Power and sample size calculations for generalized regression models with covariate measurement error

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Covariate measurement error is a common issue in scientific data, leading to reduced statistical power in regression analyses.
• Existing power and sample size calculations often overestimate power and underestimate required sample sizes when measurement error is ignored.

Purpose of the Study:

• To derive a novel measurement error-corrected power function for generalized linear models.
• To provide accurate power and sample size calculations that account for covariate measurement error.

Main Methods:

• Developed a generalized score test based on quasi-likelihood methods.
• Derived a flexible power function applicable to discrete or continuous covariates, with or without error.
• Applied the function to logistic regression with a continuous exposure, normal measurement error, and a normal confounder.

Main Results:

• The novel power function corrects for covariate measurement error, leading to more accurate power and sample size estimations.
• Simulations and numerical studies demonstrated the improved properties of the corrected power calculations.
• A program was developed to implement these calculations for specific regression models.

Conclusions:

• Accurate power and sample size calculations are crucial in research involving covariate measurement error.
• The developed methodology offers a more reliable approach for study design in fields like environmental epidemiology, as exemplified by arsenic exposure studies.