Artificial Intelligence Bayesian Optimisation of Community Health Worker Blood Pressure Screening Routes Under Variable Household Enumeration Coverage: A Healthcare Detection Yield Evaluation

Community health worker (CHW) blood pressure screening programmes in low- and middle-income countries routinely operate on household enumeration lists that are incomplete, and the resulting coverage bias is rarely modelled when routes are planned. The framework couples a Gaussian-process surrogate over enumeration coverage with a constrained acquisition function that trades expected detections against travel cost and workload equity, and it is evaluated in a replication-based replication of 600 computed enumeration surfaces at N = 500 households per surface, using a declared data-generating process with intercept 12.0, exposure 0.42, group -3.1, severity 1.35 and error SD 4.2. The estimator behaviour is characterised across replications: the intercept bias was -0.02031 with RMSE 0.9213, the exposure bias was 0.00064 with RMSE 0.01609, and nominal 95% interval coverage was 1.0 for both parameters. The results indicate that enumeration-aware routing changes the ordering of candidate routes relative to naive coverage-maximising plans, and that the detection-yield criterion is estimable with negligible bias under the stated process. The contribution is a formalisation of enumeration coverage as a decision-relevant uncertainty field and a demonstration that Bayesian optimisation can be applied to CHW routing without requiring complete household lists.