Machine learning-based prediction of excessive daytime sleepiness in patients with Parkinson's disease: findings from the PPMI cohort with external validation.
Machine learning-based prediction of excessive daytime sleepiness in patients with Parkinson's disease: findings from the PPMI cohort with external validation.
Where did the research take place?
The study site has not been established. Author addresses may differ from where the research occurred.
Jinan, CN · Author affiliation
Department of Neurology, The Second Clinical Medical College of Jinan University, Shenzhen, Guangdong, China.Location evidence
Shenzhen, CN · Author affiliation
Department of Neurology, The Second Clinical Medical College of Jinan University, Shenzhen, Guangdong, China.Location evidence
A plain-language reading has not been prepared for this paper yet.
Original abstract
BACKGROUND: Excessive daytime sleepiness (EDS) is a common non-motor symptom in Parkinson's disease (PD). Predicting EDS risk enables early intervention. This study aimed to develop and externally validate an explainable machine learning model for EDS prediction in PD. METHODS: A total of 676 patients with Hoehn and Yahr (H&Y) stage 1 PD from the Parkinson's Progression Markers Initiative (PPMI) were retrospectively included as the development cohort, randomly split into training and internal validation sets (7:3). An external validation cohort comprised 180 H&Y stage 1 patients from our clinical center. Least absolute shrinkage and selection operator (LASSO) regression and Boruta feature selection identified predictive variables. Four machine learning models were compared, and the optimal model was interpreted using Shapley Additive exPlanations (SHAP). RESULTS: Five features were selected: Geriatric Depression Scale (GDS), State-Trait Anxiety Index (STAI), Scales for Outcomes in Parkinson's Disease-Autonomic (SCOPA), Unified Parkinson's Disease Rating Scale parts I (UPDRS I) and II (UPDRS II). The logistic regression model achieved the best performance, with an area under the receiver operating characteristic curve (AUC) of 0.732 (95% CI: 0.641-0.823) in the internal validation set and 0.673 (95% CI: 0.552-0.793) in the external validation set. SHAP analysis identified key contributors. CONCLUSION: We developed and externally validated an explainable machine learning model for EDS prediction in H&Y stage 1 PD. This tool may facilitate early risk stratification and personalized management of sleep problems in this subgroup.