A1374
Title: Identifying multiple hotspot clusters via within-cluster heterogeneity
Authors: Kunihiko Takahashi - Institute of Science Tokyo (Japan) [presenting]
Hideyasu Shimadzu - Kitasato University (Japan)
Abstract: In epidemiology, spatial scan statistics are a powerful tool for detecting disease clusters with significantly elevated disease risk relative to surrounding regions. Recent methodological developments allow such scan-based approaches, including circular and flexibly shaped scan statistics, to detect multiple hotspot clusters simultaneously. However, conventional approaches generally assume homogeneous risk within a detected cluster and therefore tend to unnecessarily merge adjacent hotspots with distinct risk levels into a single cluster, masking important spatial heterogeneity. To address this limitation, a new scan-based procedure has been developed to identify multiple hotspot clusters by assessing within-cluster heterogeneity. The proposed method integrates Cochran's Q-statistic, a classical measure of heterogeneity, into the scan statistic framework to quantify risk heterogeneity among candidate hotspot regions and distinguish adjacent hotspots with differing disease risks. This approach enables more accurate identification of complex hotspot structures, while retaining the flexibility of existing scan statistics. Simulation studies and real-world disease-mapping applications are used to evaluate the method and compare its performance with conventional scan-based procedures. The results demonstrate improved detection and interpretation of adjacent hotspot clusters exhibiting heterogeneous disease risks.
