Some aspects of measurement error in explanatory variables for continuous and binary regression models

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Measurement error in explanatory variables is a common issue in epidemiological studies.
• Classical and Berkson error models, along with additive and multiplicative errors, are frequently encountered.
• Existing methods may lead to biased estimates, particularly in risk assessment.

Purpose of the Study:

• To develop and evaluate a robust measurement error model for epidemiological data.
• To address both continuous and logistic regression models with various error types.
• To provide accurate estimation procedures for cohort and case-control data.

Main Methods:

• Development of amended likelihood functions for statistical analysis.
• Incorporation of classical and Berkson error models with additive or multiplicative errors.
• Simulation studies to assess performance under diverse error structures.
• Application to cohort and case-control study designs.

Main Results:

• The proposed method significantly reduces bias compared to conventional analyses.
• Estimates obtained using amended likelihood functions are within 5% of true values.
• Corrects for attenuation where conventional methods show an average 50% reduction in slope estimates.

Conclusions:

• The developed measurement error model offers a substantial improvement in the accuracy of epidemiological risk estimates.
• The methodology is applicable to various study designs and error types, including complex scenarios.
• This approach provides a reliable tool for analyzing risks, such as lung cancer from radon exposure.