Development of a Machine Learning Model to Predict the Expanded Disability Status Scale in Multiple Sclerosis Patients

Abstract

Objective: The assessment of disability in multiple sclerosis (MS) patients is crucial for treatment decisions and prognosis estimation. The Expanded Disability Status Scale (EDSS) provides a standardized way to quantify disability in MS. However, predicting EDSS scores can be challenging due to the complex and heterogeneous nature of the disease. Machine learning techniques offer a promising approach to predict EDSS scores based on various patient characteristics. Methods: 231 people with MS (pwMS) who had an assessment of physical, psychosocial, and cognitive functions in three timelines (baseline (T0), first year (T1), and second year (T2)) were enrolled. The dataset used for the study consists of 126 features. Feature selection was based on feature saliency and correlation analysis. Three machine learning models -XGBoost, Random Forest, and Linear Regression -were trained on the selected features. Hyperparameter tuning was also carried out on the models. Model performance was evaluated using standard evaluation metrics, including MAE, MSE, and R2. Results: The Machine Learning model based on the XGBoost algorithm performed best in predicting EDSS scores (T2). The MAE value obtained with the XGBoost model is 0.2361, the MSE value is 0.2408, and the R2 value is 0.9705. These results indicate that XGBoost's predictive ability on the current dataset is promising. Conclusion: Our study demonstrates the feasibility of using machine learning techniques to predict EDSS scores in MS patients. The developed models show promising performance and have the potential to enhance clinical decision-making and patient management in MS care.