Machine Learning Predicts Risk of Falls in Parkison's Disease Patients in a Multicenter Observational Study

• DOI: 10.1111/ene.70118
• Publication Date: 2025-04-30T09:41:22Z
• Authors: Maria Chiara Mala...; Chiara Longo; Monica Moroni; Flavio Ragni; Stefano Bovo; Marco Chierici; Lorenzo Gios; Laura Avanzino; Roberta Marchese; Francesca Di Biasio; Matteo Pardini; Denise Cerne; Paola Mandich; Manuela Marenco; Antonio Uccelli; Bruno Giometto; Giuseppe Jurman; Venet Osmani
• Keywords: Clinical prediction rule

ABSTRACT Background Postural instability and gait difficulties are key symptoms of Parkinson's disease (PD), elevating the risk falls substantially. Falls afflict 35% to 90% PD patients, representing a major challenge in managing condition. Accurate prediction fall identification contributing factors essential for timely interventions. Objectives Our objective was develop validate machine learning (ML) algorithm across multiple centers Italy accurately forecast identify related using routinely collected clinical data. Methods Patient data from two Italian ( N = 251) were divided into training cohort 164) ML model development validation 87). External conducted on subset PPMI study patients 65). We compared performance logistic regression (LR) Support Vector Classifier (SVC) models trained The Shapley Additive exPlanations (SHAP) method employed examine predictive power individual variables. Results In set, SVC outperformed LR slightly (AUC: 0.779 ± 0.054, 0.792 0.056). However, demonstrated better accuracy both internal 0.753, 0.733) external cohorts 0.714, 0.676). SHAP analysis revealed associations between motor non‐motor Conclusions ML‐based effectively estimate different centers, enabling tailored interventions enhance patients' quality life. Challenges persist predicting US‐based due demographic healthcare system differences.