Introduction

Food insecurity is defined as the lack of access, availability, and utilization of sufficient and diverse foods [1] and tends to be more common among HIV-positive individuals than among HIV-negative individuals [2, 3]. The adverse effects of food insecurity on HIV-related outcomes are well documented and include suboptimal adherence to combination antiretroviral therapy (cART) [4,5,6,7], a detectable HIV viral load [5, 8], increased mortality [9], and lower CD4 cell count [5, 10]. Although food insecurity tends to be common among HIV-positive adults initiating cART [8, 9, 11,12,13,14], little is known about how cART initiation subsequently influences food insecurity. On one hand, cART may improve one’s physical ability to engage in agriculture or other income-generating activities, which may in turn reduce food insecurity [12, 15,16,17]. On the other hand, cART increases appetite, which may in turn exacerbate food insecurity by increasing the quantity of food needed to satisfy the household [13, 16, 18, 19].

Changes in food insecurity following initiation of cART may depend on whether the patient receives a food ration [20, 21]. Provision of a food ration is recommended by international agencies, including the World Health Organization [22], Joint United Nations Programme on HIV/AIDS [23], and World Food Programme [24], as a key component of comprehensive HIV care [25]; however, competing program needs and limited resources have prevented the implementation of this recommendation in many settings. It is possible that food support has an independent effect on food insecurity or that it synergizes with cART to yield greater reductions in food insecurity than those that could be achieved with cART alone [26]. An understanding of how food insecurity changes during the first year of cART and whether these changes are influenced by the receipt of food support, is critical to enabling programs to understand and anticipate food insecurity challenges in their patient populations.

In this study, we examined how food insecurity changed in two groups of HIV-positive adults during the first 12 months of cART. One group of individuals received clinic-based cART care, which did not include food support, while one group additionally received a comprehensive, community-based intervention that included a monthly food ration. We hypothesized that any reduction in food insecurity during the first 12 months of cART would be greater among those receiving a food ration through the community-based intervention.

Methods

Study Setting

This study was conducted in Rwanda, where 85% of the population lives in rural areas, and individuals are more than twice as likely to be living in extreme poverty or to be poor [27]. Food insecurity remains a major concern, especially in rural areas where 28% of the population is food insecure and 24% is highly vulnerable to food insecurity [28]. The national prevalence of HIV in Rwanda is 3% [29], and 91% of those who need cART are receiving it [30]. Throughout the study period, cART was provided free of charge using a clinic-based model of delivery and was initiated according to Rwanda national guidelines (for individuals with a CD4 cell count <350 cells/μL, or World Health Organization HIV disease stage III or IV) [29].

Community-Based Accompaniment (CBA) with Food Support

As part of efforts to scale-up access to HIV services, including cART, community-based accompaniment (CBA) was implemented by Partners In Health, in partnership with the Rwanda Ministry of Health in select rural districts in Eastern Rwanda beginning in 2005. The target catchment area and health facilities were identified by the Rwanda Ministry of Health. The cornerstone of the CBA model consisted of daily home visits by a community health worker who directly observed treatment at least once per day, provided social support, identified socioeconomic barriers to care, and monitored patients for treatment-related adverse events. Patients received a transportation stipend for routine clinic visits during the first 10 months of cART, and socioeconomic interventions (e.g., housing assistance, school fees, and microfinance) were offered to those with grave socioeconomic need [31]. Food support in the form of a monthly food ration based on a family of four was provided for the first 10 months of cART and was available for pick up after routine clinic visits. For the first 4 months, the monthly food ration included: corn flour (38 kg), beans (10 kg), oil (4 L), sugar (4 kg), and sosoma (4 kg), and cost $36 per month per family. For the last 6 months, the monthly food ration included: corn flour (18 kg), beans (5 kg), oil (4 L), sugar (2 kg), and sosoma (2 kg) and cost $20 per family.

Study Population

We performed secondary analyses of data from a prospective cohort study of the clinical and psychosocial benefits associated with the addition of CBA with food support, when added to a clinic-based cART delivery model in rural Rwanda [32]. Between June 2007 and August 2008, 610 patients were enrolled from five health centers in Kayonza/Kirehe districts, where CBA with food support was provided in addition to the clinic-based care (N = 304), and from four health centers in Musanze district, which provided clinic-based care without CBA (N = 306). Musanze district was chosen for comparison because it was considered to have a well-functioning governmental cART program but lacked a nongovernmental organization implementation partner adding additional support for cART delivery. Participation was restricted to adults (≥21 years) who were initiating lifelong cART for the first time. We previously reported baseline differences in the two study groups: patients receiving CBA with food support tended to have less severe disease (e.g., higher CD4 cell counts, lower HIV disease stage), lower literacy, live further from the clinic, and have more depression, lower BMIs, and lower health-related quality of life [32]. We also found that CBA was associated with improved rates of retention with viral load suppression, greater reductions in depression and larger gains in perceived social support, and physical and mental health quality of life [32, 33].

The study was approved by the Rwanda National Ethics Committee and Partners Human Research Committee, and all study participants provided written informed consent.

Data Collection

Study workers conducted interviews with participants at the time of cART initiation (i.e., baseline), and after 3, 6, and 12 months using standardized questions to assess sociodemographic and psychosocial characteristics, and food insecurity. Sociodemographic information included sex, age, civil status, literacy, income, and travel time and cost to reach the health facility. Interviews were conducted in the local language, Kinyarwanda. Clinical data, including CD4 cell count, HIV disease stage, tuberculosis disease, height, and weight, were prospectively recorded by clinicians and entered into an electronic medical record that was used for study purposes.

Instruments

All instruments were translated from English to Kinyarwanda and then back-translated to English to ensure accuracy. We assessed food insecurity using the Household Food Insecurity and Access Scale (HFIAS) developed by the Food and Nutrition Technical Assistance Program, which has been validated in Tanzania [34] has been widely used to assess food insecurity in HIV-positive patients in sub-Saharan Africa [35,36,37] and demonstrated good internal consistency in this cohort (Cronbach’s alpha: 0.77). Participants were asked nine questions regarding the frequency of food inadequacy, lack of variety, size of meals and hunger during the last 30 days. Responses fell into four categories ranging from never to often (score 0–3) for each question. Numeric scores were summed into a continuous score of 0–27 (least to most food insecure), and food insecurity was categorized as secure, mildly food insecure, moderately food insecure, or severely food insecure using the responses to specific questions as shown in the categorization scheme in the HFIAS Indicator Guide [38]. We focused analyses on moderate and severe food insecurity because households meeting one of these classifications are characterized by a reduction in food intake, either by reducing the size or number of meals. A household with severe food insecurity is one in which the meal size or number of meals was decreased often, and/or in which any of the three most severe conditions (running out of food, going to bed hungry, or going a whole day and night without eating) was reported [38].

Depression was measured using the Hopkins Symptom Checklist-15 (HSCL-15) which asks respondents to classify 15 symptoms (such as low energy, loneliness, and poor appetite) during the past week. Responses are scored on a severity scale from 1 to 4, with a higher score indicating more depressive symptoms. Individuals with a mean score greater than 1.75 were classified as depressed [39]. The mental health and physical health subscales of the Medical Outcomes Survey HIV Scale were used to assess self-perceived mental and physical health quality of life [40]. The mental health subscale is comprised of five items that reflect major mental health dimensions including anxiety, depression, loss of behavioral/emotional control, and psychological wellbeing. The six items of the physical health subscale reflect physical function, pain, and role functioning. Subscales are scored through summation of the response choice item values, with higher scores indicating better health. To facilitate comparability, subscale scores are transformed to a 0–100 scale [40]. To measure social support, we used the 8-item Duke UNC Functional Social Support Questionnaire (DUFSSQ). The DUFSSQ asks respondents to classify supportive situations such as ‘‘You get help when you are sick in bed’’ by the amount of support received relative to their ideal, and their responses are scaled 1–5 with higher scores indicating greater social support (range of possible scores 8–40) [41]. We previously reported good reliability and construct validity of these scales in this cohort [42].

Statistical Analysis

We compared food insecurity at 3, 6, and 12 months, relative to baseline, and stratified these comparisons by whether the participants received care at a site providing CBA with food support or the clinic-based model of cART alone. First we calculated the median HFIAS scores and compared them at 3, 6 and 12 months, relative to baseline, using Friedman’s test. Next, we used McNemar’s test to test for differences in the proportions of people with [1] severe food insecurity; and [2] moderate or severe food insecurity at baseline and each of the follow-up time points. These two approaches used all available measurements at a given time point (i.e., anyone who completed at least one HFIAS assessment was included). We conducted a sensitivity analysis in which we calculated the within-person change in food insecurity score from baseline to each of the three follow-up time points and tested for differences using one sample t-tests. To be included in this analysis, an individual was required to have a baseline HFIAS assessment and at least one follow-up HFIAS assessment.

To examine whether any association between CBA with food support and severe food insecurity and moderate or severe insecurity at 12 months was independent of baseline differences between the groups, we conducted multivariable binomial regression analyses and Poisson regression with a robust variance estimator to calculate the relative risk of having a favorable program outcome. We included in the multivariable model all potential baseline confounders of the relationship between delivery model and severe food insecurity. To identify potential confounders, we first identified biologically- or socially-plausible food insecurity risk factors and then examined whether they were associated with both the receipt of CBA and severe food insecurity at 12 months at a p value <0.20 in univariable analyses. Due to few missing data, the complete case method was used. To assess the degree of unmeasured confounding that would need to be present to completely explain our findings, we calculated the E-value, which represents the minimum strength of association on the risk ratio scale that an unmeasured confounder would need to have on both exposure and outcome, conditional on covariates, to explain away an observed association [43]. We reported E-values for the adjusted observed risk ratio as well as the confidence interval limit closest to the null value of one.

Because the food ration was provided in the context of CBA, we were unable to examine the role of food rations alone in reducing food insecurity. We conducted two secondary analyses to gain insight on the mechanism through which CBA with food support might lead to reductions in food insecurity. First, a small subset of individuals in the clinic-based care group received a food ration from an external source during the study period; we therefore conducted a secondary analysis in which we compared the prevalence of severe food insecurity at 12 months among patients in this group who did and did not receive a food ration. Second, we considered whether community-based support might lead to improved health and ability to work, independently of the food ration, which could in turn improve food insecurity. If this were true one might expect changes in clinical outcomes, such as CD4 cell count or body mass index, to correlate with changes in food insecurity. We therefore calculated correlations for change in CD4 count, BMI, and food insecurity between baseline and 12 months and stratified by delivery model. All analyses were completed using SAS University Edition v3.3 (Cary, NC: SAS Institute Inc.) and statistical significance was established at α = 0.05.

Results

Of the 610 adults included in the larger cohort study, nine (1.5%) who lacked a food insecurity assessment at baseline were excluded from baseline descriptive analyses and those examining change in food insecurity from baseline. Figures 1 through 3 show the prevalence of food insecurity, over time, stratified by receipt of CBA with a food ration.

Fig. 1
figure 1

Food insecurity scores among HIV-positive adults in Rwanda, during the first year of cART

Full size image

Change in Median Food Insecurity Score Over Time

Among individuals receiving food support under the CBA model, median food insecurity scores were lower after 3 months (−7; Friedman’s test statistic = 112.97; p-value = <0.0001), 6 months (−7; Friedman’s test statistic = 89.56; p-value = <0.0001), and 12 months (−9; Friedman’s test statistic = 136.38; p-value = <0.0001) of cART, relative to baseline (Fig. 1). Conversely, in the clinic-based model that did not receive food support, the food insecurity scores were similar to baseline after 3 and 12 months and significantly higher after 6 months (+7; Friedman’s test statistic = 18.13; p-value = <0.0001) (Fig. 1).

Proportion with Severe and Moderate or Severe Food Insecurity

Among individuals receiving food support under the CBA model, the proportion of severe food insecurity was lower at 3 months (−32%; McNemar’s test statistic = 60.83; p-value = <0.0001), 6 months (−28%; McNemar’s test statistic = 61.54; p-value = <0.0001), and 12 months (−42%; McNemar’s test statistic = 89.63; p-value = <0.0001), relative to baseline (Fig. 2). For those who did not receive CBA, the proportion with severe food insecurity was not significantly different after 3 months (−4%; McNemar’s test statistic = 2; p-value = 0.20), 6 months (+4%; McNemar’s test statistic = 2.18; p-value = 0.18), or 12 months (0%; McNemar’s test statistic = 0.20; p-value = 0.74), relative to baseline (Fig. 2). We observed a similar pattern when we repeated this analysis for the proportion of people with moderate or severe food insecurity at each site: statistically significant decreases in moderate or severe food insecurity were observed among participants at sites offering CBA with food support, but not among sites that received clinic-based care without CBA (Fig. 3).

Fig. 2
figure 2

Percentage of HIV-infected adults in Rwanda with severe food insecurity during the first year of cART

Full size image
Fig. 3
figure 3

Percentage of HIV-infected adults in Rwanda with moderate or severe food insecurity, during the first year of cART

Full size image

Within Person Change in Food Insecurity Score

On an individual level, those in the CBA with food support group experienced a statistically significant decrease in food insecurity from baseline to 3 months (6 point decrease; t-statistic = 13.07; p-value = <0.0001), from baseline to 6 months (6 point decrease; t-statistic = 12.61; p-value = <0.0001), and from baseline to 12 months (8 point decrease; t-statistic = 15.44; p-value = <0.0001). In contrast, among those in the clinic-based group, food insecurity remained similar from baseline to 3 and 12 months. From baseline to 6 months, those in the clinic-based group experienced a statistically significant increase (2 point increase; t-statistic = −4.29; p-value = <0.0001) in food insecurity.

CBA with Food Support and Severe Food Insecurity at 12 Months

Of the 610 study participants, 498 (82%) had a 12-month food insecurity assessment and were included in analyses of risk factors for severe food insecurity at 12 months. The 12-month assessment was similarly available for patients receiving each model of care (80% for patients receiving CBA; 84% for patients receiving clinic-based care alone); however, more patients in the clinic-based care group were missing the 12-month assessment due to attrition from cART or transfer to a different health facility as opposed to simply missing the 12-month assessment (35 of 50 (70%) for non-CBA group, 24 of 62 (39%) for CBA group). Among people missing a 12-month assessment, the prevalence of moderate or severe baseline food insecurity was similar regardless of the reason for the missed assessment (93% for missing interview and 94% for attrition or transfer out in clinic-based care group; 85% for missing interview and 90% for attrition or transfer out in CBA group). Relative to those with a 12-month interview, baseline moderate or severe food insecurity tended to be more common about participants missing a 12-month interview in the non-CBA (94% in those missing an assessment versus 84% among those with an assessment; test statistic = 3.32; p-value = 0.06); however, the opposite was true in the CBA arm (82 vs. 91%, respectively; test statistic = 3.44; p-value = 0.06).

Receipt of CBA with food ration was strongly inversely predictive of severe food insecurity at 12 months (Risk Ratio (RR): 0.14; 95% CI: 0.10, 0.20; Wald χ2-statistic = 52.1; p-value = <0.0001) and moderate or severe insecurity at 12-months (RR: 0.70; 95% CI: 0.63, 0.79; Wald χ2-statistic = 35.6; p-value = <0.0001). Ninety-seven percent of participants with a 12-month food insecurity measurement (N = 480) were included in the final multivariable model adjusting for baseline CD4 cell count, depression, MOS HIV physical and mental health score, social support, receipt of TB treatment, HIV disease stage, and transportation cost to clinic. CBA with food support was associated with a lower risk of severe food insecurity (adjusted RR: 0.13; 95% CI 0.08, 0.20; Z-statistic = −9.08; p-value = <0.0001) and moderate or severe insecurity at 12-months (adjusted RR: 0.58; 95% CI: 0.49, 0.69; Z-statistic = −6.12; p-value = <0.0001). For severe food insecurity, E-values were 14.9 and 9.5 for the observed risk ratio and the upper bound of the confidence interval, respectively. This means that an unmeasured confounder would need to be associated with both receipt of CBA with food support and severe food insecurity at 12 months by a risk ratio of 9.5 each to explain away the risk ratio from the confidence bound closest to one. For moderate or severe food insecurity, these values were 2.2 and 1.8, respectively.

Food Support and Severe Food Insecurity at 12 Months in the Comparison Group

A small number of patients who received clinic-based care without CBA and had a 12-month food insecurity assessment reported receiving at least some unspecified amount of food support during the follow-up period (n = 31; 12%), whereas the majority (n = 223; 88%) of patients in this group did not report any food support. The 12-month prevalence of severe food insecurity among those in the clinic-based group who reported any support was 61% as compared to 68% among patients who received no food support (χ2-statistic: 3.91; p-value = 0.05).

Correlations Between Change in Food Insecurity, CD4 Count and BMI

In patients who received CBA with food support, we found no statistically significant correlations between changes in BMI, CD4 count, and changes in food insecurity between baseline and 12 months (results not shown). Among those in the clinic-based group, weak positive correlations were observed between changes in BMI and CD4 count (Pearson correlation coefficient: 0.14; p-value = 0.04), and between CD4 count and food insecurity (Pearson correlation coefficient: 0.14; p-value = 0.03).

Discussion

We found significant reductions in food insecurity during the first year of cART in a comprehensive community-based cART program that offered 10 months of a food ration with a standard ration sized for a family of four. After 1 year, the prevalence of self-reported food insecurity among those in the comprehensive community-based cART program was 10%, even less than the background rate of food insecurity reported for this region, which ranged from 25% to 33% [44]. These results contrast with those observed in a high performing clinic-based model that did not include a food ration: among those patients we found no improvement in food insecurity during the same time period. Taken together we conclude that a food ration, in the context of a comprehensive community-based HIV program, may be an effective intervention to reduce food insecurity among adults newly initiating cART.

Few studies have examined changes in food insecurity over time among HIV-positive adults on cART. Longitudinal studies have found improvements in food insecurity following cART initiation both in the presence [17, 45,46,47] and absence [15, 47] of a food ration. In this sense our results deviate: we found no reductions in food insecurity among a group of adults newly initiating cART who did not receive a food ration. There are several possible explanations for this. First, food insecurity is prevalent in rural Rwanda, even among individuals without HIV-infection [28]. It is possible that cART alone was insufficient to counter other forces, including those unrelated to HIV infection, which contribute to food insecurity in this setting. Second, it is possible that external factors led to an increase in food insecurity over the study period, but that this was concealed by a mitigating effect of cART. The lack of an HIV-negative group or an HIV-positive group that had not yet initiated cART for comparison precludes us from exploring other hypotheses. Of note, while baseline prevalence of moderate or severe food insecurity was comparable across study groups, the baseline prevalence of severe food insecurity was lower in the CBA group. This is likely because some participants in the CBA group received their first food ration prior to cART initiation and administration of the baseline questionnaire and therefore the food package may have already mitigated severe food insecurity.

The most dramatic reduction in food insecurity among patients receiving food support through CBA occurred within the first 3 months of cART initiation, a precarious time that is often marked by elevated mortality due to advanced HIV disease [9]. An intervention that reduces food insecurity at a time when patients may be especially vulnerable to malnutrition and non-adherence, two consequences of food insecurity [4,5,6,7,8,9, 11,12,13,14], may also reduce HIV-related morbidity and mortality. Although the food ration was restricted to the first 10 months following cART, food insecurity remained low at 12 months among food support recipients, lower in fact than the prevalence in the general population of this area [28]. Among the many possible explanations for this is that early supplementation permits stabilization during a key period, which may lead to lasting effects. We lack data, such as whether patients were able to work, to further examine this hypothesis.

Patients in the CBA group received a food ration as part of a larger comprehensive community-based intervention, and therefore we are unable to conclude with certainty that improved food security was a result of the food ration versus another pathway related to the CBA intervention. Some patients with grave socioeconomic need who received CBA with food support were invited to participate in microfinance and other income-generating activities that may explain the sustained improvements in food insecurity. Similarly, transportation stipends may have freed up resources for food. It is also possible that greater gains in health and productivity as a result of improved cART adherence through daily community health worker visits may have facilitated one’s ability to work and thus obtain food; however, comparable changes in CD4 count in the CBA and clinic-based groups over the study period and their lack of correlation with changes in food insecurity, are not consistent with this hypothesis. Moreover, an inverse association between food rations and severe food insecurity among patients in the clinic-based group, which did not receive the other elements of CBA, suggests that the food ration may be a key contributor to the reductions in food insecurity. Additional study, including qualitative work, could elucidate how time-limited food rations in the context of a comprehensive community-based intervention can have long-term impacts on food insecurity.

Unmeasured confounding and selection bias due to missing follow-up assessments are potential sources of bias in this study; however, we feel that neither likely explains our study findings. First, we considered and adjusted for a number of sociodemographic, psychosocial and clinical factors. Second, our E-value sensitivity analysis suggested that any unmeasured factor would need to be exceptionally strongly correlated with both CBA and severe food insecurity at 12 months to explain away our study findings. Last, we know of no geographic trends, events, or differences that would have resulted in such drastic reductions in reported household food insecurity in the CBA group but not the clinic-based group. Missing 12-month assessments is similarly unlikely to explain our findings, because completion rates were similar across groups and participants in the CBA group who were missing 12-month assessments had slightly less food insecurity at baseline, while the opposite was true in the clinic-based care group. Assuming a positive correlation between baseline and 12- month assessments, the inclusion of these individuals would likely strengthen rather than explain our findings.

Additional limitations include a lack of data on patient “dose” of the food ration (i.e., whether a given participant received their food ration each month) and on household size, which could have impacted the effectiveness of a food ration designed for a family of four. Notably, the 2010 Demographic and Health Survey in Rwanda suggests that family sizes are comparable in the two study regions with an average of 4.4 people per household [48]. Last, because our study included only adults newly initiating cART, our results cannot be generalized to patients on cART for longer durations.

We previously reported the clinical and psychosocial benefits of CBA with a food ration, as part of a comprehensive, community-based delivery model for HIV care [32, 33, 49]. Based on the present findings, we conclude that a comprehensive community support program, of which a time-limited monthly food ration was a critical component, also contributes to an alleviation of food insecurity among adults newly initiating cART in regions with high HIV and food insecurity rates. Interventions, such as CBA with food support, that lead to improved food insecurity among HIV-positive adults should be prioritized in areas of food insecurity and will likely have positive upstream impacts on HIV- and non-HIV related health outcomes.