Measurement error in covariates in the marginal hazards model for multivariate failure time data

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Measurement error in covariates is a common issue in statistical modeling.
• This error can lead to biased results in marginal hazards models for correlated failure times.
• Understanding these bias implications is crucial for accurate data analysis.

Purpose of the Study:

• To investigate the bias introduced by normal additive measurement error in covariates within marginal hazards models.
• To propose and evaluate the SIMEX procedure for correcting this measurement-error-induced bias.
• To assess the performance of the SIMEX method in both large and small sample sizes.

Main Methods:

• Utilizing the SIMEX (Simulation and Extrapolation) procedure to correct for measurement error.
• Analyzing multivariate failure time data with correlated outcomes.
• Applying the method to the Lipid Research Clinics Coronary Primary Prevention Trial data.

Main Results:

• The SIMEX procedure effectively corrects bias in regression coefficients of marginal models.
• Demonstrated large and small sample properties of the SIMEX method.
• Illustrated the application using total cholesterol as a covariate with correlated cardiovascular outcomes.

Conclusions:

• The SIMEX procedure offers a robust solution for handling measurement error in covariates in marginal hazards models.
• This method improves the accuracy of survival analysis when dealing with imperfect covariate measurements.
• Findings are applicable to studies with correlated failure time data and covariates measured with error.