A Bayesian Hierarchical Model for Evaluating Clinical Outcomes in South African Urban Primary Care Networks

"background": "Urban primary care networks are a cornerstone of health system reform in South Africa, yet robust methods for evaluating their clinical performance across heterogeneous sites are lacking. Existing evaluations often fail to account for multi-level data structures and produce unreliable estimates for smaller clinics. ", "purpose and objectives": "This study aimed to develop and apply a novel Bayesian hierarchical model to measure and compare clinical outcomes across an urban primary care network, using hypertension control as a primary indicator. The objective was to provide a more statistically robust framework for health system evaluation. ", "methodology": "We conducted an intervention study analysing routine clinical data from multiple primary healthcare facilities. The core model was specified as y{ij \ (), \\; () = \ + \ Xij + uj, where yij is the control status for patient i in facility j, Xij are patient-level covariates, and uj \ N (0, \²ᵤ) are facility-level random effects. Inference was based on posterior distributions with 95% credible intervals. ", "findings": "The model revealed substantial variation in adjusted hypertension control rates between facilities, with a posterior probability of 0. 87 that the true inter-facility standard deviation exceeded 0. 2 on the log-odds scale. One network demonstrated a significantly higher pooled control rate (posterior median: 58%, 95% CrI: 52% to 64%) compared to others. ", "conclusion": "The Bayesian hierarchical model successfully quantified clinical performance variation while stabilising estimates for smaller facilities, offering a superior methodological approach for network evaluation. ", "recommendations": "Health authorities should adopt hierarchical modelling techniques for routine health system performance assessment. Future research should integrate this model with cost-effectiveness analyses. ", "key words": "Bayesian hierarchical model, primary care networks, health systems evaluation, clinical