Expected estimating equation using calibration data for generalized linear models with a mixture of Berkson and...

Area of Science:

• Epidemiology
• Biostatistics
• Statistical Modeling

Background:

• Epidemiological and clinical studies frequently encounter measurement errors in collected data.
• These errors can be of the classical type, Berkson type, or a mixture of both.
• Ignoring these errors can lead to unreliable statistical inference and biased results.

Purpose of the Study:

• To develop a robust statistical method for regression analysis in generalized linear models when covariates are subject to a mixture of classical and Berkson errors.
• To address situations where calibration data for error assessment are available only for a subsample.
• To provide a method that consistently estimates error variances without prior knowledge of the mixture percentage.

Main Methods:

• Proposed an expected estimating equation approach to simultaneously accommodate both classical and Berkson measurement errors.
• Developed a method to consistently estimate classical and Berkson error variances using available calibration data from a subsample.
• Investigated the finite-sample performance of the proposed method through numerical simulations.

Main Results:

• The proposed expected estimating equation method demonstrated consistent estimation of classical and Berkson error variances.
• The method proved effective even without knowing the specific mixture percentage of the errors.
• Numerical simulations confirmed the method's reliable finite-sample performance.

Conclusions:

• The developed statistical approach effectively handles mixed classical and Berkson errors in covariate data within generalized linear models.
• This method enhances the reliability of statistical inference in epidemiological and clinical research.
• The approach was successfully illustrated using real-world data from an HIV vaccine study.