Optimising fall risk classification models in Parkinson’s disease utilising clinical and mobility outcomes

Background: Falls are a serious concern for people with Parkinson’s disease (PD), often leading to hospitalisation, dependence and reduced quality of life. Effective fall management requires identification of those at risk. Although many clinical and mobility-related outcomes have been linked with falls, it remains unclear which selection of outcomes best discriminate fallers from non-fallers.

Methods: Participants with PD were recruited as part of the ICICLE-GAIT study. Data presented are from the 54-month and 72-month follow-up. Participants were stratified into fallers and non-fallers based on monthly prospective fall reports.

A total of 299 outcomes across 4 domains were collected: clinical (n=9), lab-based mobility (gait n=60, turning n=99) and real-world mobility (n=131).

Receiver operating characteristic analysis evaluated classification models distinguishing fallers from non-fallers. Area under the curve (AUC) determined which models were optimal. Models were re-evaluated at 72-months.

Results: Of the 48 participants, 32 (67%) were classified as fallers and 16 (33%) as non-fallers. Significant group differences (faller vs. non-faller) were found in all domains at 54-months; clinical (n=2/9), lab-based gait (n=4/60), lab-based turning (n=19/99) and real-world mobility (n=5/131). At 54-months, the turning model performed the best (AUC=0.89), followed by the real-world mobility (AUC=0.82). Any model combination of 2 or more domains that included turning yielded an AUC=1.00. At 72-months, three non-fallers had fallen, and the turning model was superior achieving an AUC=1.00.

Conclusion: Turning measures showed superior discriminative power in identifying PD fallers and are robust over time, outperforming clinical, lab-based and real-world mobility assessments. Real-world mobility also had strong discriminative value, highlighting the importance of ecologically valid continuous monitoring. Future models should explore whether real-world turning would have superior discriminative value to lab-based turning.