Regression calibration to correct correlated errors in outcome and exposure

Area of Science:

• Biostatistics
• Epidemiology
• Health Research Methodology

Background:

• Measurement error is a common issue in health research, particularly in outcome assessment and nonclassical covariate measurement error.
• Existing methods primarily address bias from covariate measurement error, with less focus on outcome errors.
• Addressing correlated errors in continuous outcomes is crucial for accurate statistical inference.

Purpose of the Study:

• To extend the regression calibration method for continuous outcomes with measurement errors.
• To develop methods applicable when outcome errors correlate with prognostic covariates or covariate measurement error.
• To provide conditions for identifiability and consistent estimation using validation or reliability subsets.

Main Methods:

• Extension of the regression calibration method to handle measurement error in continuous outcomes.
• Application using a validation subset (true data observed) or a reliability subset (second error-prone measurement).
• Simulation studies to evaluate performance across various measurement error scenarios and subset sizes.

Main Results:

• The proposed method achieves consistent regression parameter estimates under specified conditions.
• Consistency is maintained even with systematic and random errors in outcomes and exposures when the reliability subset has no or classical error.
• Performance is robust across different measurement error types and reliability subset sizes.

Conclusions:

• The extended regression calibration method effectively adjusts for measurement error in continuous outcomes.
• The approach is valuable for studies with correlated errors, enhancing the reliability of health research findings.
• Demonstrated utility in the Women's Health Initiative Dietary Modification Trial data.