A Bayesian nonparametric model for zero-inflated outcomes: Prediction, clustering, and causal estimation.

Area of Science:

• Biostatistics
• Health Economics
• Data Science

Background:

• Predicting outcomes, subgroup detection, and causal inference are key research goals.
• Skewness and zero-inflation in data distributions pose significant modeling challenges.
• Existing methods struggle with these data pathologies, impacting analysis accuracy.

Purpose of the Study:

• To present a multipurpose Bayesian nonparametric model for continuous, zero-inflated outcomes.
• To simultaneously predict structural zeros, capture skewness, and cluster patients.
• To enable robust causal effect estimation using a novel standardization procedure.

Main Methods:

• Developed a flexible, data-adaptive Bayesian nonparametric model.
• Incorporated the model into a standardization procedure for causal inference.
• Utilized posterior predictive checks for positivity assumption verification.

Main Results:

• The proposed model better captures joint data distributions compared to standard zero-inflated methods.
• Simulation studies show low bias in point estimates and nominal coverage for interval estimates.
• The method yields robust causal effect estimates with uncertainty quantification.

Conclusions:

• The developed Bayesian model effectively addresses skewness and zero-inflation in continuous data.
• It provides a flexible framework for prediction, clustering, and robust causal effect estimation.
• The approach is valuable for analyzing complex health economic data, such as medical costs.