Bayesian causal inference for observational studies with missingness in covariates and outcomes.

Area of Science:

• Biostatistics
• Epidemiology
• Health Data Science

Background:

• Missing data are prevalent in electronic health records and patient registries, complicating statistical and causal inference.
• Standard causal inference methods often assume complete data or require separate missing data imputation steps.
• Observational health data frequently contain mixed-type variables, posing modeling challenges for complex joint distributions.

Purpose of the Study:

• To develop a novel Bayesian nonparametric causal model to address missing data in observational health studies.
• To simultaneously handle missing value imputation and causal effect estimation within a unified framework.
• To evaluate the model's performance in complex data settings with missing covariates and outcomes.

Main Methods:

• Introduction of a Bayesian nonparametric causal model.
• Simultaneous imputation of missing values and estimation of causal effects using the potential outcomes framework.
• Validation through three simulation studies and two real-world case studies.

Main Results:

• The proposed model effectively handles missing data in both covariates and outcomes.
• The method demonstrates robust performance in complex data scenarios.
• The approach facilitates causal inference in the presence of mixed-type variables and missingness.

Conclusions:

• The Bayesian nonparametric causal model offers a powerful tool for causal inference with missing data in observational health studies.
• This method addresses limitations of traditional approaches by integrating imputation and causal estimation.
• Applicable to chronic disease management studies, improving comparative effectiveness research.