A Bayesian unified framework for risk estimation and cluster identification in small area health data analysis

Area of Science:

• Biostatistics
• Spatial Epidemiology
• Disease Surveillance

Background:

• Analyzing small area disease data is crucial for understanding spatial disease risk.
• Existing statistical models often focus on either risk estimation or cluster identification, but not simultaneously.
• Geographically separated areas may share common disease risks, posing analytical challenges.

Purpose of the Study:

• To propose a novel Bayesian hierarchical model for small area disease data analysis.
• To enable simultaneous estimation of disease risk and identification of disease clusters.
• To develop a method that accounts for an unknown number of risk classes and can incorporate covariates.

Main Methods:

• A Bayesian hierarchical model is formulated with independent allocation variables for assigning areas to risk classes.
• A novel procedure estimates the posterior distribution of the number of risk classes by combining prior distribution with marginal likelihood.
• An extension incorporating covariates is presented to include additional information or account for spatial correlation.

Main Results:

• The proposed model successfully estimates disease risk and identifies clusters in small areas.
• The method effectively handles an unknown number of risk classes.
• The model's performance is validated through simulation and a case study of varicella in Valencia, Spain.

Conclusions:

• The developed Bayesian hierarchical model offers a robust framework for analyzing small area disease data.
• It provides a unified approach for risk estimation and cluster identification, even for geographically dispersed areas.
• The model's flexibility in incorporating covariates enhances its applicability in epidemiological research.