Machine Learning–Driven Personalized Prediction of Disease-Modifying Therapy in Multiple Sclerosis

Abstract

Motivation: Multiple Sclerosis (MS) is a heterogeneous autoimmune disease characterized by variable clinical trajectories and treatment responses. A growing number of Disease-Modifying Therapies (DMTs) are available, each differing in efficacy, mechanism of action, and risk profile. Selecting the most appropriate therapy for an individual patient remains a complex, data-intensive decision that relies heavily on clinician experience and evolving guidelines. Traditional rule-based or cohort-specific statistical approaches often struggle to capture the nonlinear interactions and patient-specific variability inherent in real-world clinical data.

Results: We propose a data-driven machine learning framework for personalized DMT prediction using structured clinical and demographic data. The problem is formulated as a supervised multiclass classification task, with alternative label groupings based on therapeutic efficacy, route of administration, or mechanism of action to explore different clinical decision granularities. Multiple models, including ensemble and gradient-boosting methods, are trained with hyperparameter optimization and evaluated under class imbalance settings. Comparative performance analysis highlights the trade-offs between predictive accuracy, calibration, and generalization across labeling strategies.

Conclusion: This work presents a systematic machine learning approach to support personalized treatment selection in MS. By leveraging population-level clinical data and flexible modeling strategies, the proposed framework aims to enhance decision support, improve treatment alignment with patient characteristics, and contribute toward more data-driven precision medicine in neuroimmunology.