BACKGROUND: Accurate prediction of falls in patients with Parkinson's disease (PWP) is crucial for effective prevention efforts. Historically, fall risk models have heavily relied on motor features, overlooking the vital cognitive-motor interplay essential for locomotion. METHODS: Baseline assessments and year-long fall data from the CYClical Lower Extremity Exercise for Parkinson's disease II (CYCLE-II) trial's control group were utilized. A LASSO logistic regression model assessed thirty-seven demographic, motor, and cognitive variables to identify key fall predictors. To explore the practical implementation of predicting falls in a clinical setting, a second model was developed using a subset of nine candidate measures conducive for retrieval from electronic medical records. Models' accuracy was validated against Paul et al.'s 3-step fall prediction model. RESULTS: Analysis included 123 participants (mean age 65.3 ± 8.3 years, 66 % males, mean disease duration 4.9 ± 4.1 years). Seventy-two participants (58.5 %) fell at least once
with 33.1 % occurring during walking, 34.4 % resulting in injuries. The initial model identified 8 predictors with an AUC of 0.68. The second model, incorporating disease duration and cognitive tests, achieved an AUC of 0.67, comparable to Paul et al.'s validation (AUC 0.66). Participants with poorer information processing and spatial memory were more prone to falling over the 12-month period. CONCLUSIONS: Impaired cognitive performance and longer disease duration were powerful predictors in identifying a future fall in PWP. The link between cognitive performance and potential for falling reinforces the strong interplay between gait and cognition.