Accurate classification of secondary progression in multiple sclerosis using a decision tree.

Area of Science:

• Neurology
• Biostatistics
• Clinical Informatics

Background:

• Accurate phenotyping in multiple sclerosis (MS) is hindered by the lack of reliable imaging or biological markers for phenotype transition.
• Current methods for assigning MS disease status are often difficult and subjective.

Purpose of the Study:

• To investigate the utility of clinical information for accurately assigning current multiple sclerosis disease phenotypes.
• To develop and validate a predictive model for MS phenotype classification.

Main Methods:

• A classification algorithm, specifically a decision tree, was developed using demographic and clinical data from 14,387 MS patients in Sweden.
• The algorithm incorporated the most recent Expanded Disability Status Scale (EDSS) score and patient age.
• Performance was validated in an independent cohort from British Columbia and compared against a prior algorithm and neurologist assessments.

Main Results:

• The decision tree classifier, using only EDSS status and age, achieved 89.3% accuracy in predicting neurologist-assigned MS phenotypes.
• Validation in an independent cohort yielded 82.0% accuracy.
• The developed classifier outperformed a modified existing algorithm (77.8% accuracy) and demonstrated comparable accuracy to clinical judgment by neurologists (85% vs. 84.3%).

Conclusions:

• A validated clinical prediction model can standardize MS phenotype definitions across diverse patient cohorts.
• This model offers valuable supplementary information to assist neurologists in clinical decision-making for multiple sclerosis management.