Parkinson’s Disease Fall-Prediction Models Show Moderate External Performance

Researchers developed models to predict falls in Parkinson’s disease using visit-pairs from the Parkinson’s Progression Markers Initiative (PPMI), with external validation in University of Pennsylvania (UPenn) cohorts. The incident-fall model included MDS-UPDRS Part III, Montreal Cognitive Assessment and Geriatric Depression Scale scores; the recurrent-fall model included Hoehn and Yahr stage, Geriatric Depression Scale and Symbol Digit Modalities Test scores. Development and validation sets comprised 773 and 589 visit-pairs for incident falls, and 302 and 137 for recurrent falls, respectively.

For incident falls, AUROC was 0.704 in PPMI and 0.630 in UPenn. In the UPenn validation cohort, observed incident-fall risk ranged from 7.1% in the low-risk group to 18.5% in the high-risk group. For recurrent falls, AUROC was 0.683 in PPMI and 0.641 in UPenn; recurrent-fall rates in UPenn ranged from 22.2% to 65.0% across risk groups.

Adding imaging and fluid biomarkers modestly improved discrimination in PPMI but not in external validation, where estimates were imprecise. The findings suggest that clinical measures can stratify fall risk to a degree, but attenuated external performance and the absence of evidence that model-guided care improves outcomes limit conclusions about clinical use.