Predicting Phenoconversion in Isolated RBD: Machine Learning and Explainable AI Approach.

Abstract

Isolated rapid eye movement (REM) sleep behavior disorder (iRBD) is recognized as a precursor to neurodegenerative diseases. This study aimed to develop predictive models for the timing and subtype of phenoconversion in iRBD. We analyzed comprehensive clinical data from 178 individuals with iRBD over a median follow-up of 3.6 years and applied machine learning models to predict when phenoconversion would occur and whether progression would present with motor- or cognition-first symptoms. During follow-up, 30 patients developed a neurodegenerative disorder, and the extreme gradient boosting survival embeddings-Kaplan neighbors (XGBSE-KN) model demonstrated the best performance for timing (concordance index: 0.823; integrated Brier score: 0.123). Age, antidepressant use, and Movement Disorder Society-Unified Parkinson's Disease Rating Scale Part III scores correlated with higher phenoconversion risk, while coffee consumption was protective. For subtype classification, the RandomForestClassifier achieved the highest performance (Matthews correlation coefficient: 0.697), indicating that higher Montreal Cognitive Assessment scores and younger age predicted motor-first progression, whereas longer total sleep time was associated with cognition-first outcomes. These findings highlight the utility of machine learning in guiding prognosis and tailored interventions for iRBD. Future research should include additional biomarkers, extend follow-up, and validate these models in external cohorts to ensure generalizability.