Modeling disease-state transition heterogeneity through Bayesian variable selection.

Area of Science:

• Biostatistics
• Epidemiology
• Computational Biology

Background:

• Markov transition models are frequently used to describe disease progression between discrete states.
• Subject variability in transition probabilities is a significant challenge, often due to unmodeled clinical factors and disease history.
• Existing models struggle to capture complex, subject-specific relationships influencing disease state changes.

Purpose of the Study:

• To develop a Bayesian variable selection approach for Markov transition models.
• To enable accurate estimation of subject-specific transition probabilities.
• To simultaneously determine the Markov process order and covariate effects on transitions.

Main Methods:

• Utilized Bayesian variable selection within Markov transition models.
• Developed a method to estimate subject-specific transition probabilities.
• Incorporated simultaneous estimation of Markov order and transition-specific covariate effects.
• Validated the approach through simulation studies.

Main Results:

• The proposed methodology accurately identified complex covariate-transition relationships in simulations.
• Applied to multiple sclerosis (MS) patient data, the model revealed significant interactions.
• A clinically significant interaction between relapse history and Expanded Disability Status Scale (EDSS) history was identified in MS patients.

Conclusions:

• Bayesian variable selection offers a robust framework for personalized Markov transition modeling.
• This approach enhances understanding of disease heterogeneity and progression drivers.
• The findings have implications for predicting and managing chronic diseases like MS.