Bayesian two-part spatial models for semicontinuous data with application to emergency department expenditures.

Area of Science:

• Health Services Research
• Biostatistics
• Spatial Statistics

Background:

• Semicontinuous data, common in health services research (e.g., medical expenditures), present a unique analytical challenge with a zero point mass and a continuous positive distribution.
• Traditional two-part mixture models are often used, but their application in spatial analysis requires careful distributional selection to avoid biased inferences.

Purpose of the Study:

• To introduce a flexible class of Bayesian two-part models for the spatial analysis of semicontinuous health data.
• To provide a joint modeling framework for health services utilization and expenditures across geographic regions.

Main Methods:

• Development of Bayesian two-part models, including lognormal, log skew-elliptical, and Bayesian non-parametric variants.
• Utilizing multivariate conditionally autoregressive priors for spatial smoothing and linking model components.
• Implementation of a fully conjugate Gibbs sampling scheme for efficient posterior computation.

Main Results:

• Demonstration of a joint spatial modeling framework for health utilization and expenditures.
• Efficient posterior computation achieved through a conjugate Gibbs sampling scheme.
• Illustration of the approach using real-world data on emergency department expenditures.

Conclusions:

• The proposed Bayesian two-part models offer a robust approach for the spatial analysis of semicontinuous health data.
• This framework enables a comprehensive understanding of geographic variations in health services utilization and associated costs.
• The methodology addresses potential model misspecification issues, leading to more reliable inferences in health services research.