Joint assessment of dependent discrete disease state processes

Area of Science:

• Neurology
• Biostatistics
• Computational Biology

Background:

• Multiple sclerosis (MS) clinical assessment relies on the Expanded Disability Severity Scale (EDSS) and relapse status.
• These discrete measures are often modeled as jointly dependent Markov processes.
• Accurate modeling of EDSS and relapse transitions is crucial for MS research and patient care.

Purpose of the Study:

• To propose and assess novel statistical methods for analyzing jointly dependent Markovian processes in multiple sclerosis.
• To formally justify modeling decisions regarding disease state combinations and dependence structures.
• To enhance the accuracy of modeling transitions between EDSS states and changes in relapse status over time.

Main Methods:

• Utilizing Bayes factors for model selection and assessment.
• Applying Bayesian variable selection techniques.
• Analyzing simulated data and real-world data from the Partners Multiple Sclerosis Center.

Main Results:

• Demonstrated the utility of proposed Bayesian methods in assessing jointly dependent Markovian processes.
• Provided a formal framework for making modeling decisions in MS progression analysis.
• Validated methods using both simulated and clinical datasets.

Conclusions:

• Bayes factors and Bayesian variable selection offer a robust approach to modeling complex disease progression in MS.
• These methods address limitations of ad hoc decision-making in statistical modeling.
• The proposed techniques can lead to more accurate predictions and improved understanding of MS.