Design, Simulation, and Evaluation of a Community-Based Malaria Early-Warning System

Abstract We design and simulate a closed-loop, network-structured community-based malaria early-warning system (CB-EWS) in which local case reports drive a spatially pooled threat estimate, and that estimate triggers a fixed-strength suppression of transmission whenever it crosses a threshold. We evaluate the system via 10-year simulations using a real 16-town road-adjacency network in Ekiti State, Nigeria. Across a wide range of suppression levels and thresholds, the threshold-triggered policy is slightly more effective (p < 0.05) in reducing symptomatic cases than a constant suppression of the same time-averaged effort, while greatly reducing the total time during which suppression measures are in force. A one-factor sensitivity screen finds this gap stays small over plausible ranges of the model's reporting and coupling parameters. The simulation uses a realistic trigger driven by a noisy, delayed, spatially pooled reported signal, which is compared against an idealised perfect-information version. The realistic trigger invariably leads to more suppression episodes and a longer total suppression duration. When suppression is normalized per active person-day, the two are close, with the idealised version ahead at somewhat more grid cells (a median difference of about 0.2 percent). Finally, we discuss a machine-learning approach for potential future improvements.