Bayesian approach to average power calculations for binary regression models with misclassified outcomes.

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Risk assessment studies often involve binomial regression.
• Response data in these studies can be subject to misclassification.
• Accurate sample size determination is crucial for reliable results.

Purpose of the Study:

• To develop a simulation-based procedure for sample size determination in binomial regression.
• To address challenges posed by misclassified response data.
• To ensure adequate power for detecting associations between predictors and event probability.

Main Methods:

• Utilized a Bayesian average power criterion for sample size calculation.
• Developed a simulation-based procedure applicable to various parametric binomial regression models (logistic, probit, complementary log-log).
• Extended methods for single and two diagnostic test surrogate outcomes.

Main Results:

• The proposed procedure provides a sample size that offers a high probability of correctly establishing the direction of association.
• Demonstrated applicability in medical scenarios with impractical true disease status ascertainment.
• Illustrated with categorical covariates and a human papillomavirus screening example.

Conclusions:

• The Bayesian sample size procedure is effective for studies with error-prone measurements.
• The method enhances the reliability of risk assessment in binomial regression.
• Applicable to diverse statistical models and diagnostic test scenarios.