A systematic comparison of additive and interaction approaches to modeling the effects of syndemic problems on HIV outcomes in South Africa.

Abstract

Much of the research on the effects of syndemics on HIV outcomes has utilized an additive approach. However, interaction effects may better account for syndemic synergy than an additive approach, but it remains difficult to specify interaction effects without empirical guidance. We sought to systematically compare additive and interaction effects approaches to modeling the effects of syndemic problems on antiretroviral therapy (ART) using empirically specified interaction terms. Participants were 194 people with HIV (PWH) who received HIV care in Khayelitsha, South Africa. In a series of linear regression models, we examined ten syndemic problems: depression, alcohol use, intimate partner violence (IPV), post-traumatic stress, social anxiety, substance use, food insecurity, poverty, housing instability, and structural barriers to care. Depression, substance use, and food insecurity were selected for interaction terms based on a prior network analysis, which found these problems to be most central. The additive models did not produce statistically significant findings. However, the interaction effects models yielded significant interaction terms in both the full model and a parsimonious model. There was a statistically significant effect of the interaction between depression and food insecurity on ART adherence (b = 0.04, Robust SE = 0.02, 95%CI [0.001-0.08], p = .012). This pattern of results was replicated in the parsimonious model. Findings suggest that when feasible, interaction effects approaches may be a helpful syndemic modeling technique. Results may inform future intervention targets, such as depression and food insecurity, and the importance of addressing both structural and psychosocial syndemic problems.