In the study of infectious diseases, cluster analysis is a useful statistical method to explore the relationship between disease hot spots and propagation patterns associated with risk factors. However, when covariates and excessive zeros also produce clustering effects for responses, traditional identification methods for geographic clusters may not address this confounding issue well. Moreover, for complex surveillance systems, scrutinizing all risk factors that could simultaneously affect health outcomes and geographic clusters would be difficult, which could cause large spatial variations and thus create a convergency issue for the Bayesian algorithm. To address these issues, a mixture scan statistic that is associated with the two-part and cluster models is developed to identify geographic clusters for spatial data with covariates and excessive zeros. A model selection procedure is proposed to evaluate the significance of geographic clusters for the mixture scan statistic. Simulation studies and data analysis for associations between dengue infection and environmental risk factors are used to illustrate the proposed method.
Date:
2024-11-04
Relation:
Journal of Agricultural Biological and Environmental Statistics. 2024 Nov 04;Article in Press.
