Social determinants of modeled HIV treatment outcomes: A global synthetic-data simulation study

Background and aim: Social determinants of health (SDH) are known to shape HIV treatment outcomes, but their multi‑level, relative contributions remain difficult to isolate from observational data. This literature‑informed, hypothesis‑generating study employs an integrated statistical simulation rather than analysing real patient data. Structural, psychosocial, and behavioural determinants are examined across hypothetical regional‑economic settings. Because the entire cohort is synthetic, the findings generate new hypotheses about SDH interplay without implying a real global population. Methods: A statistical simulation generated 4,793 fully synthetic patient profiles. The 40 countries represent scenario labels for contrasting socioeconomic conditions; no actual patient records or sensitive data were accessed. Fifteen social determinants of health (structural, psychosocial, and behavioural) were assessed. Three outcomes were modelled: ART adherence (continuous 0–100%), virological suppression (binary), and 12‑month care retention (binary). No time‑to‑event or survival analysis was conducted. Results: Within the synthetic cohort, healthcare accessibility showed the strongest positive association with ART adherence (r = 0.551), followed by housing stability (r = 0.482), education (r = 0.470), and social support (r = 0.434). Perceived stigma (r = −0.523) and depression (r = −0.301) exhibited strong negative associations. In scenario conditions with high stigma, modelled mean adherence decreased from 52.71% to 33.06%. Severe food insecurity reduced virological suppression to 32.9% and 12‑month retention to 44.7%. All reported associations reflect the deterministic structure of the simulation and are not inferential. Conclusions: The simulation reproduced plausible social gradients and identifies hypotheses for empirical testing. The observed patterns, including the pronounced impact of stigma and food insecurity, warrant evaluation in real‑world settings. Estimates are not country‑level prevalence estimates, observed clinical effects, or causal effects.