Misspecification of a binary dependent variable in the logistic model controlling for the repeated longitudinal...

Area of Science:

• Biostatistics
• Medical Informatics
• Epidemiology

Background:

• Accurate disease status determination is crucial for medical applications.
• Serial measurements are often used, but outcomes can be mismeasured.
• Mismeasured outcomes lead to biased statistical estimators.

Purpose of the Study:

• To develop a statistical model accounting for multiple serial measurements and misspecified binary outcomes.
• To derive accurate maximum-likelihood estimators for disease status.
• To assess the impact of outcome misspecification on statistical estimates.

Main Methods:

• Derivation of a complete data likelihood function.
• Application of the Expectation-Maximization (EM) algorithm for parameter estimation.
• Monte Carlo simulations to evaluate estimator performance.
• Retrospective analysis of atrial fibrillation recurrence data.

Main Results:

• The proposed model successfully incorporates both serial measurements and misspecified outcomes.
• Maximum-likelihood estimators were derived using the EM algorithm.
• Monte Carlo simulations demonstrated the impact of misspecification on estimates.
• The model was applied to real-world atrial fibrillation data.

Conclusions:

• The developed statistical approach improves the accuracy of disease status estimation in the presence of measurement errors.
• The model provides a robust method for analyzing longitudinal data with potential outcome inaccuracies.
• This methodology enhances the reliability of medical data analysis, particularly for conditions like atrial fibrillation recurrence.