Space-time Bayesian small area disease risk models: development and evaluation with a focus on cluster detection.

Area of Science:

• Statistics
• Spatial Analysis
• Epidemiology

Background:

• Existing spatial models lack space-time capabilities for cluster detection.
• Evaluating model performance in identifying disease clusters is crucial.

Purpose of the Study:

• To extend spatial local-likelihood and mixture models to the space-time domain.
• To compare these extended models with standard space-time random effect models for cluster detection.
• To evaluate model performance using space-time cluster detection diagnostics and goodness-of-fit criteria.

Main Methods:

• Extension of spatial local-likelihood and mixture models to the space-time (ST) domain.
• Utilized ST counterparts of spatial cluster detection diagnostics, including posterior estimates (misclassification rate) and post-hoc analysis (exceedance probability).
• Compared proposed models with standard random effect space-time (SREST) models using Georgia throat cancer mortality data (1994-2005) and simulated data.

Main Results:

• Standard SREST models demonstrate good performance in space-time cluster detection and overall goodness-of-fit.
• The proposed local likelihood ST model shows superior performance in detecting extreme risks.
• Model evaluation included criteria like misclassification rate, exceedance probability, and mean square error (MSE).

Conclusions:

• The local likelihood space-time model is best suited for extreme risk detection.
• Standard SREST models are effective for general space-time cluster detection and model fit.
• The study provides a framework for evaluating space-time models in epidemiological research.