When Psychological Signals Meet Neurological Disability: An Explainable Xgboost Study of Dissociative Symptoms in People With Multiple Sclerosis Publisher



Radfar F ; Shahidi S ; Farahani H ; Naser Moghadasi A ; Eskandrieh S
Source: Clinical Neuropsychologist Published:2026

Abstract

Objective: This study examined whether psychoform and somatoform dissociation, assessed with the Dissociative Experiences Scale–II (DES-II) and the Somatoform Dissociation Questionnaire–20 (SDQ-20), provide incremental predictive information for disability severity, measured by the Expanded Disability Status Scale (EDSS), in people with multiple sclerosis (PWMS), within a predictive (non-causal) framework using an explainable extreme gradient boosting (XGBoost) model. Method: We analyzed 257 clinically confirmed PWMS cases (mean age = 37.8 years; 84% female) recruited from the Sina MS Research Center, Tehran, Iran. Participants completed the DES-II, SDQ-20, and demographic/clinical questionnaires; disability was assessed with EDSS. The model employed a 70/30 train–test split, five-fold cross-validation, and grid-search hyperparameter tuning. Results: On the independent test set, the model achieved R2 = 0.122, RMSE = 1.89, and MAE = 1.30. Relative to a mean EDSS of 2.38 (SD = 1.92), this indicates moderate prediction error. Prediction-interval coverage was 86.96%, suggesting reasonably calibrated uncertainty. Feature importance indicated that age, SDQ-20 scores, disease duration, and DES-II scores contributed most to model predictions, whereas demographic variables showed smaller contributions. Importance values reflect relative contribution to prediction error reduction rather than effect sizes or causal associations. Conclusions: Dissociative symptoms provide modest but non-trivial predictive information for disability severity in PWMS beyond conventional indicators. Incorporating dissociation into predictive frameworks may support early risk stratification without implying causal relationships and may inform integrative clinical formulations. Explainable machine learning (e.g. XGBoost) can help characterize biopsychosocial patterns relevant to disability in chronic neurological disease. Findings warrant replication in longitudinal, multi-center cohorts. © 2026 Informa UK Limited, trading as Taylor & Francis Group.