Logistic regression with exposure biomarkers and flexible measurement error

Area of Science:

• Nutritional Epidemiology
• Biostatistics
• Measurement Error Models

Background:

• Accurate exposure assessment is crucial in nutritional and physical activity epidemiology.
• Biomarker data, often costly, are typically available only for a subset of study participants.
• Self-reported data are prone to measurement error, necessitating robust statistical methods.

Purpose of the Study:

• To extend existing regression calibration and conditional scores estimation procedures for measurement error models.
• To incorporate person-specific random effects in self-report assessment models, accounting for covariates like BMI and ethnicity.
• To evaluate and compare different estimation procedures for logistic regression models relating disease risk to unmeasured exposures.

Main Methods:

• Extension of regression calibration, refined regression calibration, and conditional scores estimation.
• Application to a measurement error model with biomarker data on a subsample and self-report data on the full cohort.
• Use of logistic regression to model disease odds ratios based on true exposures, adjusting for measurement error.

Main Results:

• Simulation studies were conducted to assess the performance of the three estimation procedures.
• The simulations provide insights into the trade-offs between different methods under various cohort configurations.
• Guidance is offered on optimal biomarker subsample sizes for reliable exposure and disease risk estimation.

Conclusions:

• The extended methods provide a framework for handling measurement error in nutritional and physical activity epidemiology.
• The choice of estimation procedure and biomarker subsample size impacts the accuracy of disease odds ratio estimation.
• The findings aid researchers in designing studies and selecting appropriate statistical methods for exposure assessment.