Uncovering Spatial Syndemics: A Bayesian Joint Model of HIV Prevalence and NCD Mortality in Southern Africa

Background HIV and non-communicable diseases (NCDs) increasingly co-occur in Southern Africa, yet region-wide evidence on shared spatial–temporal structure and the value of joint models remains limited. Methods We analysed 13 Southern African countries (2000–2019) using a bivariate Bayesian areal model estimated via Integrated Nested Laplace Approximation (INLA). NCD deaths were modelled with a Poisson likelihood and exposure; adult HIV prevalence was modelled with a Beta–logit likelihood. Spatial dependence used the BYM2 specification; temporal dependence used a first-order random walk (RW1). Penalised-complexity priors regularised hyperparameters. We compared four covariate tiers (Gross Domestic Product (GDP) only to fully adjusted) via DIC/WAIC and summarised predictive adequacy. We mapped BYM2 spatial effects, RW1 temporal trends, and country-level residual correlation between HIV and NCDs after removing shared structure and covariates. Results Non-zero spatial variability and smooth regional time trends were detected for both outcomes and remained stable across model tiers, indicating persistent shared structure after adjustment. Residual correlations attenuated with additional covariates but remained positive in several countries, suggesting potential syndemic overlap beyond shared components. A parsimonious joint model (GDP-only tier) minimized DIC/WAIC, while predictive diagnostics showed good calibration for both outcomes. Fixed-effect estimates changed with adjustment but did not eliminate the latent spatial–temporal signals. Conclusions HIV prevalence and NCD mortality in SADC exhibit shared spatial and temporal structure and retain pockets of residual co-clustering after adjustment. Joint modelling via BYM2–RW1 adds interpretive value over separate analyses by estimating common latent processes and coherent residual co-movement. Findings support geographically targeted integration of HIV–NCD services and investment in subnational data to refine shared components and identify programme mechanisms. Trial registration: Not applicable.