Objective: to use machine learning on detailed clinical, cognitive, balance, and gait data to identify which features best predict falls risk in people with Parkinson’s disease, so that individuals at higher risk of becoming frequent fallers can be detected early and offered targeted, personalised interventions.

Background: Falls are a major cause of injury, disability, and loss of independence, leading to a significant decline in quality of life—particularly among individuals with Parkinson’s disease (PD). The onset of frequent falling often represents a key milestone in disease progression, marking an accelerated deterioration in independence and overall well-being. Falls arise from a complex interplay of factors such as balance, muscle strength, and cognitive function. To date, there remains no reliable method to predict who is most at risk of falls, as the influence of these factors is highly individualised. Untangling this complex aetiology is key to developing more effective, personalized interventions to prevent falls and enhance quality of life in PD.

Method: Eighty-two participants (aged 39–86 years), completed a comprehensive battery of clinical and cognitive assessments, including the MDS_UPDRS MDS-UPDRS, Mini-Balance Evaluation Systems Test (MiniBEST), Montreal Cognitive Assessment (MoCA), Beck Anxiety and Depression Inventories, and fear-of-falling questionnaires. The cohort comprised 62 individuals with Parkinson’s disease (PD); 16 classified as frequent fallers, 23 as infrequent fallers, and 23 as non-fallers, as well as 20 unaffected controls. Symptom duration among participants with PD ranged from 19 to 480 months. Machine learning techniques were applied to preprocess and analyze the dataset, extract relevant gait, balance, cognitive and clinical features, and develop predictive models.

Results: We present results on the individual contributions of the above parameters to falls risk in PD as well as a machine predictive model for falls risk.

Conclusion: This approach highlights how machine learning techniques can be used to identify key features most predictive of falls, providing a data-driven framework to identify individuals at higher risk of converting from non- or infrequent fallers to frequent fallers. This approach will enable more targeted, personalized interventions aimed at reducing falls and improving safety, independence, and overall quality of life for people living with PD.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/untangling-the-web-of-fall-risk-in-parkinsons-disease-through-machine-learning/