Objective: To develop and validate a machine learning model integrating clinical and gait data to identify key clinical markers associated with faller status in Parkinson’s disease (PD).

Background: Falls are common in PD and contribute substantially to morbidity. However, accurate classification of fallers remains challenging due to the heterogeneous presentation of motor and non-motor symptoms.

Method: Clinical and gait data were collected from 468 participants. After excluding cases with incomplete data, 396 individuals were analyzed, with 298 assigned to a training cohort from one center and 98 to an external validation cohort from another. Demographic variables, motor and non-motor assessments, and instrumented walkway system (GAITRite)–derived gait parameters were obtained. Fall history was used to classify participants as PD fallers, PD non-fallers, or healthy controls. Features were selected using statistical and importance-based approaches, and seven machine learning algorithms were trained and evaluated through internal and external validation.

Results: The Extra Trees classifier using statistics-based feature selection achieved the highest performance, with accuracies of 88% in internal validation and 89% in external validation. Feature selection consistently identified three principal domains associated with faller status: fear of falling, balance/gait impairment, and autonomic dysfunction. Fear of falling, captured by multiple Korean version of the Activities-specific Balance Confidence scale (K-ABC) items, emerged as the most discriminative feature. Balance and gait measures, including Tinetti test items (360° rotation) and GAITRite-derived parameters (stride length and velocity), were also highly ranked. Autonomic dysfunction, represented by Scales for Outcomes in Parkinson’s Disease–Autonomic (SCOPA-AUT) scores and orthostatic symptoms, was a notable non-motor contributor.

Conclusion: Machine learning–based models integrating multidomain clinical and gait features can effectively classify faller status in PD with robust external validation. These findings highlight fear of falling, balance/gait impairment, and autonomic dysfunction as key clinical domains associated with faller status.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/classification-of-fallers-in-parkinsons-disease-through-machine-learning-based-feature-analysis/