Objective: This study aimed to evaluate the performance of our previously developed cognitive-based fall prediction model (Anis et al. 2025) in a real-world setting using data captured as part of the “Waiting Room of the Future” modality in Cleveland Clinic (Alberts et al. 2023). We compared its performance to a validated motor-based fall prediction model (Paul et al. 2013).

Background: Existing fall prediction models in Parkinson’s disease primarily rely on motor assessments, such as the established Paul et al. model, which is anchored on previous fall history, comfortable gait speed, and the presence of freezing of gait. However, we recently developed a model based on standard clinical and retrievable measures, with cognitive performance (iPad-based Processing Speed Test and Visual Memory Test) and disease duration identified as the strongest predictors of future falls.

Method: The cognitive-based model and the motor-based model by Paul et al. were tested using 10-fold cross-validation, retaining the original predictors while refitting their coefficients using logistic regression. Model fit was quantified via the average AUC across folds. Missing data were imputed using regression techniques. Since our focus is on predicting falls before they occur, we also performed a sub-analysis comparing both models in patients without prior falls.

Results: A total of 1,557 visits from 1,001 participants were analyzed (age 69.6±8.8, 62.7% male, 91.1% white, disease duration 3.6±4.3 years, MDS-UPDRS III 29.9±13.8). The cognitive-based model achieved an AUC of 0.69, while the motor-based model yielded an AUC of 0.76 (Figure 1A), comparable to their respective AUCs of 0.67 and 0.66 in our previous report. Among patients with no prior falls in the year before evaluation (n=605), the cognitive-based model maintained an AUC of 0.69, whereas the motor-based model dropped to 0.58 (Figure 1B).

Conclusion: Our proposed model, based on cognitive measures, demonstrated consistent predictive performance for future falls across real-world data, including in a subgroup of non-fallers, compared to the motor-based model by Paul et al. These findings support integrating these cognitive assessments into routine clinical care for fall risk evaluation.

Figure 1

References: Alberts JL, Shuaib U, Fernandez H, Walter BL, Schindler D, Miller Koop M, et al. The Parkinson’s disease waiting room of the future: measurements, not magazines. Front Neurol. 2023;14:1212113.
Anis S, Zimmerman E, Jansen AE, Kaya RD, Fernandez HH, Lopez-Lennon C, et al. Cognitive measures predict falls in Parkinson’s disease: Insights from the CYCLE-II cohort. Parkinsonism Relat Disord. 2025 Feb 11;133:107328.
Paul SS, Canning CG, Sherrington C, Lord SR, Close JCT, Fung VSC. Three simple clinical tests to accurately predict falls in people with Parkinson’s disease. Mov Disord Off J Mov Disord Soc. 2013 May;28(5):655–62.

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/single-center-real-world-validation-of-cognitive-vs-motor-based-fall-prediction-models-in-parkinsons-disease/