"background": "Community-based health systems are a cornerstone of primary care in many African nations, yet robust, context-specific evaluations of their cost-effectiveness remain scarce. Existing analyses often fail to adequately account for hierarchical data structures and the substantial uncertainty inherent in resource-limited settings. ", "purpose and objectives": "This study aimed to develop and apply a Bayesian hierarchical model to evaluate the cost-effectiveness of community health centres in Ghana, providing a methodological framework that explicitly quantifies uncertainty for decision-makers. ", "methodology": "We constructed a probabilistic cost-effectiveness model using primary data on costs and health outcomes from a sample of community health centres. The core model is a Bayesian hierarchical linear regression: y{ij \ (\ + \ Xij, \²), \ \ (\\, \²), where yij is the incremental net health benefit for individual i in centre j, \ are centre-specific random effects, and Xij are covariates. Parameters were estimated using Hamiltonian Monte Carlo. ", "findings": "The model estimated a 95% credible interval for the incremental cost-effectiveness ratio (ICER) of £amount per disability-adjusted life year averted, indicating a high probability of cost-effectiveness at a willingness-to-pay threshold of one times the national gross domestic product per capita. Centre-level heterogeneity was substantial, with the standard deviation of the random effects (\) estimated at 0. 31 (95% CrI: 0. 22, 0. 45), highlighting significant variation in performance across sites. ", "conclusion": "The Bayesian hierarchical approach provides a statistically rigorous and practically informative framework for cost-effectiveness analysis in decentralised health systems, formally incorporating variability between implementation sites. ", "recommendations": "Health policymakers should adopt probabilistic, hierarchical modelling techniques for economic evaluations to better capture geographical and operational heterogeneity. Resources should be allocated with consideration for
Asante et al. (Tue,) studied this question.