The sample consisted of 456 participants followed from August 2007 to July 2010. Twenty (4.4%) participants were lost to follow up. The median duration of follow-up was 2.1 years (interquartile range, 1.6–2.8 years). Baseline summary statistics are provided in . The sample was predominantly female (71.4%). Median self-reported travel time to the clinic was 40 minutes. A majority of study participants were ART-naive at the time of the baseline food insecurity sub-study visit (56.5%). Median treatment duration among ART-experienced participants was 5.4 months. At baseline, participants reported a median social support score of 3.9 and internalized stigma score of 1.0. Ninety-three (20.4%) participants had ever experienced enacted stigma. Slightly less than one-half (46.1%) of participants had disclosed his or her serostatus to a neighbor.
Summary statistics at baseline (N=456)
The overall prevalence of severe food insecurity was 37.9%, and this was greater among women than among men (41.7% vs. 28.8%; P=0.01). Frank macronutrient malnutrition appeared to be rare, as only a minority (10.3%) were underweight or severely underweight with a BMI <18.5. HFIAS was highly variable both between participants and within (i.e., over time) participants over the study period. The overall person-quarter mean HFIAS score was 7.0, with a standard deviation between study participants of 5.4 and a standard deviation within participants of 4.0. The intra-class correlation coefficient was 0.53, indicating that there was considerable variation between participants as well as within participants over time. Nearly one-half (41.0%) of participants were food insecure, and 9.0% were severely food insecure, for the entire duration of follow-up.
Univariable pooled cross-sectional estimates of the association between HFIAS and demographic, clinical, and social factors are displayed in the first column of . Women had a higher mean level of food insecurity: compared to men, women had a 2.10 point greater HFIAS (95% CI, 1.17–3.02), a 32.5% relative increase in intensity compared to the mean HFIAS across men at baseline. Indicators of low socioeconomic status were all associated with greater food insecurity: low educational attainment, low household asset wealth, and unemployment. Serostatus disclosure, lower social support, and greater internalized stigma were also associated with greater food insecurity.
Associations between food insecurity and social support, HIV stigma, and serostatus disclosure (N=456)
Many of these associations remained statistically significant in the multivariable regression model (second column, ). Baseline household asset wealth was inversely associated with food insecurity (P<0.001), and its effects were also large in magnitude. Participants with the lowest asset index values had a predicted mean HFIAS of 9.99 (95% CI, 8.87–11.11), whereas participants with the highest asset index values had a predicted mean HFIAS of 1.25 (95% CI, −0.85 to 3.34), an eightfold difference in food insecurity intensity. Seasonality was also observed in food insecurity.
Serostatus disclosure, social support, and HIV stigma had statistically significant associations with food insecurity. A one point greater difference in the functional social support scale was associated with a 2.02 point lower HFIAS (95% CI, −2.79 to −1.26). There was nearly a twofold difference in food insecurity scores across the range of social support intensity: participants with the lowest social support scores had a predicted mean HFIAS of 11.36 (95% CI, 9.60–13.12), whereas participants with the highest social support scores had a predicted mean HFIAS of 6.30 (95% CI, 5.86–6.75). Internalized stigma also had a statistically significant association with HFIAS: the predicted mean HFIAS was 6.46 (95% CI, 6.00–6.91) at the lowest stigma score and 9.43 (95% CI, 8.18–10.69) at the highest score. We tested serostatus disclosure, social support, and HIV stigma for interactions by sex, but these were not statistically significant (P-values ranged from 0.46 to 0.99). With lagged covariates (third column, ), the association of serostatus disclosure with food insecurity was no longer statistically significant. However, both social support and internalized stigma retained statistically significant associations with food insecurity.
Results from the fixed effects specification are displayed in . Because observed time-invariant variables (e.g., sex) are collinear with the fixed effects, their associations with the outcome could not be estimated. In this specification, social support and internalized stigma retained statistically significant associations with food insecurity. Each one-point increase (i.e., from one time period to the next) in social support was associated with a 0.66 decrease (95% CI, −1.21 to −0.10) in HFIAS. Similarly, increases in internalized and enacted stigma intensity were associated with increases in HFIAS. New HIV serostatus disclosures were not associated with changes in food insecurity, suggesting that unobserved time-invariant confounders (e.g., resilience or coping style) could explain the statistically significant associations observed in the pooled cross-sectional results. The joint statistical significance of the fixed effects was supported by a F-test (P<0.001) indicating that changes in HIV stigma and social support did not fully subsume the influence of individuals on changes food insecurity.
Fixed-effects estimates of the associations between change in food insecurity and changes in social support, HIV stigma, and serostatus disclosure (N=456)