Machine-learning-based prediction of disability progression in multiple sclerosis: An observational, international,...

Area of Science:

• Neurology
• Artificial Intelligence
• Biostatistics

Background:

• Disability progression is a critical indicator in multiple sclerosis (MS) evolution.
• Current prediction models lack clinical trust and a standardized benchmark.
• There is a need for reliable tools to forecast MS disability progression.

Purpose of the Study:

• To develop and validate machine learning models for predicting disability progression in people with MS (PwMS).
• To establish a benchmark for assessing MS disability progression prediction models.
• To evaluate the utility of routinely collected variables in predicting MS progression.

Main Methods:

• Utilized data from 15,240 PwMS across 146 centers in 40 countries (MSBase consortium).
• Applied state-of-the-art machine learning models following TRIPOD guidelines for prediction over two years.
• Validated models using external datasets five times, assessing discrimination (ROC-AUC, AUC-PR) and calibration (Brier score, ECE).

Main Results:

• Machine learning models demonstrated good performance with ROC-AUC of 0.71 ± 0.01 and AUC-PR of 0.26 ± 0.02.
• Model calibration was adequate, with a Brier score of 0.1 ± 0.01 and ECE of 0.07 ± 0.04.
• Past disability progression was a more significant predictor than treatment history or relapses.

Conclusions:

• Routinely collected variables are sufficient for accurate MS disability progression prediction.
• Machine learning models show promise for clinical application, informing decisions about future MS progression.
• The developed models and code serve as a benchmark for future MS progression prediction research.