A Spatiotemporal Permutation Scanning Method for Epidemic Early Warning That Integrates Adjacency Relations of Geographic Units With Spatial Distance

ABSTRACT The spatiotemporal permutation scanning technique is extensively utilized for the early detection of epidemic and criminal events that exhibit spatial and temporal clustering. The conventional algorithm defines the spatial extent by a progressively enlarging circular area based on a fixed growth radius. However, this approach may lead to inaccurate predictions when geographic observations are irregularly shaped due to terrain or present uneven distributions. To overcome these limitations, this paper introduces an enhanced space–time permutation scanning method for epidemic outbreak alerts that combines the adjacency of geographic units with their spatial proximity. The proposed method extends spatially according to the boundaries of adjacent geographic units, progressing from near to distant areas, and sequentially incorporates each unit based on their spatial distance, employing the maximum likelihood function to discern anomalous spatiotemporal clusters for regional alerts. The method's efficacy was corroborated using data from daily new COVID‐19 cases in Texas, USA, during 2021. The outcomes indicate that our approach surpasses traditional scanning methods, with an average increase of 7%–9% in the hit rate and 5%–6% in the recall rate for early warning zones, thus proving superior efficiency and enhancing the performance of the space–time permutation scanning technique. These results affirm the proposed method's effectiveness in refining the accuracy of public health event outbreak forecasts.