[Paper Review] Assessing the predictive ability of the UPDRS for falls classification in early stage Parkinson's disease

Sarini Abdullah, Nicole White+3 more|arXiv (Cornell University)|Oct 3, 2019

Parkinson's Disease Mechanisms and Treatments4 citations

TL;DR

This study evaluates the predictive power of individual Unified Parkinson’s Disease Rating Scale (UPDRS) items versus composite scores for classifying fallers vs. non-fallers in early-stage Parkinson’s disease. Using logistic regression, decision trees, random forests, and Bayesian model averaging on 51 patients over 12 months, it finds that individual UPDRS items—particularly from Parts II and III—achieve 80% accuracy, 85% sensitivity, and 77% specificity, outperforming aggregate measures and offering superior fall prediction in early PD.

ABSTRACT

Identification of risk factors associated with falls in people with Parkinson's Disease (PD) is important due to their high risk of falling. In this study, various ways of utilizing the Unified Parkinson's Disease Rating Scale (UPDRS) were assessed for the identification of risk factors and for the prediction of falls. Three statistical methods for classification were considered:decision trees, random forests, and logistic regression. UPDRS measurements on 51 participants with early stage PD, who completed monthly falls diaries for 12 months of follow-up were analyzed. All classification methods applied produced similar results in regards to classification accuracy and the selected important variables. The highest classification rates were obtained from model with individual items of the UPDRS with 80% accuracy (85% sensitivity and 77% specificity), higher than in any previous study. A comparison of the independent performance of the four parts of the UPDRS revealed the comparably high classification rates for Parts II and III of the UPDRS. Similar patterns with slightly different classification rates were observed for the 6- and 12-month of follow-up times. Consistent predictors for falls selected by all classification methods at two follow-up times are: thought disorder for UPDRS I, dressing and falling for UPDRS II, hand pronate/supinate for UPDRS III, and sleep disturbance and symptomatic orthostasis for UPDRS IV. While for the aggregate measures, subtotal 2 (sum of UPDRS II items) and bradykinesia showed high association with fall/non-fall. Fall/non-fall occurrences were more associated with individual items of the UPDRS than with the aggregate measures. UPDRS parts II and III produced comparably high classification rates for fall/non-fall prediction. Similar results were obtained for modelling data at 6-month and 12-month follow-up times.

Motivation & Objective

• To assess whether individual UPDRS items or aggregate measures better predict falls in early-stage Parkinson’s disease (PD).
• To identify the most informative UPDRS items and sub-scores for fall classification using multiple statistical methods.
• To evaluate the consistency and stability of fall prediction models over 6- and 12-month follow-up periods.
• To compare the performance of logistic regression, decision trees, random forests, and Bayesian model averaging in classifying fall risk.
• To determine whether shorter follow-up (6 months) yields comparable predictive power to longer follow-up (12 months).

Proposed method

• Applied three classification methods: logistic regression, decision trees, and random forests to predict fall status from UPDRS data.
• Used stepwise selection and Bayesian model averaging (BMA) with log-marginal likelihood for logistic regression model selection.
• Employed the Gini index criterion for variable importance and split selection in decision trees and random forests.
• Evaluated model performance using classification accuracy, sensitivity, specificity, and receiver operating characteristic (ROC) analysis.
• Conducted cross-validation to assess model robustness and stability across 6- and 12-month follow-up periods.
• Compared results across individual UPDRS items, sub-scores (e.g., subtotal 2, bradykinesia), and full parts (I–IV) to identify optimal predictors.

Experimental results

Research questions

• RQ1 Which UPDRS components—individual items or aggregate scores—yield higher classification accuracy for fall prediction in early-stage PD?
• RQ2 Which specific UPDRS items are most consistently predictive of falls across multiple statistical models?
• RQ3 How do classification performance metrics (accuracy, sensitivity, specificity) vary between 6-month and 12-month follow-up periods?
• RQ4 Do decision tree, random forest, logistic regression, and Bayesian model averaging methods produce comparable or divergent results in fall prediction?
• RQ5 Can a 6-month follow-up period provide sufficient predictive information comparable to a 12-month follow-up in early PD?

Key findings

• The highest classification accuracy of 80% (85% sensitivity, 77% specificity) was achieved using individual UPDRS items, surpassing results from any prior study.
• UPDRS Parts II and III yielded the highest classification rates, with Part III showing 71–79% accuracy and 73–83% sensitivity.
• Consistent predictors across all models and follow-up times included: thought disorder (UPDRS I), dressing and falling (UPDRS II), hand pronate/supinate (UPDRS III), and sleep disturbance and symptomatic orthostasis (UPDRS IV).
• Aggregate measures such as subtotal 2 (UPDRS II items) and bradykinesia showed strong associations with fall status, but were less predictive than individual items.
• No significant difference in classification performance was observed between 6-month and 12-month follow-up periods, suggesting 6 months may be sufficient for fall risk assessment.
• UPDRS IV, often overlooked, showed high purity and odds ratios >1 for falls, indicating it contributes meaningfully to prediction despite lower stability in cross-validation.