A hidden Markov model for continuous longitudinal data with missing responses and dropout.

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Longitudinal studies often encounter missing data, complicating analysis.
• Existing methods may not adequately address diverse missing data patterns like intermittent missingness and informative dropout.
• Accurate statistical modeling is crucial for reliable interpretation of longitudinal health data.

Purpose of the Study:

• To propose a novel hidden Markov model (HMM) for multivariate continuous longitudinal responses.
• To effectively handle three distinct missing data patterns: partial, intermittent, and monotone (dropout).
• To provide a robust framework for analyzing longitudinal data with covariates under missingness assumptions.

Main Methods:

• Developed a hidden Markov model (HMM) accommodating multivariate continuous longitudinal data.
• Incorporated strategies to manage partially missing outcomes, intermittent missingness, and informative dropout.
• Utilized the missing-at-random (MAR) assumption for partial and intermittent missingness.
• Employed an extra absorbing state to model informative dropout.
• Parameter estimation via maximum likelihood using an expectation-maximization (EM) algorithm with recursions.

Main Results:

• The proposed HMM effectively models longitudinal data with complex missing patterns.
• Simulation studies demonstrated the model's performance.
• The model was successfully applied to real-world data from a primary biliary cholangitis study.

Conclusions:

• The hidden Markov model provides a flexible and robust approach for analyzing longitudinal data with various missing data types.
• This methodology enhances the analytical capabilities for complex biomedical and health-related longitudinal studies.
• The model facilitates more accurate insights from data affected by missingness and dropout.