference that is statistically significantly different (p=0.011). Figure 6 shows the survival functions for each of the two treatment arms. However, the difference in the survival functions, according to the log-rank test for equality of survivor functions, is not statistically significantly different (p=0.227). At the completion of the intervention, study participants were debriefed and asked to sign a formal termination of experimental intervention notice. ARV patients in the control group were provided with two month’s stock of the canned 6    food that the peer adherence and nutritional support intervention group received for the duration of the experimental intervention. In the subsequent pages, patient-level data are used to investigate the impact of peer adherence and nutritional support on measures of ARV treatment adherence and success. 2.2 Data collection Trained enumerators conducted structured, face-to-face baseline interviews with ART patients recruited into the study.2 Written, informed consent was obtained from study participants by the nursing personnel at the respective clinics (for ARV patients), as well as by the enumerator (for ARV patients and ARV patient/comparison households). A facility survey was also conducted, during which semi-structured interviews were conducted with health care providers at each study site. At follow-up, enumerators again obtained written, informed consent from study participants. Two rounds of follow-up interviews were conducted with patients. The median time between consecutive interviews was 11.7 months [IQR 10.6-15.9 months]. A total of 1,588 interviews were conducted with 653 individual patients, 422 of whom were interviewed at both follow-ups. By the second and final round of follow-up interviews, 218 patients had been lost to follow-up, primarily due to mortality among study participants (42.4%) and unknown whereabouts (34.1%), which translates into an aggregate attrition rate of 33.6%. Of the 337 patients that consented to enrollment in the adherence and nutritional support interventions at the commencement of the intervention, 288 or 85.4% were interviewed at the first round of follow-up interview, whereas 175 or 86.2% of the 203 study participants that completed the full 18-months of FEATS treatment were interviewed at the second and final round of follow-up interviews. On completion of the study clinical bio-markers and other additional patient information were obtained from paper and electronic files. Patients at baseline, when first recruited into the study,                                                              2 The study also included a household survey component, with interviews being conducted with patient and comparison households, this to achieve various other objectives of the larger study (refer CONSORT statement). 7    consented that the research team access their clinic and hospital records to collect the latter information. 2.3 Socio-demographic and other characteristics of the study population Table 1 reports selected baseline characteristics of the FEATS patient study population. Study participants were mostly in their late 30s. As is the case in the ARV program in general, approximately three in four study participants were women. Two-thirds of patients enrolled into the FEATS study were single, while one in four lived with their partner or spouse. Three-quarters lived in formal dwellings and one in five in informal dwellings. Only in one in five participants had completed secondary school or held some tertiary education, while almost half had some level of secondary education [grade 8-11]. Households generally included 2-4 members, one in three of which represented a child or elderly person. With the exception of age (attrited study participants were marginally, but statistically significantly younger than study participants interviewed in both rounds of follow-up interviews), socio-demographics did not differ statistically significantly across FEATS study participants that stayed in the study and those that were lost to follow-up, thus suggesting the absence of major attrition bias. 3. Methods Adherence outcomes for study participants were measured as follows: a continuous and dichotomous version of the CASE index of self-reported adherence (Mannheimer et al.. 2006); a continuous, normalized index of self-reported adherence derived from five categorical variables on self-reported adherence using multiple correspondence analysis (MCA adherence index); a self- reported, dichotomous 0/1 variable for a missed clinic or hospital visit; and timeliness for scheduled clinic and hospital visits, which represents a more objective proxy of adherence than self-reported adherence (Blacher et al., 2010; Kunutsor et al., 2010). Delays in clinic and hospital visits, used here to measure timeliness of visits, were calculated by subtracting the scheduled visit dates from the actual visit dates for consecutive clinic and hospital visits. The latter variable was also converted into a dichotomous variable indicating whether the patient arrived three or more days late for the particular visit (=1) or not (=0). Immunologic response to ARV treatment is measured using CD4 counts (cells per microliter). (In the case of timeliness of clinic and hospital 8    visits and CD4 counts, the data includes all outcomes observed between recruitment into the study and completion of the FEATS experiment. The advantage of the latter approach is that the statistical power of the analysis increases dramatically, because observations included in the analysis originates from patients’ records and is therefore not dependent on a patient interview having been conducted with the said FEATS study participant.) The analysis conducted in this study comprised the following: Firstly, we assess balance at baseline in adherence outcomes and other observable socio-demographic characteristics, by original intent-to-treat status and actual treatment arms (Tables 2-5). On the basis of this evidence, we reflect on the presence of potential selection bias. Secondly, we employ basic bivariate analysis to assess differences in our main outcomes, i.e. self-reported adherence outcome, timeliness of clinic and hospital visits, and immunologic response, across treatment arms for intent-to-treat and actual treatment status (Tables 6-7; Figure 7-8). The statistical significance of differences in means and medians are assessed using t-tests, F-tests and Wilcoxon rank-sum tests. Finally, we employ fixed effects (FE) linear regression models with Instrumental Variables (IV) and random effects (RE) linear regression models adjusted for selection bias to investigate the impact of the FEATS experiments on self-reported adherence, timeliness of clinic and hospital visits, and immunologic response to ARV treatment (Tables 8-10; Figure 9). In each case, we control for basic socio- demographics and ARV treatment duration. Following suggestions by Angrist et al (1996) and Greenland (2000), the original intent-to-treat status is used as instrument to adjust for selection bias in the fixed effects (FE) regression analysis, while in the random effects (RE) regression analysis the inverse Mills ratio is used to adjust for selection bias. To be deemed significant for the purposes of discussion, results need to be statistically significant at least at the 95% level (although results significant at a 90% level are reported). 4. Results At baseline, age, sex, education, monthly income and pre-ART CD4 count, in terms of the intent- to-treat comparison, did not differ significantly across treatment arms, as did all self-reported measures of adherence. However, differences in marital status are statistically significant when comparing the peer adherence support only and control groups. The peer adherence support only arm comprise a larger proportion of single adults compared to the control group. 9    Timeliness of clinic and hospital visits at baseline does though differ statistically significantly across treatment arms. Delays in clinic and hospital visits for clients in the two treatment arms exceeded delays for the control group. More specifically, delays in clinic visits were significantly lower in the peer adherence support only arm, and delays in hospital visits significantly lower in the peer adherence and nutritional support arm. These differences possibly are the result of patient files not being available for data extraction for all study participants, which implies that outcomes are not missing in a non-random manner. Yet, balance does exist across treatment arms in self- reported adherence and most observable socio-demographic characteristics. Tables 4-5, however, illustrate the presence of a clear selection bias into the treatment arms of the experiment. In the case of the arm for peer adherence support only, those with higher levels of education selected into treatment. According to our results, study participants who exhibited poorer adherence than those in the control group in terms of record-based timeliness of clinic and hospital visits selected themselves into the peer adherence and nutritional support interventions. The opposite, however, is true for self-reports of missed clinic and/or hospital visits. Here, those selecting into the peer adherence support only arm, reported better adherence than in the control group. (As the former are considered more objective measures of adherence, we discount the latter evidence of selection bias in favor of the former.) For these reasons, regression results were adjusted for selection bias using instrumental variables (IV). Moving to study outcomes as observed before and during the provision of peer adherence and nutritional support, there are few significant differences in adherence outcomes by the intent-to- treat assignment of subjects. Only in the case of the proportion of delayed clinic visits is the outcome significantly better in the two treatment arms as well as in the combined treatment arm for any peer adherence support. The same is true for comparisons based on actual treatment assignment, including for the duration of delays in clinic visits (see Figures 7a/b). This is not true for hospital visits (Figures 8a/b). The attention now shifts to the results of the regression analysis. Table 8 provides little evidence that the experiment’s interventions impacted significantly on measures of self-reported adherence. (Two of the three results significant at the 10% level are counter-intuitive (support increases the reporting of missed clinic and/or hospital visits), whereas one result only supports the argument 10    that peer adherence and nutritional support results in fewer missed visits for routine medical check- ups.) No results are statistically significant at the 5% level. Table 9 reports the results for the intent-to-treat and the Fixed Effects (FE) with instrumental variables (IV) regression analysis of timeliness of clinic and hospital visits. There is strong evidence that the combination of peer adherence and nutritional support impacts positively on adherence. Clients receiving peer adherence and nutritional support are less likely to miss clinic and hospital visits and less late for hospital visits, regardless of whether the analysis is based on intent-to-treat assignment or actual treatment assignment, correcting for selection. The results hold in part when the analysis focuses on the combined treatment arms (i.e. any peer adherence support), particularly for missing a clinic visit. Table 10 suggests that peer adherence support only impact positively on CD4 counts. Estimates show that a person in the peer adherence support only group will have a CD4 count that is permanently larger by 60 cells per microliter. According to simulations based on the coefficients in table 4, a person receiving peer adherence support only will have a CD4 count 12 points higher after 1 month, 24 points higher after 4 months, 50 points higher after 16 months, and 60 points higher after 25 months. These significant improvements in CD4 counts are maintained across the duration of the peer adherence support intervention, regardless of whether adherence support commences at 6, 12 or 18 months on ARV treatment (Figure 9). The result that nutritional support (in combination with peer adherence support) does not impact on clinical outcomes is perhaps not surprising, given the relatively small size of the supplement (2 cans of food per week) and the fact that the vast majority of clients (>95%) reported sharing the food with others. It is also likely that patients took up the combined treatment, not because of a need for peer adherence support, but due to the opportunity of receiving nutritional support that helps strengthen the household’s food security. It is also interesting here that the impacts of treatment on behavioral as opposed to biomedical outcomes are different: peer adherence and nutritional support impact on timeliness of visits, while peer adherence support only impacts on immunologic response. Further research is required to untangle the behavioral complexities underlying such results. 11    5. Conclusion The quantitative findings presented in this paper suggest that peer adherence and nutritional support, though not impacting significantly on self-reported adherence, do improve the timeliness of adults´ clinic and hospital visits. In addition, adherence support has been shown to impact positively on immunologic response to ART treatment. These results suggest that scaled-up effective and sustainable community-based, peer-driven adherence and nutritional support interventions would be a useful component of the UNAIDS Treatment 2.0 strategy´s community mobilization and health systems strengthening pillars. 12    References Amico, K.R., Harman, J.J. & Johnson, B.T., 2006. Efficacy of Antiretroviral Therapy Adherence Interventions. Journal of Acquires Immune Deficiency Syndromes, 41(3): 285-297. Angrist, J.D., Imbens, G.W. & Rubin, D.B., 1996. Identification of Causal Effects Using Instrumental Variables. Journal of the American Statistical Association, 91: 444–455. Babudieri, S., Dorrucci, M., Boschini, A., Carbonara, S., Longo, B., Monarca, R., Ortu, F., Congedo, P., Soddu, A., Maida, I.R., Caselli, F., Madeddu, G. & Rezza, G., 2011. Targeting Candidates for Directly Administered Highly Active Antiretroviral Therapy: Benefits Observed in HIV-infected Injecting Drug Users in Residential Drug-Rehabilitation Facilities. AIDS Patient Care and STDs, 25(6): 359-364. Berg, K.M., Litwin, A., Li, X., Heo, M. & Arnsten, J.H., 2011. Directly observed antiretroviral therapy improves adherence and viral load in drug users attending methadone maintenance clinics: A randomized controlled trial. Drug and Alcohol Dependence, 113: 192-199. Bhargava A. (2005). AIDS epidemic and the psychological well-being and school participation of Ethiopian orphans. Psychology, Health and Medicine, 10, 263-275. Blacher, R.J., Muiruri, P., Njobvu, L., Mutsotso, W., Potter, D., Ong'ech, J., Mwai, P., Degroot, A., Zulu, I., Bolu, O., Stringer, J., Kiarie J. & Weidle, P.J., 2010. How late is too late? Timeliness to scheduled visits as an antiretroviral therapy adherence measure in Nairobi, Kenya and Lusaka, Zambia. AIDS Care, 22:11: 1323-1331. Ford, N., Nachega, J.B., Engel, M.E. & Mills, E.J., 2009. Directly observed antiretroviral therapy: a systematic review and meta-analysis of randomised clinical trials. Lancet, 374: 2064-2071. Garland, W.H., Wohl, A.R., Valencia, R., Witt, M.D., Squires, K., Kovacs, A., Larsen, R., Potterat, N., Anthony, M.N., Hader, S. & Weidle, P.J., 2007. The acceptability of a directly-administered antiretroviral therapy (DAART) intervention among patients in public HIV clinics in Los Angeles, California. Aids Care, 19(2): 159-167. Greenland, S., 2000. An introduction to instrumental variables for epidemiologists. International Journal of Epidemiology, 29: 722-729. 13    Hart, J.E., Yean, C.Y., Ivers, L.C., Behforouz, H.L., Caldas, A., Drobac, P.C. & Shin, S.S., 2010. Effect of Directly Observed Therapy for Highly Active Antiretroviral Therapy on Virologic, Immunologic, and Adherence Outcomes: A Meta-Analysis and Systematic Review. Journal of Acquires Immune Deficiency Syndromes, 54(2): 167-179. Havlir D.V. & Lange J.M., 1998. New antiretrovirals and new combinations. AIDS, 12(Supplement A): S165–S174. Kaptchuk, T., 2001. The double-blind, randomized, placebo-controlled trial: Gold standard or gold calf? Journal of Clinical Epidemiology, 54(6): 541-549. Kunutsor, S., Walley, J., Katabira, E., Muchuro, S., Balidawa, H., Namagala, E. & Ikoona, E., 2010. Clinic Attendance for Medication Refills and Medication Adherence amongst an Antiretroviral Treatment Cohort in Uganda: A Prospective Study. AIDS Research and Treatment, 2010: 1-8. Lange J., Perriens J., Kuritzkes D. & Zewdie D., 2004. What policymakers should know about drug resistance and adherence in the context of scaling-up treatment of HIV infection. AIDS, 18(Supplement 3): S69-S74. Liechty, C.A. & Bangsberg, D.R., 2003. Doubts about DOT: antiretroviral therapy for resource-poor countries. AIDS, 17: 1383-1387. MacLehose, R.R., Reeves, B.C., Harvey, I.M., Sheldon, T.A., Russell, I.T. & Black, A.M.S., 2001. A systematic review of comparisons of effect sizes derived from randomised and non-randomised studies. Health Technology Assessment, 4(34): 1-153. Mannheimer, S.B., Mukherjee, R., Hirschhorn, L.R., Dougherty, J., Celano, S.A., Ciccarone, D., Graham, K.K., Mantell, J.E., Mundy, L.M., Eldred, L., Botsko, M. & Finkelstein, R., 2006. The CASE adherence index: A novel method for measuring adherence to antiretroviral therapy. AIDS Care, 18:7, 853-861. Motsoaledi, A., 2011. Health Budget Vote Policy speech by Minister of Health, National Assembly, 31 May, Cape Town. Nachega, J.B., Mills, E.J. & Schechter, M., 2011. Antiretroviral therapy adherence and retention in care in middle-income and low-income countries: current status of knowledge and research priorities. Current Opinion in HIV & AIDS, 5(1): 70-77. 14    Nene, N., 2015. Budget speech by Minister of Finance, National Assembly, 25 February, Cape Town. Page-Shipp, L.S., Charalambous, S., Roux, S., Dias, B., Sefuti, C., Churchyard, G.J. & Grant, A.D., 2007. Attitudes to directly observed antiretroviral therapy treatment in a workplace HIV care program in South Africa. Sexually Transmitted Infections, 83: 383-386. Rains, C. & Penzien, D.B., 2005. Behavioural research and the double-blind placebo-controlled methodology: Challenges in applying the biomedical standard to behavioural headache research. Headache, 45(5): 479-486. Santos, C., Adeyemi, O. & Tenorio, A., 2006. Attitudes towards directly administered antiretroviral therapy (DAART) among HIV-positive inpatients in an inner-city public hospital. AIDS Care, 18(7): 808- 811. Shin, S., Munoz, M., Zeladita, J., Slavin, S., Caldas, A., Sanchez, E., Callacna, M., Rojas, C., Arevalo, J., Sebastian, J.L. & Bayona, J., 2011. How does directly observed therapy work? The mechanisms and impact of a comprehensive directly observed therapy intervention of highly active antiretroviral therapy in Peru. Health and Social Care in the Community, 19(3): 261-271. Sterne J.A.C., Hernán M., Ledergerber B., Tilling K., Weber R., Sendi P., Rickenbach M., Robins J., Egger M. & Swiss HIV Cohort Study. (2005). Long-term effectiveness of potent antiretroviral therapy in preventing AIDS and death: a prospective cohort study. Lancet, 366, 378–84. Torpey, K.E., Kabaso, M.E., Mutale, L.N., Kamanga, M.K., Mwango, A.J., Simpungwe, J., Suzuki, C. & Mukadi, Y.D., 2008. Adherence Support Workers: A Way to Address Human Resource Constraints in Antiretroviral Treatment Programs in the Public Health Setting in Zambia. PLoS ONE, 3(5): e2204. UNAIDS. 2011. The Treatment 2.0 Framework for Action: Catalysing the Next Phase of Treatment, Care and Support. Geneva: UNAIDS. Zewdie D., Lange J. & Kuritzkes D., 2004. Editorial. AIDS, 18 (Supplement 3): S1-S3. 15    Figure 1: UNAIDS’s Treatment 2.0 Strategy Source: UNAIDS (2011: 5). 16    Figure 2: Mean Effect Sizes of ARV Therapy Adherence Interventions Source: Figure 2 from Amico et al (2006: 292). 17    Figure 3a: Impact on Virologic Response of Directly Observed ARV Therapy Sources: Figure 2 in Ford et al (2009: 173). 18    Figure 3b: Impact on Virologic Response of Directly Observed ARV Therapy Sources: Figure 2 in Hart et al (2010: 2068). 19    Figure 4: FEATS study design as per original protocol Group A: Group B: 216 ART households receiving 216 ART households: A + treatment and support provided in bi-weekly visits by trained ARV the existing programme peer adherence supporter (PAS) Group C: Group D: 216 ART households: A + B + 208 comparison households nutritional supplement (canned randomly sampled from the food) delivered by PAS relevant communities 20    Figure 5: Assignment and uptake of FEATS experiments, by intervention arm adherence and nutritional adherence support only control support A. Assignment phase: master reserve cross-over master reserve cross-over 1. master 216 216 216 [a = master] 2. reserve 74 75 149 [b = reserve] 3. total [1+2] 290 291 67 [c = a - b] B. Recruitment phase: 4. contacted 146 28 154 31 174 185 5. not contacted 70 62 132 [d] 46 44 90 [e] 116 106 222 [f = d + e] 6. refused treatment offer 9 2 8 3 22 [g] 11 11 7. offered treatment [4-6] 137 26 146 28 163 174 8. cross-over during enrolment 2 2 4 7 6 1 9. enrolment [7-8] 135 24 7 140 27 4 311 [h = c + f + g] 166 171 311 [i] C. Treatment phase: 10. pre-treatment dropout 9 1 0 4 0 4 10 8 18 [j] 11. uptake [9-10] 126 23 7 136 23 4 156 163 329 [k = i + j] 12. cross-over between treatment arms 22 2 2 26 14 1 0 13 10 10 20 [l] 24 36 13. treatment 154 185 309 [m = k - l] 64 [n] 21    Table 1: Selected baseline socio-demographic and pre-ART baseline CD4 count Respondents Full cohort at Balanced lost to follow- baseline cohort up (n=654) (n=421) (n=233)               Median age (years) [IQR] 37 [31-43] 37 [32-44] 36 [31-41] **              Female (%) 76.8 69.2 62.4             Marital status: Single 66.9 69.2 62.4 Not cohabiting with partner 9.6 8.9 11.0 Cohabiting with partner 23.5 21.9 26.6 Total       100.0   100.0 100.0      Dwelling: Formal 74.3 74.1 74.7 Informal 19.7 20.1 18.9 Traditional 6.0 3.7 6.4 Total    100.0    100.0  100.0      Education: None 2.5 3.4 0.9 Primary 27.0 26.0 29.2 Some secondary 47.8 47.6 48.1 Grade 12 19.9 20.0 19.9 Tertiary 2.7 3.1 1.9 Total       100.0    100.0 100.0      Median household size [IQR]    3 [2-4]    3 [2-4]   3 [3-4]      Mean dependency ratio (%) 33.3 33.3 33.3              Median baseline CD4 [IQR] 127 [74.5-183.5] 129.5 [77-180] 185 [66-185] Note: Results are for patients recruited into the FEATS study cohort at baseline in 2007/08. Baseline CD4 values represent the CD4 count observed closest to but prior to ARV treatment initiation. Some percentages may not add up to 100% across categories due to rounding. One, two and three asterisks denote differences that are statistically significant at the 10%, 5% and 1% levels, respectively. Source: Authors’ own calculations. 22    Figure 6: Kaplan-Meier survival function, by treatment arm Source: Authors’ own calculations. 23    Table 2: Socio-demographic and other selected participant characteristics at baseline, by original intent-to-treat status Peer adherence support only Peer adherence and nutritional Control [1] [2] support [3] Total mean n mean n p-value [a] mean n p-value [b] mean n Mean age (years) 37.6 215 37.1 214 0.306 37.5 212 0.470 37.4 641 Female (%) 76.8 212 75.8 211 0.399 77.5 209 0.439 76.7 632 Education: None 3.3 7 0.9 2 0.291 3.3 7 0.975 2.5 16 Primary 27.0 57 28.1 59 26.0 54 27.0 170 Some secondary 46.4 98 50.9 107 45.4 94 47.6 299 Grade 12/post-secondary 23.2 49 20.0 42 25.1 52 22.7 143 Total 100.0 211 100.0 210 100.0 207 100.0 628 Marital status: Single 61.1 129 73.9 156 0.016 65.8 137 0.603 66.9 422 Not cohabiting with partner 10.9 23 8.5 18 9.6 20 9.6 61 Cohabiting with partner 27.9 59 17.5 37 24.5 51 23.3 147 Total 100.0 211 100.0 211 100.0 208 100.0 630 Median pre-ART initiation CD4 [IQR] 127.0 197 132.5 198 0.724 126,0 187 0.823 [69-186] [79-183] [73-178] Mean monthly income [ZAR] 445 211 475 210 0.350 552 205 0.069 490 626 Source: Authors’ own calculations. 24    Table 3: Self-reported adherence at baseline, by original intent-to-treat status Peer adherence support only Peer adherence and nutritional Control [1] [2] support [3] Total mean n mean n p-value [a] mean n p-value [b] mean n Self-reported adherence: MCA adherence_baseline -0.016 215 -0.020 214 0.485 0.065 213 0.182 0.009 642 CASE_continuous 15.4 215 15.4 213 0.477 15.5 212 0.163 15.4 640 CASE_cutoff (%) 95.3 215 94.3 214 0.327 97.1 213 0.159 95.6 642 Clinic/hospital visits: Missed a drug collection visit (%) 3.7 215 2.3 214 0.202 2.3 213 0.204 2.8 642 Missed a routine medical check-up (%) 1.3 215 1.4 213 0.495 1.8 212 0.345 1.5 640 Missed either drug collection/routine check-up visit (%) 5.1 215 3.2 213 0.173 3.7 212 0.251 4.0 640 Timeliness of clinic/hospital visits: Mean days delayed clinic visit 5.9 64 11.4 132 0.023 19.1 70 0.158 12.1 266 Median days delayed clinic visit [IQR] 0 1 0.010 0 0.303 0 [0-3] [0-20] [0-6] [0-7] Delayed clinic visit (%) 23.4 40.1 0.010 30.0 0.197 33.4 Mean days delayed hospital visit 10.8 37 13.0 74 0.316 69.7 19 0.036 20.7 130 Median days delayed hospital visit [IQR] 0 0 0.335 24 0.010 0 [0-13] [0-21] [0-58] [0-24] Delayed hospital visit (%) 32.4 37.8 0.290 68.4 0.004 40.7 Source: Authors’ own calculations. 25    Table 4: Socio-demographic and other selected participant characteristics at baseline, by actual treatment status Peer adherence support only Peer adherence and nutritional Control [1] [2] support [3] Total mean n mean n p-value [a] mean n p-value [b] mean n Age (years) 37.4 63 37.8 153 0.385 37.8 183 0.392 37.7 399 Female (%) 73.0 63 77.3 150 0.251 76.5 179 0.288 76.2 392 Education: None 4.8 3 0.0 - 0.010 4.3 8 0.425 2.7 11 Primary 20.9 13 33.7 51 24.5 45 27.5 109 Some secondary 59.6 37 47.6 72 48.6 89 50.0 198 Grade 12/post-secondary 14.5 9 18.5 28 22.4 41 19.7 78 Total 100.0 62 100.0 151 100.0 167 100.0 396 Marital status: Single 61.9 39 72.3 110 0.240 71.2 129 0.259 70.2 278 Not cohabiting with partner 9.5 6 9.2 14 9.9 18 9.6 38 Cohabiting with partner 28.5 18 18.4 28 18.7 34 20.2 80 Total 100.0 63 100.0 152 100.0 181 100.0 396 Median pre-ART initiation CD4 [IQR] 117.0 59 120.5 144 0.362 136.0 168 0.059 127.0 371 [63-172] [72-179] [80-186] [72-183] Mean monthly income [ZAR] 439 63 416 150 0.401 447 179 0.460 434 392 Source: Authors’ own calculations. 26    Table 5: Self-reported adherence at baseline, by actual treatment status Peer adherence support only Peer adherence and nutritional Control [1] [2] support [3] Total mean n mean n p-value [a] mean n p-value [b] mean n Self-reported adherence: MCA adherence_baseline 0.037 63 -0.113 153 0.178 -0.036 183 0.326 -0.054 399 CASE_continuous 15.4 63 15.1 153 0.162 15.4 183 0.410 15.3 399 CASE_cutoff (%) 95.2 63 92.8 153 0.256 95.0 183 0.488 94.2 399 Clinic/hospital visits: Missed a drug collection visit (%) 6.3 63 1.9 153 0.049 3.2 183 0.144 3.2 399 Missed a routine medical check-up (%) 1.5 63 1.3 152 0.439 1.6 182 0.486 1.5 397 Missed either drug collection/routine check-up visit (%) 7.9 63 2.6 152 0.038 4.9 182 0.190 4.5 397 Timeliness of clinic/hospital visits: Mean days delayed clinic visit 4.1 16 11.6 122 0,075 13.6 128 0,314 12.1 266 Median days delayed clinic visit [IQR] 0 0 0,025 0 0,049 0 [0-16] [0-28] [0-5] [0-7] Delayed clinic visit (%) 18.7 40.1 0,048 28.9 0,198 Mean days delayed hospital visit 2.8 15 14.1 64 0,053 33.8 51 0,166 20.7 130 Median days delayed hospital visit [IQR] 0 0 0,047 5 0,004 0 [0-0] [0-22] [0-28] [0-24] Delayed hospital visit (%) 13.3 39.0 0,029 50.9 0,004 40.7 Source: Authors’ own calculations 27    Table 6: Self-reported adherence, by intent-to-treat status Peer adherence and Peer adherence nutritional support Any peer adherence Control [1] support only [2] [3] support [4] p-value [a] p-value [b] Self-reported adherence: MCA adherence index -0.008 (0.06) 0.027 (0.06) 0.006 (0.07) 0.929 0.017 (0.05) 0.749 CASE index 15.3 (0.1) 15.4 (0.1) 15.3 (0.1) 0.956 15.3 (0.1) 0.855 CASE_cutoff (%) 94.9 (1.4) 95.4 (1.4) 93.8 (1.7) 0.781 94.7 (1.1) 0.912 Clinic/hospital visits: Missed either drug collection/routine check-up visit (%) 6.9 (1.6) 8.5 (2.0) 3.4 (1.3) 0.126 6.2 (1.2) 0.737 Timeliness of clinic/hospital visits: Mean days delayed clinic visit 11.1 (1.0) 9.6 (1.0) 8.8 (1.0) 0.319 9.2 (0.7) 0.159 Delayed clinic visit (%) 30.1 (1.7) 24.1 (1.5) 20.8 (1.3) <0.001 22.4 (1.0) <0.001 Mean days delayed hospital visit 15.3 (1.9) 12.2 (1.7) 15.3 (2.5) 0.476 13.7 (1.5) 0.511 Delayed hospital visit (%) 36.2 (2.6) 34.3 (2.6) 34.2 (2.7) 0.832 34.3 (1.9) 0.545 Source: Authors’ own calculations. 28    Table 7: Self-reported adherence, by treatment status Peer adherence and Peer adherence nutritional support Any peer adherence Control [1] support only [2] [3] support [4] p-value [a] p-value [b] Self-reported adherence: MCA adherence index -0.100 (0.13) -0.095 (0.09) 0.089 (0.06) 0.210 0.004 (0.05) 0.435 CASE index 15.1 (0.2) 15.3 (0.1) 15.5 (0.1) 0.325 15.4 (0.1) 0.291 CASE_cutoff (%) 93.5 (2.7) 94.4 (1.9) 95.2 (1.6) 0.874 94.8 (1.2) 0.662 Clinic/hospital visits: Missed either drug collection/routine check-up visit (%) 8.8 (3.2) 7.3 (2.2) 1.5 (0.9) 0.021 4.2 (1.1) 0.112 Timeliness of clinic/hospital visits: Mean days delayed clinic visit 13.7 (1.4) 9.6 (1.0) 8.6 (0.8) 0.016 9.0 (0.6) 0.005 Delayed clinic visit (%) 37.8 (48.5) 22.8 (42.0) 22.1 (41.5) <0.001 22.4 (41.7) <0.001 Mean days delayed hospital visit 17.6 (2.5) 13.5 (2.0) 13.3 (1.8) 0.375 13.4 (1.3) 0.162 Delayed hospital visit (%) 39.3 (48.9) 34.2 (47.5) 33.5 (47.5) 0.342 33.8 (47.3) 0.146 Source: Authors’ own calculations. 29    Figure 7a: Mean and median delay in scheduled clinic visits (days), by treatment arm 18 16 14 13,8 12 10 9,7 9,0 9,1 8 6 4 2 0 Control PAS only Any PAS Food Figure 7b: Late for a scheduled clinic visit (%), by treatment arm 45 40 37,8 35 30 25 22,8 22,4 22,1 20 15 10 5 0 Control PAS only Any PAS Food Note: Figures represent means with 95% confidence intervals. Source: Authors’ own calculations. 30    Figure 8a: Mean and median delay in scheduled hospital visits (days), by treatment arm 25 20 17,6 15 13,6 13,4 13,3 10 5 0 Control PAS only Any PAS Food Figure 8b: Late for a scheduled hospital visit (%), by treatment arm 50 45 40 39,4 35 34,2 33,9 33,6 30 25 20 15 10 5 0 Control PAS only Any PAS Food Note: Figures represent means with 95% confidence intervals. Source: Authors’ own calculations. 31    Table 8: Effects of Peer Adherence Support (PAS) and Food on Self-Reported Adherence MCA adherence CASE adherence Missed a drug Missed a routine Missed a clinic or index index CASE cut-off collection visit check-up hospital visit A. Intent-to-treat analysis Treatment -0.304 * -0.783 *** -0.072 ** 0.017 0.028 0.014 Treatment * ITT PAS only 0.086 0.275 0.041 0.052 * 0.006 0.065 * Treatment * ITT PASnFood -0.020 -0.071 -0.006 0.013 -0.038 0.001 n (N) 618 (1,270) 617 (1,268) 618 (1,270) 618 (1,267) 618 (1,263) 618 (1,263) F-test (p-value) 1.05 (0.384) 2.11 (0.062) 1.6 (0.157) 1.65 (0.143) 1.95 (0.084) 1,61 (0.154) Treatment -0.307 * -0.792 *** -0.073 ** 0.016 0.027 0.012 Treatment * ITT AnyPAS 0.034 0.106 0.018 0.033 -0.015 0.034 n (N) 618 (1,270) 617 (1,268) 618 (1,270) 618 (1,267) 618 (1,263) 618 (1,263) F-test (p-value) 1.19 (0.315) 2.19 (0.068) 1.32 (0.261) 1.68 (0.153) 1.73 (0.142) 1.20 (0.310) B. Instrumental variable (IV) analysis Treatment -0.150 -0.706 -0.080 -0.028 0.065 -0.019 Treatment * PAS only -0.085 0.223 0.084 0.094 0.011 0.120 Treatment * PASnFood -0.097 0.055 0.026 0.027 -0.099 * -0.000 n (N) 388 (794) 388 (794) 388 (794) 388 (792) 388 (790) 388 (790) F-test (p-value) 4.05 (0.670) 60,634 (<0.001) 16,730 (<0.001) 35.35 (<0.001) 31.45 (<0.001) 47.04 (<0.001) Treatment -0.150 -0.699 -0.077 -0.026 0.067 -0.017 Treatment * AnyPAS -0.093 0.113 0.046 0.051 -0.059 0.043 n (N) 388 (794) 388 (794) 388 (794) 388 (792) 388 (790) 388 (790) F-test (p-value) 4.06 (0.541) 60,703 (<0.001) 16,780 (<0.001) 33.55 (<0.001) 26.00 (<0.001) 42.80 (<0.001) Note: Results are for Fixed Effect (FE) Instrumental Variables (IV) linear panel regression models. Results are adjusted for age, sex, education and treatment duration. Source: Authors’ own calculations. 32    Table 9: Effects of Peer Adherence Support (PAS) and Food on Timeliness of Visits Clinic - delayed Hospital - delay Hospital - delayed Clinic - delay (days) (yes/no) (days) (yes/no) A. Intent-to-treat analysis Treatment 6.197 0.089 10.321 0.192 * Treatment * ITT PAS only -0.589 -0.105 7.372 -0.110 Treatment * ITT PASnFood -5.794 -0.177 * -22.587 *** -0.456 ** n (N) 192 (2,108) 192 (2,108) 114 (844) 114 (844) F-test (p-value) 2.07 (0.082) 2.37 (0.050) 3.60 (0.006) 1.78 (0.131) Treatment 6.163 0.089 10.091 * 0.189 * Treatment * ITT AnyPAS -2.979 -0.138 * -3.235 -0.232 n (N) 192 (2,108) 192 (2,108) 114 (844) 114 (844) F-test (p-value) 2.34 (0.071) 2.95 (0.031) 1.79 (0.147) 1.26 (0.288) B. Instrumental variable (IV) analysis Treatment 9.798 0.335 24.396 ** 0.525 ** Treatment * PAS only -0.650 -0.301 -1.047 -0.408 Treatment * PASnFood -10.623 -0.439 * -39.827 ** -0.821 ** n (N) 192 (2,108) 192 (2,108) 108 (801) 108 (801) F-test (p-value) 374.48 (<0.001) 793.73 (<0.001) 219.99 (<0.001) 420.28 (<0.001) Treatment 12.134 0.367 17.127 0.447 ** Treatment * AnyPAS -8.924 -0.415 * -14.296 -0.549 ** n (N) 192 (2,108) 192 (2,108) 108 (801) 108 (801) F-test (p-value) 374.24 (<0.001) 795.81 (<0.001) 214.44 (<0.001) 423.11 (<0.001) Note: Results are for Fixed Effect (FE) Instrumental Variables (IV) linear panel regression models. Results are adjusted for age, sex, education and treatment duration. Source: Authors’ own calculations. 33    Table 10: Effects of Peer Adherence Support (PAS) and Food on CD4 counts Variables [1] [2] [3] [4] PAS only 60.329 55.539 (1.96)* (1.62) PAS and Food 25.565 27.507 (0.92) (0.89) Female 14.565 14.979 18.730 18.984 (0.96) (0.99) (1.26) (1.41) ARV treatment duration (months) 65.189 65.804 66.720 65.893 (5.57)** (5.58)** (5.68)** (4.44)** Female * ARV treatment duration (months) 12.056 11.963 11.730 9.635 (4.14)** (4.11)** (4.04)** (2.65)** Age ‐9.716 ‐9.670 ‐9.423 ‐9.868 (3.71)** (3.68)** (3.62)** (4.18)** Age * ARV treatment duration (months) ‐0.633 ‐0.653 ‐0.651 ‐0.477 (1.25) (1.29) (1.28) (0.75) 2 ARV treatment duration (months) ‐2.502 ‐2.592 ‐2.980 ‐3.243 (3.81)** (3.23)** (3.94)** (3.62)** Mills ratio ‐16.024 ‐17.793 7.319 61.056 (0.14) (0.15) (0.07) (0.98) PAS only * FEATS duration (months) 3.213 7.510 12.624 (0.67) (1.84) (2.37)* PAS and Food * FEATS duration (months) ‐1.081 0.920 ‐3.097 (0.26) (0.26) (0.68) Constant 315.331 315.222 323.130 319.195 (5.15)** (5.13)** (5.33)** (5.95)** N (n) 2,733 2,733 2,733 2,733 (549) (549) (549) (549) Chi2 1,504.01 1,506.60 1,511.13 1,044.06 Note: Results are for Random Effects (RE) Instrumental Variables (IV) linear panel regression models with robust standard errors. Continuous variables are entered into the regression as square roots. Adjusted for selection bias using the Mills ratio. Source: Authors’ own calculations. 34    Figure 9: Effects of Peer Adherence Support (PAS) and Food on CD4 counts 35