Adaptive Sequential Surveillance of Infectious Diseases Under Limited Sampling Capacity

This paper develops a closed-loop adaptive sequential surveillance framework for infectious disease monitoring with limited surveillance resources. In this framework, we use a Bayesian spatio-temporal model to predict disease prevalence at each location, and propose several adaptive site-allocation strategies based on the posterior probabilities that the predictions exceed a predefined threshold. These strategies exploit high-risk locations while also exploring, by random or recency-based sampling or by incorporating the frontiers of emerging outbreaks. To address the unequal inclusion probabilities induced by adaptive sampling, we integrate inverse-probability weighting (IPW) into each model update. A simulation study showed that adaptive surveillance strategies improved outbreak detection relative to uniform sampling and fixed sentinel sampling. Frontier-based strategies improved detection and timeliness relative to risk-proportional allocation. When IPW was used to update the model, some proposed strategies yielded lower inferential bias compared with the standard unweighted update, particularly for strategies that allowed more flexible exploration. Credible-interval coverage varied by strategy. The results identify trade-offs between exploitation of known high-risk locations and exploration of potential outbreak frontiers.