Machine Learning-Based Prediction of Sleep Quality in Patients with Multiple Sclerosis.

Nearly half of MS patients in the cohort were classified as poor sleepers based on PSQI scores.

• 173 patients with MS were enrolled (mean age 39.66 ± 11.86 years, 69.9% female, 90.2% relapsing-remitting MS).
• 48.0% of patients were classified as poor sleepers (PSQI > 5).
• Mean PSQI global score was 6.06 ± 3.47.
• Study design was cross-sectional, observational, and single-center.

XGBoost outperformed Support Vector Regression and Random Forest models in predicting sleep quality as measured by PSQI global scores.

• Three regression models were trained: Support Vector Regression (SVR), Random Forest, and XGBoost.
• All models were trained on 28 predictors covering clinical, anthropometric, hemodynamic, and laboratory parameters including inflammatory and metabolic indices.
• XGBoost achieved the best predictive performance with a test R2 of 0.451.
• Models were evaluated on an internal test set.

IRLS (restless legs syndrome) severity was identified as the strongest predictor of sleep quality (PSQI) across all three machine learning models.

• SHAP (SHapley Additive exPlanations) analysis was used to examine feature contributions across all three models.
• IRLS severity ranked as the top predictor in SVR, Random Forest, and XGBoost models.
• EDSS score (neurological disability) and depression followed as the next strongest predictors in the Random Forest and XGBoost models.
• Daytime sleepiness (ESS) ranked consistently among the top predictors across models.

Cardiometabolic and inflammatory parameters contributed inconsistently to sleep quality prediction and showed effects opposite to physiological expectation in several instances.

• Cardiometabolic and inflammatory indices were included as predictors alongside disease-related and symptomatic parameters.
• These parameters contributed inconsistently across the three models.
• Several cardiometabolic and inflammatory features showed effects opposite to physiological expectation based on SHAP analysis.
• Disease-related and symptomatic factors outweighed cardiometabolic and inflammatory contributions overall.

Sleep impairment in MS was most strongly associated with restless legs syndrome severity and neurological disability, with depression also contributing in the two best-performing models.

• IRLS, EDSS, and depression were the dominant predictors identified through SHAP analysis.
• Depression appeared as a significant contributor specifically in Random Forest and XGBoost, the two best-performing models.
• The authors note that pain, nocturia, fatigue, and mood symptoms are established contributors to poor sleep in MS.
• The exploratory design and modest sample size (n=173) were noted as limitations warranting confirmation in larger, independent cohorts.

The study assessed sleep using multiple validated instruments alongside a broad panel of clinical and laboratory predictors.

• Sleep quality was assessed with the Pittsburgh Sleep Quality Index (PSQI).
• Daytime sleepiness was measured with the Epworth Sleepiness Scale (ESS).
• Restless legs syndrome severity was measured with the International Restless Legs Scale (IRLS).
• Laboratory parameters included both inflammatory and metabolic indices.
• Anthropometric and hemodynamic parameters were also collected.