Propensity Score-Based Estimators With Multiple Error-Prone Covariates.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Propensity score methods are crucial for reducing confounding in observational studies.
• Standard methods assume covariates are measured without error, but real-world data often contains measurement error.
• Measurement error in covariates can lead to biased causal effect estimates, particularly when multiple confounders are mismeasured.

Purpose of the Study:

• To investigate the impact of multiple mismeasured covariates on propensity score-based causal effect estimation.
• To examine the effects of correlated measurement errors in covariates.
• To evaluate strategies for mitigating bias from mismeasured covariates.

Main Methods:

• Extensive simulation studies were conducted to assess bias under various measurement error scenarios.
• Real-world data analyses were performed to validate simulation findings.
• Propensity score models were analyzed with correctly measured, mismeasured, and auxiliary variables.

Main Results:

• Causal effect estimates showed less bias when mismeasured covariates with strongly correlated true values were included in the propensity score model.
• Correlated measurement errors among covariates introduced additional bias.
• Including correctly measured auxiliary variables correlated with mismeasured confounders proved beneficial for reducing bias.

Conclusions:

• The correlation structure of both true covariate values and measurement errors significantly impacts causal effect estimates.
• Careful consideration of covariate measurement error and the use of auxiliary information are essential for robust propensity score analysis.
• Findings highlight the need for methods that account for complex measurement error patterns in observational research.