External validation of a clinical prediction model in multiple sclerosis

Area of Science:

• Neurology
• Immunology
• Data Science in Medicine

Background:

• Timely initiation of disease-modifying therapy (DMT) is critical for managing multiple sclerosis (MS).
• Predictive models can aid in optimizing individual treatment strategies.
• Validating these models in diverse populations is essential for broad clinical applicability.

Purpose of the Study:

• To validate a previously published predictive model of individual treatment response in multiple sclerosis (MS).
• To assess the generalizability of the model using a non-overlapping cohort from the Middle East.
• To evaluate the model's accuracy in predicting various clinical outcomes.

Main Methods:

• Utilized the MSBase registry, including patients with relapsing MS or clinically isolated syndrome.
• Applied established models to predict relapses, disability worsening/improvement, conversion to secondary progressive MS, and treatment discontinuation.
• Assessed prediction accuracy using predefined criteria, focusing on a cohort from the Middle East.

Main Results:

• Models demonstrated high accuracy (81%-96%) for predicting disability worsening and improvement.
• Moderate accuracy (73%-91%) was observed for predicting relapses.
• Suboptimal accuracy (<44%) was found for predicting changes in area under disability-time curve (ΔAUC) and treatment discontinuation.
• Accuracy for predicting conversion to secondary progressive MS varied (50%-98%).

Conclusions:

• The validated predictive models are generalizable to patients with diverse baseline characteristics in different geographic regions.
• These findings support the use of predictive modeling for guiding DMT initiation and management in MS.
• Further refinement may be needed for predicting specific outcomes like ΔAUC and treatment discontinuation.