Machine learning in secondary progressive multiple sclerosis: an improved predictive model for short-term disability...

Area of Science:

• Neuroscience
• Biostatistics
• Computational Medicine

Background:

• Predicting disease progression in secondary progressive multiple sclerosis (SPMS) is crucial for optimizing clinical trial design.
• Machine learning (ML) offers advanced computational methods for developing predictive models with reduced human input.

Purpose of the Study:

• To compare the performance of decision tree (DT)-based ML algorithms against logistic regression (LR) and support vector machines (SVMs) in predicting disability progression in SPMS.
• To evaluate both individual and ensemble model performances.

Main Methods:

• Utilized data from 485 SPMS participants in a 2-year placebo-controlled trial.
• Defined progression based on sustained increases in Expanded Disability Status Scale (EDSS).
• Included variables such as EDSS, Multiple Sclerosis Functional Composite scores, imaging data (T2 lesion volume, brain parenchymal fraction), disease duration, age, and sex. Area under the receiver operating characteristic curve (AUC) was the primary evaluation metric.

Main Results:

• DT-based models demonstrated superior predictive performance with AUCs of 61.8%, 60.7%, and 60.2%.
• These DT models outperformed ensemble SVM (52.4%, 51.0%) and LR (49.5%, 51.1%) models.

Conclusions:

• Non-parametric ML, particularly DT-based approaches, showed the highest accuracy in predicting SPMS disability progression.
• Validated ML models could refine the selection of high-risk individuals for SPMS clinical trials, potentially reducing the exposure of low-risk participants to experimental treatments.