Causal Covariate Selection for the Regression Calibration Method for Exposure Measurement Error Bias Correction

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Measurement error in continuous exposures can bias regression calibration results.
• Accurate estimation of exposure-outcome relationships requires addressing this bias.

Purpose of the Study:

• To determine minimal and efficient covariate adjustment sets for regression calibration.
• To provide guidance on selecting covariates using subject-matter knowledge and directed acyclic graphs.

Main Methods:

• Utilized directed acyclic graphs (DAGs) to illustrate covariate selection principles.
• Investigated imputation-based regression calibration for bias correction.
• Applied the approach to the Health Professionals Follow-up Study data.

Main Results:

• Identified two key covariate sets: common causes of exposure-outcome and common causes of measurement error-outcome.
• Demonstrated that adjusting for specific covariates in the measurement error model (MEM) and outcome model is crucial for unbiased correction.
• Showcased efficiency gains by adjusting for non-risk factors in the MEM.

Conclusions:

• Subject-matter knowledge is vital for selecting appropriate covariates in regression calibration.
• The proposed method ensures unbiased correction of exposure measurement error.
• Caution is advised against using regression calibration for nutritional intake estimation via biomarkers.