Generalized Linear Models with Covariate Measurement Error and Zero-Inflated Surrogates.

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Exposure measurement error is a significant challenge in epidemiological studies, biasing exposure-disease association estimates.
• Surrogate variables, often used to approximate true exposures, frequently exhibit zero-inflation (e.g., nutrient intake, physical activity levels).
• Naive regression calibration methods are biased when applied to zero-inflated surrogate data.

Purpose of the Study:

• To develop and evaluate statistical methods for regression analysis with zero-inflated surrogate exposure variables.
• To propose an improved regression calibration estimator that accounts for the probability mass at zero.
• To introduce a consistent estimator based on expected estimating equations for zero-inflated surrogate regression models.

Main Methods:

• Investigated regression analysis techniques for handling zero-inflated surrogate exposure data.
• Developed a novel regression calibration estimator to mitigate bias.
• Proposed an expected estimating equation estimator for improved consistency.
• Conducted extensive simulations to assess estimator performance.

Main Results:

• The proposed regression calibration estimator demonstrated reduced bias compared to naive methods.
• The expected estimating equation estimator was found to be consistent under zero-inflated surrogate regression models.
• Simulation studies confirmed the effectiveness of the proposed methods in bias correction.

Conclusions:

• The developed statistical methods effectively address bias in exposure-disease association estimation when using zero-inflated surrogate variables.
• The proposed estimators offer improved accuracy for epidemiological studies with such data characteristics.
• These methods are applicable to real-world studies, including physical activity intervention research.