Table 1.
Characteristics of participants.
Table 2.
Hierarchical linear regression models investigating the association between covariates and cognitive domains with risk of conversion.
Table 3.
Hierarchical logistic regression models investigating the association between covariates and cognitive domains with conversion risk status.
Table 4.
Predictive performance of Random Forest models with different input features for estimating risk of conversion from RRMS to SPMS.
Fig 1.
SHAP summary plot illustrating the relative impact of cognitive and clinical variables on Random Forest model predictions of RRMS-to-SPMS conversion risk.
Positive SHAP values indicate increased contribution to higher predicted risk. EDSS and cognitive measures (PASAT, BVMT-R, SDMT, CVLT-II) emerged among the strongest predictors.
Fig 2.
Feature importance (Mean Decrease Impurity, MDI) ranking of cognitive and clinical predictors in the Random Forest model for RRMS-to-SPMS conversion.
EDSS contributed the highest importance, followed by SDMT, PASAT, BVMT-R, and CVLT-II, whereas demographic and treatment-related variables showed comparatively lower contributions.