Bayesian Approach for Addressing Differential Covariate Measurement Error in Propensity Score Methods.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Propensity score methods are vital for reducing confounding in non-experimental studies.
• Standard methods assume covariates are error-free, but measurement error can bias causal effect estimates.
• Differential measurement error, varying across treatment groups, presents an additional challenge.

Purpose of the Study:

• To propose flexible Bayesian approaches for handling differential measurement error in propensity score methods.
• To estimate average causal effects more accurately in the presence of such errors.
• To evaluate methods under systematic, heteroscedastic, and mixed measurement error scenarios.

Main Methods:

• Developed two Bayesian frameworks to account for differential covariate measurement error.
• Investigated systematic (location shift), heteroscedastic (variance differences), and mixed error types.
• Explored sensitivity analyses using various prior specifications for error parameters.

Main Results:

• Simulation studies demonstrated the performance of the proposed Bayesian methods.
• The methods effectively addressed bias introduced by differential measurement error.
• Application to adolescent drug use disorders showed the practical utility of the approaches.

Conclusions:

• The proposed Bayesian methods offer a robust solution for differential measurement error in propensity score analysis.
• Accurate causal inference requires addressing measurement error, especially when it differs between groups.
• These techniques enhance the reliability of estimates in observational studies with complex data.