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

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Parkinson's disease (PD) is characterized by postural instability and gait difficulties, significantly increasing fall risk.
• Falls affect a large proportion of PD patients (35-90%), posing a major management challenge.
• Accurate fall risk prediction and identification of contributing factors are crucial for timely interventions.

Purpose of the Study:

• To develop and validate a machine learning (ML) algorithm for forecasting fall risk in PD patients.
• To identify factors associated with fall risk using routinely collected clinical data.
• To assess the algorithm's performance across multiple Italian clinical centers.

Main Methods:

• Utilized patient data from two Italian centers (N=251) for training (N=164) and internal validation (N=87).
• Performed external validation on a subset of Parkinson's Progression Markers Initiative (PPMI) study patients (N=65).
• Compared logistic regression (LR) and Support Vector Classifier (SVC) models, employing Shapley Additive exPlanations (SHAP) for variable importance.

Main Results:

• Support Vector Classifier (SVC) showed a slight edge over Logistic Regression (LR) in the training set (AUC: SVC=0.792, LR=0.779).
• Logistic Regression (LR) demonstrated superior prediction accuracy in both internal (AUC: LR=0.753, SVC=0.733) and external validation cohorts (AUC: LR=0.714, SVC=0.676).
• Shapley Additive exPlanations (SHAP) analysis of the LR model identified both motor and non-motor variables associated with fall risk.

Conclusions:

• Machine learning models effectively estimate fall risk across diverse clinical settings, facilitating personalized interventions.
• These models can enhance the quality of life for Parkinson's disease patients.
• Predicting falls in US-based populations presents challenges due to demographic and healthcare system variations.