Although the last decade has seen significant improvements in access to antiretroviral therapy (ART) across the developing world,1,2 widespread food insecurity and undernutrition in eastern and southern Africa, where the HIV epidemic is most severe, continue to complicate an effective response. There is substantial evidence on the interconnectedness between HIV/AIDS, food insecurity, and undernutrition.3–6 In addition to the well-documented negative impacts of HIV/AIDS on food security and other socioeconomic outcomes,7 there is growing recognition that food insecurity and undernutrition contribute to poor health outcomes among people living with HIV (PLHIV). For PLHIV, undernutrition is a strong risk factor for mortality,8,9 even among people receiving ART.10–12 Food insecurity and diet quality are associated with undernutrition and decreased quality of life among PLHIV,13,14 and diet quality is inversely associated with disease severity and mortality.15 Food insecurity is also a significant barrier to prevention,16 access to care and treatment services,8,9 and adherence to ART.10–12
National governments and international agencies consequently recognize the importance of integrating food and nutrition security interventions into their response to the AIDS epidemic.13–15,17 Presently, several HIV/AIDS service providers in sub-Saharan Africa already incorporate food and nutrition security interventions into their programming. The 2 dominant program models are nutrition supplementation, often using specialized food products, targeted to undernourished PLHIV and food assistance targeted to food insecure households affected by the epidemic. Some of the key hypothesized benefits to PLHIV from such programs include: (1) improved nutritional status, (2) slowed disease progression, (3) improved food security, (4) improved health status and quality of life, and (5) improved ART adherence and retention in care.
Although policy makers and program implementers acknowledge the potential for food security interventions to mitigate the effects of the AIDS epidemic, even after more than 10 years of programmatic experience, there is little evidence on the impacts of food assistance on PLHIV. Recent reviews identified only a small number of food-based supplementation studies in HIV/AIDS care and treatment programs in developing county settings and highlighted the need for additional research.18–20 To the best of our knowledge, there are no randomized controlled trials of the impact of food assistance in an HIV program context. Although such randomized controlled trials remain the gold standard for evidence generation, implementing them in routine program settings brings substantial challenges associated with randomizing individuals to a control group while at the same time adhering to ethical, programming, and funding guidelines that may have limited flexibility, for example, with respect to altering eligibility criteria.
When randomization is not feasible, we must turn to other approaches to establish a valid counterfactual. With previous planning, it is possible for researchers to work within routine program operations to conduct an evaluation that incorporates a strong design and applies rigorous statistical methods to estimate the causal impact of such interventions. Such evaluations, of program effectiveness rather than efficacy, reflect the real-world operating environment of many implementing organizations.21,22 Consequently, this type of evidence is especially valuable in guiding programs and policies as they integrate food-based interventions into the expansion of programs tackling the AIDS epidemic.
We capitalized on an existing intervention for ART-naive PLHIV in Uganda, coordinated by The AIDS Support Organization (TASO), a large HIV care and treatment organization, and the World Food Program (WFP) to design and conduct a quasi-experimental longitudinal study evaluating the impact of food assistance on nutritional status, disease progression, and 2 dimensions of food security: individual diet quality and access and the experiential dimension of household food security. Changes in these outcomes are among the routinely hypothesized benefits of standard food assistance programs, targeted to PLHIVs. We hypothesize that food assistance alleviates food insecurity and improves nutritional status and disease severity, through improved food intake. Additionally, with the provision of a micronutrient-fortified commodity, in this case Corn Soy Blend (CSB), we hypothesize improvements in anemia. Finally, food assistance may relax household budget constraints, and therefore, we hypothesize improvements in a household's ability to consume diverse foods.
TASO is the largest indigenous nongovernmental organization providing comprehensive HIV prevention and AIDS care and support services in Uganda. Recognizing the interconnectedness between HIV prevention, care, and treatment and food and nutrition insecurity, TASO began partnering with organizations such as WFP to provide food assistance to its clients in the early 2000s. WFP distributed monthly household food rations for 12 months to registered HIV-positive TASO clients, eligible for food assistance based on WFP's poverty assessment criteria. The criteria used to determine eligibility for food assistance covered the following domains: household composition, ownership of valuable assets, employment status and income, housing characteristics, and past experience of food insecurity. Continued food assistance eligibility was conditional on remaining an active TASO client; there were no other criteria or conditionalities. The WFP standard food basket (consisting of 200 g of maize meal, 40 g of pulses, 10 g of vitamin A–fortified vegetable oil, 5 g of iodized salt, and 50 g of micronutrient fortified CSB per person per day) provided approximately 1100 kcal/d per person for up to 7 adult household members.
To evaluate the impact of this monthly household food basket, we conducted a 12-month, quasi-experimental, longitudinal study nested within the routine programmatic context of both TASO and WFP in 2 districts in northern Uganda. The study districts, Gulu and Soroti, are both highly food insecure, with a history of conflict and displacement.23 During the study, WFP operated in Gulu but not in Soroti; thus, Gulu served as the intervention district and Soroti the nonrandomized comparison district.
We estimated that a sample size of 180 HIV-positive individuals in each of the 2 study arms was sufficient to detect a minimum increase in (1) body mass index (BMI) of 0.5 kg/m2, assuming a standard deviation of 1.9 kg/m2, (2) CD4 count of 30 cells per microliter, assuming a standard deviation of 100 cells per microliter, and (3) hemoglobin (Hb) of 0.5 g/dL, assuming a standard deviation of 1.5 g/dL. The calculation was made for a 1-sided test and based on specificity (1−α) of 95% and power (1−β) of 80%. This sample was increased to 405 to accommodate a design effect of 2.25. Allowing for 10% loss to follow-up, we recruited 450 individuals per study arm. This sample size also allowed us to detect changes in 2 additional outcomes: individual dietary diversity24 and household food access.25
We recruited HIV-positive non-pregnant adults (over 18 years of age), during their visits to the TASO clinics in Gulu and Soroti, using the following criteria: (1) initial eligibility for food assistance based on WFP's poverty assessment criteria described above, (2) nonreceipt of food assistance from any source in the previous 12 months (based on self-report), (3) CD4 count between 200 and 450 cells per microliter, and (4) ART naive. Contact information of study participants in Gulu were given to WFP after baseline interviews were completed. Food distribution, conducted at a location away from the TASO clinic, began 7–30 days after recruitment and lasted 12 months.
Multipurpose individual and household level surveys were administered at baseline recruitment and at approximately 12 months follow-up. On recruitment, an individual questionnaire was administered to the study participant in a private room at the TASO clinic. Trained and standardized anthropometrists took anthropometric measurements and a TASO laboratory technician drew blood for CD4 count and Hb concentration. Within 7 days, a study team member visited the home of the participant to administer the household questionnaire. The interviewer was trained to maintain strict confidentiality and did not refer to TASO or the HIV status of the study participant in the household interview.
The protocol was the same for the 12-month follow-up interview. We were unable, however, to reinterview all study participants 12 months after recruitment. Two developments outside our control necessitated either earlier than 12-month follow-up interviews, or exclusion, for some participants: (1) individuals becoming eligible for ART after recruitment, which may have been accelerated by changes in WHO recommendations for ART eligibility based on CD4 counts (WHO 2010) and (2) another organization offering food assistance to select individuals in the comparison district after study initiation (Fig. 1). We reinterviewed as many of the affected individuals as possible before receipt of ART or food assistance. Consequently, whereas 77% of study participants were reinterviewed between 10 and 13 months, 15% were interviewed before 10 months (mean, 8.2; SD, 1.8) and 8% after 13 months (mean, 14.6; SD, 2.2). If such previous interview was not feasible, the individual was considered to be lost to follow-up.
The principal outcome variables include changes in: (1) BMI (kg/m2), MUAC (mm), and Hb concentration (g/dL) as measures of nutritional status, (2) CD4 count (cells/μL) as a measure of disease progression, and (3) diet quality using the Individual Dietary Diversity Score (IDDS)24 and food access using the Household Food Insecurity Access Scale (HFIAS)25 as measures of food security; and the number of self-reported physical symptoms associated with HIV in the previous 30 days, from a list of 16 items, including weight loss, skin rash, body pains, and dizziness, among others. The IDDS and HFIAS are validated scales widely used in food security programming and research. The HFIAS consists of 9 items and 4 frequency responses using a 30-day recall period and was developed to reflect 3 apparently universal domains of the experience of inadequate household-level food access: (1) anxiety about household food supply, (2) insufficient quality, which includes variety and preferences, and (3) insufficient quantity of food supply, the amount consumed, and the physical consequences of insufficiency. IDDS is the sum of different food groups consumed in the 24-hour preceding the recall and is a proxy measure of the quality of food at the individual level. Hb was included as an outcome given the inclusion of micronutrient fortified CSB in the food assistance transfer. Viral load measures were not available for study participants.
To estimate the impact of food assistance, we employed nonexperimental matching techniques combined with difference-in-difference estimators.26 We compared the change over time in outcomes for individuals in the intervention district with the change over time for matched individuals in the comparison district. This approach is increasingly used in impact evaluations when randomization is infeasible.27–30
First, guided by theoretical considerations and exploiting the rich individual and household survey data, we determined a set of “balance” variables on which the matches would be based (Table 1), including baseline values of the key outcome measures.31–34 Second, we constructed a propensity score for each individual. The propensity score model was the logit of living in the intervention district as explained by all of these variables measured at baseline. We transformed some of the underlying variables used in the logit until the distributions of the variables for each quantile of the propensity score were statistically the same across intervention and comparison individuals, a procedure referred to as balancing.34,35 We then used the propensity score, as well as a subset of the balance variables (logarithmic per capita expenditures, the share of food to total household expenditures, and the time between surveys), to match each individual in the intervention district to the most similar individual, or “nearest neighbor,” of the same gender in the comparison district.26 The estimated average treatment effect on the treated (ATT), then, is the difference-in-difference in mean outcomes over 12 months for the intervention group compared with the matched comparison group. We implemented a heteroskedasticity robust variance estimator developed for this nearest neighbor matching technique.26
One concern is that despite all of these controls, there still may have been different unobservable trends in the 2 districts influencing nutritional status or disease progression, but not captured by our matching variables. We therefore replicated the above analyses on a subsample for which, in addition to baseline CD4 count, we had available a previous CD4 measurement (from TASO records). Using this earlier measurement, we calculated the average monthly change in CD4 count before the baseline and included it as an additional matching variable (second row for select outcomes in Table 3). In this analysis, then, we control for previous trends in disease progression for each individual. We do this only for the outcomes most likely to be influenced by such trends—the individual level nutritional outcomes.
We also undertook several analyses to assess the sensitivity of the results to the choice of methodology and matching variables used, as estimates may be sensitive to such choices.34 These included the following: (1) alternative matching methods such as nearest neighbor matching with between 1 and 5 neighbors. Results reported in Table 3 are from nearest neighbor matching with the 1 and 3 nearest neighbors; (2) alternative sets of matching variables, for example, including all of the balancing variables directly in the matching procedure (as opposed to including only the propensity score and a subset of the variables), while always matching exactly on gender; (3) alternative subsamples, in particular limiting to the sample for which estimated propensity scores were between 0.1 and 0.9 only; and (4) an alternative estimation procedure using Gaussian kernel matching based only on the estimated propensity scores and with bootstrapped standard errors using 1000 replications.34,36
All analyses were performed using Stata version 12.37 We set statistical significance at a 2-tailed P < 0.05.
The ethics review boards of TASO, the Uganda National Council on Science and Technology, and the International Food Policy Research Institute (IFPRI) approved the study protocol. Study participants provided signed informed consent.
Between August 2008 and October 2009, we recruited a total of 904 study participants from the intervention and comparison districts and followed up 640 participants between August 2009 and October 2010 (Fig. 1). We did not reinterview individuals who began ART, lived in the comparison district and had been provided food assistance from another program, could not be tracked, or died.
Baseline characteristics are presented for the intervention group (column A), the full comparison group (column B), and the weighted matched comparison group (column C) (Table 2). Average baseline characteristics for the intervention and weighted matched comparison groups are very similar for most indicators, including the time interval between baseline and follow-up surveys, providing evidence of the credibility of the matching exercise. There are 3 significant differences between the treatment and matched comparison (age, HFIAS, and number of self-reported symptoms), but all are relatively small (<0.2 SDs).
ATT estimates are presented in Table 3. For each outcome, we present impact estimates for the complete sample of 318 observations. In addition, for the nutritional status outcomes, which are likely to have been influenced by previous CD4 levels, we present estimates that control for previous CD4 trend on a subsample of 126 observations. Food assistance increased BMI by 0.6 kg/m2 and MUAC by 6.7 mm. These impacts were larger, though less significant, when controlling for CD4 trend (estimates based on half of the total sample). They were also similar, and more highly significant, when matching to the 3 nearest neighbors. We find no statistically significant impact of food assistance on Hb concentration or CD4 count. However, there were large and significant impacts on Hb concentration (increases of approximately 1.0 g/dL; results not presented) when restricting the analysis to individuals with CD4 counts >350 cells per microliter.
There was no statistically significant effect on individual dietary diversity, based on previous day consumption of the 9 different food groups measured in the IDDS. Consumption of the food group categories containing 2 of the components of the food basket (beans and cooking oil), however, showed statistically significant improvements. At the household level, we find large significant decreases in food insecurity as a result of the intervention; the mean HFIAS score decreased by 2.1 points, equivalent to an approximately 50% point reduction in the probability of being designated as severely food insecure. Finally, there was a statistically significant 4.1 reduction in the number of reported symptoms related to HIV in the previous 30 days.
Although many have theorized about the potential benefits of integrating food assistance into HIV care and treatment programs operating in food insecure areas, and such combined program models have been widely implemented, there is little evidence documenting their impacts. We demonstrated that food assistance, provided to households of ART-naive PLHIV in northern Uganda, significantly improved BMI, MUAC, and household food security, compared with a matched comparison group receiving otherwise similar HIV care and treatment. We found no impact of food assistance on CD4 count or IDDS. Impacts on Hb concentrations were seen only among those with less advanced disease progression.
We selected individuals with CD4 counts between 200 and 450 cells per microliter because of the vulnerability of these individuals with compromised immunity, who were at the time of the study not eligible for ART, but were at a critical point of disease progression. Targeted interventions during this vulnerable period could delay the need for ART initiation. We wanted to test whether food assistance can improve nutritional status and food security, 2 key facilitators of ART success when initialized. Additionally, it is important to note that not all individuals who qualify for ART receive it, particularly in resource-limited settings,38 and long-term dropout rates are high.39 Therefore, understanding the impact of food assistance on people not receiving ART has wider policy implications.
Our study contributes to the literature in 3 important ways. First, it is one of only a handful of studies investigating the impact of food assistance to adult PLHIV in a developing country context29,40–42; and the first to focus exclusively on ART-naive adult PLHIV and their households. Second, the design and analytical techniques are more rigorous than previous published studies. We recruit intervention and comparison subjects, based on an identical set of program eligibility criteria and use difference-in-difference propensity score matching methods, using multiple matching variables, to estimate the impact of food assistance. In addition to avoiding any selection bias or confounders associated with several explicit eligibility criteria, we attempt to control for all other observable differences between groups, except the presence or absence of food assistance. Third, this study examines a broad range of individual and household level outcomes (nutritional status, disease progression, food security) of interest in HIV and food security programming.
Results from the few published observational studies examining the impact of food assistance on nutritional status among PLHIV in developing countries are inconsistent. In a prospective cohort study in Haiti, a monthly WFP family food transfer to adult PLHIV was associated with improved BMI at 6 and 12 months, compared with individuals who did not receive food assistance.41 In that study, the food assistance and the control group differed substantially at recruitment since, by design, individuals who did not meet the eligibility criteria for food assistance were enrolled as the comparison group. In a retrospective cohort study from Uganda using a program administrative database, individuals receiving food assistance for a period of 12 months had a higher mean weight gain compared with their matched control and had statistically significant, but biologically insignificant, slower disease progression.29 In Zambia, a monthly WFP individual or household food ration was not associated with weight gain or BMI increases at 6 and 12 months among PLHIV on ART, compared with a control group.40 Finally, a cohort study in India found a small daily WFP individual supplement provided to adult PLHIV on ART was not associated with improvements in weight, BMI, MUAC, or Hb concentrations at 6 months, compared with those in 2 comparison clinics.42
There seems to be consistency in the published literature on the lack of impact of food assistance on immunological outcomes. Similar to our study, results from Zambia40 and India42 show no impacts of food assistance on changes in CD4 counts. Ours is the only study to examine impacts on CD4 count of exclusively ART-naive individuals. There is also consistency in the published literature on the impacts of food assistance on ART adherence. Because our study was among ART-naive PLHIV, adherence as a result of food assistance was not a measurable outcome.
The lack of any impact on Hb concentrations was unexpected, and we propose 3 possible explanations for it. First, the prevalence of anemia in this population was low (15%), and therefore, the potential to benefit at the population level was muted. Second, we are unable to determine whether the anemia that is present is due to iron deficiency or HIV infection itself, an important distinction when examining the potential of a nutritional intervention to improve Hb concentrations. This is supported by the positive impacts of food assistance on Hb concentrations only among those individuals with less advanced disease progression. It is plausible that in these individuals, the comparatively lower level of immunosuppression may provide greater potential to benefit from nutritional interventions, compared with those with more severe immunosuppression in whom Hb concentrations may be more strongly influenced by disease severity. Third, although the food assistance basket contained micronutrient-fortified commodities (CSB and vitamin A–fortified vegetable oil), the composition of the basket may be inadequate to meet the micronutrient requirements of PLHIV, as they are for infants,43 and therefore, might not be expected to have an impact on anemia. The food basket provided approximately 2.15 mg or iron per day, per adult, based on the provision of 50 g of CSB per adult per day.18 This provides approximately 25% of the iron recommended dietary allowance (RDA) for male adults (8 mg/d), and 12% of the RDA for reproductive aged females (18 mg/d).44 There are no specific iron RDA guidelines for PLHIVs. Additionally, we are unable to determine the actual quantity of CSB (or other commodities) consumed by individuals in our study.
The results on dietary diversity should be interpreted with caution. IDDS captures food groups consumed (and had been shown to be associated with micronutrient density of the diet among infants, and micronutrient adequacy among women), but it does not capture within food group diversity. It is possible that the food assistance may have had an impact on the total number of food items consumed within a group. Further research is required to understand the impact that food and nutrition security interventions have on overall dietary patterns within the context of HIV.
Several aspects of the results increase confidence in the findings. First, there is internal consistency between the impacts on BMI, MUAC, and household food security, suggesting that improved household food security is a likely mechanism underlying the observed improvements in nutritional status. Second, our extensive analyses, including the use of alternative matching techniques to test the sensitivity of the results (see Statistical Methods) yielded similar results. Third, we applied the same methodology to individual self-reports of 16 physical symptoms associated with HIV. There were highly significant reductions in the number of reported symptoms consistent with improved health. Estimation of reports of the individual symptom of losing weight was also significantly negative, consistent with the results for measured BMI and MUAC.
There are several limitations to our study. First, because the comparisons were drawn from a different geographical area, it is possible that unobserved geographical, sociocultural, or other factors explain part of the observed differences between groups. Gulu district, for example, may have suffered more intensively from civil conflict in the past,23 though it is difficult to assess whether and how this would alter the impact of food assistance on PLHIV. We attempted to mitigate this potential bias by differencing the outcomes (thereby controlling for time invariant factors) and by including a large number of matching variables, many of which reflect conditions specific to the geographic areas and trends in those areas. Second, we are unable to document actual food consumption by target PLHIV who received the household food transfer. Receipt of the household food transfer was monitored by the program, but beyond the food distribution point, no detailed assessment of individual-level intake was made. Third, a relatively large proportion of individuals (21%) became eligible for and received ART, during the study period and were therefore excluded from the impact analysis. We examined baseline characteristics of individuals who were put on ART, and found no significant differences by study group. Finally, we are unable to examine the sustainability of impacts over the long term, an outcome of interest to program implementers and policy makers. We have however, published results from a study, within the same study context, that investigates the transition of beneficiaries from food assistance to sustainable livelihoods programs, which is one element of program sustainability that implementers are exploring.45
Although the past few years have seen a dramatic expansion of ART access across sub-Saharan Africa, significant challenges to universal access persist.46,47 Food insecurity remains a critical barrier that compromises the effectiveness of HIV programs. By conducting this study in a routine program context, and in coordination with one of the largest providers of food assistance in the region, the results of this study are valuable to programmers and policy makers as they have a degree of external validity that would not be possible if the evaluation had altered the program design. This study demonstrates the benefit of food assistance in the absence of ART and highlights the potential for food assistance programming to be part of the standard of care for PLHIV in areas of widespread food insecurity.
The authors gratefully acknowledge the support of staff from TASO in Uganda, particularly those at the TASO centres in Gulu and Soroti who were directly involved in the study; the field team and the field supervisor, Moses Odeke, for their efforts in data collection. Robert Ochai and Frances Babirye at TASO headquarters provided tremendous support to the study overall and were instrumental in the success of this study. The authors thank WFP in Uganda for its support of the study. At IFPRI, Wahid Quabili provided support in data management and cleaning. Finally, they thank all TASO clients for their participation.
1. World Health Organization. HIV/AIDS Fact Sheet. Geneva, Switzerland: World Health Organization; 2012.
2. Joint United Nations Programme on HIV/AIDS (UNAIDS). UNAIDS World AIDS Day 2012 Global Fact Sheet. Geneva, Switzerland: Joint United Nations Programme on HIV/AIDS; 2012.
3. Gillespie S, Kadiyala S. HIV/AIDS and Food and Nutrition Security: From Evidence to Action. Washington DC, WA: International Food Policy Research Institute; 2005.
4. Food and Agriculture Organization. HIV/AIDS, Food Security and Rural Livelihoods. Geneva, Switzerland: Food and Agriculture Organization; 2002.
5. Barnett T, Whiteside A. AIDS in the Twenty-first Century: Disease and Globalization. New York, NY: Palgrave Macmillan; 2008:485–487.
6. Joint United Nations Programme on HIV/AIDS (UNAIDS). HIV, Food Security and Nutrition Policy Brief. Geneva, Switzerland: UNAIDS; 2008.
7. Chapoto A, Jayne TS. Socioeconomic characteristics of individuals afflicted by AIDS related prime age mortality in Zambia. In: Gillespie S, ed. AIDS, Poverty, and Hunger: Challenges and Responses. Washington DC, WA: International Food Policy Research Institute; 2006:33–55.
8. Tang AM. Weight loss, wasting, and survival in HIV-positive patients: current strategies. AIDS Read. 2003;13:S23–S27.
9. van der Sande MAB, van der Loeff MFS, Aveika AA, et al.. Body mass index at time of HIV diagnosis—a strong and independent predictor of survival. J Acquir Immune Defic Syndr. 2004;37:1288–1294.
10. Mangili A, Murman DH, Zampini AM, et al.. Nutrition and HIV infection: review of weight loss and wasting in the era of highly active antiretroviral therapy from the nutrition for healthy living cohort. Clin Infect Dis. 2006;42:836–842.
11. Johannessen A, Naman E, Ngowi BJ, et al.. Predictors of mortality in HIV-infected patients starting antiretroviral therapy in a rural hospital in Tanzania. BMC Infect Dis. 2008;8:52.
12. Zachariah R, Fitzgerald M, Massaquoi M, et al.. Risk factors for high early mortality in patients on antiretroviral treatment in a rural district of Malawi. AIDS. 2006;20:2355–2360.
13. Kadiyala S, Rawat R. Food access and diet quality independently predict nutritional status among people living with HIV in Uganda. Public Health Nutr. 2013;16:164–170.
14. Palermo T, Rawat R, Sheri D, et al.. Food access and diet quality are associated with quality of life outcomes among HIV-infected individuals in Uganda. PLoS One. 2013;8:e62353.
15. Rawat R, McCoy SI, Kadiyala S. Poor diet quality is associated with low CD4 count and anemia and predicts mortality among antiretroviral therapy-naive HIV-positive adults in Uganda. J Acquir Immune Defic Syndr. 2013;62:246–253.
16. Weiser SD, Leiter K, Bangsberg DR, et al.. Food insufficiency is associated with high-risk sexual behavior among women in Botswana and Swaziland. PLoS Med. 2007;4:1589–1598.
17. Frega R, Duffy F, Rawat R, et al.. Food insecurity in the context of HIV/AIDS: a framework for a new era of programming. Food Nutr Bull. 2010;31:S292–S312.
18. de Pee S, Semba RD. Role of nutrition in HIV infection: review of evidence for more effective programming in resource-limited settings. Food Nutr Bull. 2010;31:S313–S344.
19. Mahlungulu S, Grobler LA, Visser ME, et al.. Nutritional interventions for reducing morbidity and mortality in people with HIV. Cochrane Database Syst Rev. 2007;CD004536.
20. Tirivayi N, Groot W. Health and welfare effects of integrating AIDS treatment with food assistance in resource constrained settings: a systematic review of theory and evidence. Soc Sci Med. 2011;73:685–692.
21. Habicht JP, Victora CG, Vaughan JP. Evaluation designs for adequacy, plausibility and probability of public health programme performance and impact. Int J Epidemiol. 1999;28:10–18.
22. Victora CG, Habicht JP, Bryce J. Evidence-based public health: moving beyond randomized trials. Am J Public Health. 2004;94:400–405.
23. Vinck P. Exposure to war crimes and implications for peace building in northern Uganda. JAMA. 2007;298:543–554.
24. Arimond M, Wiesmann D, Becquey E, et al.. Simple food group diversity indicators predict micronutrient adequacy of women's diets in 5 diverse, resource-poor settings. J Nutr. 2010;140:2059S–2069S.
25. Coates J, Swindale A, Bilinsky P. Household Food Insecurity Access Scale (HFIAS) for Measurement of Food Access: Indicator Guide (V3). Washington DC, WA: Food and Nutrition Technical Assistance Project, Academy for Education; 2007.
26. Abadie A, Drukker D, Leber J, et al.. Implementing matching estimators for average treatment effects in Stata. Stata J. 2004;4:290–311.
27. Imbens GW, Wooldridge JM. Recent developments in the econometrics of program evaluation. J Econ Lit. 2009;47:5–86.
28. Donegan S, Maluccio JA, Myers C, et al.. Two food assisted maternal and child health nutrition programs helped mitigate the impact of economic hardship on child stunting in Haiti. J Nutr. 2010;140:1139–1145.
29. Rawat R, Kadiyala S, McNamara P. The impact of food assistance on weight gain and disease progression among HIV-infected individuals accessing AIDS care and treatment services in Uganda. BMC Public Health. 2010;10:316.
30. Mahal A, Canning D, Odumosu K, et al.. Assessing the economic impact of HIV/AIDS on Nigerian households: a propensity schore matching approach. AIDS. 2008;22:s95–s101.
31. Abadie A, Imbens GW. Large sample properties of matching estimators for average treatment effects. Econometrica. 2006;74:235–267.
32. Heckman J, Navarro LS. Using matching, instrumental variables, and control functions to estimate economic choice models. Rev Econ Stat. 2004;86:30–57.
33. Smith J, Todd P. Does matching overcome LaLonde's critique of non-experimental estimators? J Econom. 2005;125:305–353.
34. Todd P. Evaluating social programs with endogenous program placement and selection of the treated. In: Schultz TP, Strauss J, eds. Handbook of Development Economics. 4th vol. New York, NY: Elsevier; 2008:3847–3894.
35. Leuven E, Sianesi B. Stata module to perform full Mahalanobis and propensity score matching, common support graphing, and covariate imbalance testing. 2003.
36. Heckman JJ, Ichimura H, Todd P. Matching as an econometric evaluation estimator. Rev Econ Stud. 1997;64:605–654.
37. StataCrop [Computer program]. Texas: Stata Corporation; 2011.
38. Vella S. The history of antiretroviral therapy and of its implementation in resource-limited areas of the world. AIDS. 2012;26:1231–1241.
39. Rosen S, Fox MP, Gill CJ. Patient retention in antiretroviral therapy programs in sub-Saharan Africa: a systematic review. PLoS One. 2007;4:e298 1691–1700.
40. Cantrell RA, Sinkala M, Megazinni K, et al.. A pilot study of food supplementation to improve adherence to antiretroviral therapy among food-insecure adults in Lusaka, Zambia. J Acquir Immune Defic Syndr. 2008;49:190–195.
41. Ivers LC, Chang Y, Gregory JJ, et al.. Food assistance is associated with improved body mass index, food security and attendance at clinic in an HIV program in central Haiti: a prospective observational cohort study. AIDS Res Ther. 2010;7:33.
42. Swaminathan S, Padmapriyadarsini C, Yoojin L, et al.. Nutritional supplementation in HIV-infected individuals in South India: a prospective interventional study. Clin Infect Dis. 2010;51:51–57.
43. Ruel MT, Menon P, Loechl C, Pelto G. Donated fortified cereal blends improve the nutrient density of traditional complementary foods in Haiti, but iron and zinc gaps remain for infants. Food Nutr Bull. 2004;25:361–376.
44. National Research Council. Dietary Reference Intakes for Vitamin A, Vitamin K, Arsenic, Boron, Chromium, Copper, Iodine, Iron, Manganese, Molybdenum, Nickel, Silicon, Vanadium, and Zinc. Washington DC, WA: The National Academies Press; 2001.
45. Yager JE, Kadiyala S, Weiser SD. HIV/AIDS, food supplementation and livelihood programs in Uganda: a way forward? PLoS One. 2011;6:e26117.
46. Joint United Nations Programme on HIV/AIDS (UNAIDS). New Reports Show Philanthropic Funding for AIDS Down at Pivotal Moment in the Response. Geneva, Switzerland: UNAIDS; 2011.
47. Amuron B, Namara G, Birungi J, et al.. Mortality and loss-to-follow-up during the pre-treatment period in an antiretroviral therapy programme under normal health service conditions in Uganda. BMC Public Health. 2009;9:290.