Bayesian adaptive algorithms for locating HIV mobile testing services.

摘要: We have previously conducted computer-based tournaments to compare the yield of alternative approaches deploying mobile HIV testing services in settings where prevalence undetected infection may be characterized by ‘hotspots’. report here on three refinements our prior assessments and their implications for decision-making. Specifically, (1) enlarging number geographic zones; (2) including spatial correlation infection; (3) evaluating a prospective search algorithm that accounts such correlation. Building work, we used simulation model create hypothetical city consisting up 100 contiguous zones. Each zone was randomly assigned infection. employed user-defined weighting scheme correlate levels between adjacent Over 180 days, algorithms selected which conduct fixed tests. Algorithms were permitted observe results own activities use information choosing test subsequent rounds. The Thompson sampling (TS), an adaptive Bayesian strategy; Besag York Mollie (BYM), hierarchical model; Clairvoyance, benchmarking strategy with access perfect information. 250 tournament runs, BYM detected 65.3% (compared 55.1% TS) cases identified Clairvoyance. outperformed TS all sensitivity analyses, except when there small zones (i.e., 16 4 × 4 grid), wherein no significant difference two strategies. Though no, low, medium, high data examined, differences these did not effect relative performance versus TS. narrowly simulation, suggesting improvements can achieved accounting However, comparative simplicity implemented makes field evaluation critical understanding practical value either as existing resources.