An approximate quasi-likelihood approach for error-prone failure time outcomes and exposures

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Research

Background:

• Measurement error is prevalent in electronic health records and observational studies, particularly with self-reported outcomes like dietary intake.
• Self-reported data, common in chronic disease research (e.g., diabetes), is prone to error, potentially biasing exposure-disease associations and misleading clinical decisions.

Purpose of the Study:

• To extend a semiparametric method for handling measurement error in discrete failure time outcomes to also address covariate error.
• To evaluate the performance of the proposed method against a naive approach that ignores measurement error.

Main Methods:

• Developed an extended semiparametric likelihood-based method to correct for measurement error in both outcomes and covariates.
• Conducted extensive numerical simulations to compare the bias and efficiency of the proposed method versus the naive approach.
• Applied the method to Women's Health Initiative data to analyze the association between dietary intake and incident diabetes mellitus.

Main Results:

• The proposed method demonstrated minimal bias and maintained coverage probability in simulations, significantly outperforming the naive analysis.
• Naive analysis ignoring measurement error exhibited extreme bias and low coverage.
• Correcting for measurement errors in both self-reported outcomes and dietary exposures yielded substantially different hazard ratio estimates for diabetes risk.

Conclusions:

• The developed method effectively corrects for measurement error in both outcomes and covariates in survival analysis.
• Failure to account for measurement error in self-reported data can lead to misleading conclusions in clinical research.
• Accurate assessment of exposure-disease relationships, particularly for diet and chronic diseases like diabetes, requires addressing measurement error in both outcomes and covariates.