A multi-outcome prognostic score for therapeutic responses in multiple sclerosis

Area of Science:

• Neuroimmunology
• Computational Neuroscience
• Clinical Neurology

Background:

• Multiple sclerosis (MS) exhibits significant heterogeneity in disease progression and treatment response.
• Predicting individual patient trajectories and therapeutic efficacy remains a challenge in MS management.

Purpose of the Study:

• To develop and validate a machine learning-based prognostic score for predicting short-term multiple sclerosis (MS) outcomes.
• To assess the score's ability to predict relapse risk, MRI activity, and disability worsening in a French population.
• To evaluate the score's utility in guiding therapeutic choices for commonly prescribed disease-modifying treatments.

Main Methods:

• Utilized data from nine industrial randomized clinical trials (RCTs) and one prospective French MS registry cohort.
• Developed multilabel binary classifiers, primarily using random forest modeling, to predict two-year risks of relapse, new T2 lesions, and sustained disability worsening.
• Prioritized model calibration for probabilistic predictions and conducted external validation on a population-based cohort.

Main Results:

• Random forest modeling effectively captured the temporal dynamics of MS risks.
• External validation showed modest discriminatory capacities (AUCs: 0.67 for relapse, 0.75 for new T2 lesions, 0.58 for disability worsening).
• The prognostic score demonstrated good generalization of MRI activity predictions across various therapeutic scenarios.

Conclusions:

• A novel prognostic score, based on routinely available variables, predicts short-term therapeutic response in multiple sclerosis (MS).
• The score's probabilistic approach aids in selecting appropriate disease-modifying treatments by emphasizing prediction certainty.
• The model's ability to predict MRI activity generalizes well, offering valuable insights for clinical decision-making in MS.