Does a Sway-Based Mobile Application Predict Future Falls in People With Parkinson Disease?

Abstract

ObjectiveTo determine whether Sway, a sway-based mobile application, predicts falls and to evaluate its discriminatory sensitivity and specificity relative to other clinical measures in identifying fallers in individuals with Parkinson disease (PD). DesignObservational cross-sectional study. SettingCommunity. ParticipantsA convenience sample of subjects with idiopathic PD in Hoehn and Yahr levels I-III (N=59). InterventionsParticipants completed a balance assessment using Sway, the Movement Disorders Systems-Unified PD Rating Scale motor examination, Mini-BESTest, Activities-specific Balance Confidence (ABC) Scale, and reported 6-month fall history. Participants also reported falls for each of the following 6 months. Binomial logistic regression was used to identify significant predictors of future fall status. Cutoff scores, sensitivity, and specificity were based on receiver operating characteristic plots. Main Outcome MeasuresSway score. ResultsThe most predictive logistic regression model included fall history, ABC Scale, and Sway (P<.001). This model explained 61% (Nagelkerke R2) of the variance in fall prediction and correctly classified 85% of fallers. However, only fall history and ABC Scale were statistically significant (P<.02). Participants were 32 times more likely to fall in the future if they fell in the past. The ABC Scale and Mini Balance Evaluation Systems Test (Mini-BESTest) demonstrated greater accuracy than Sway (area under the curve=0.76, 0.72, and 0.65, respectively). Cutoff scores to identify fallers were 85% for the ABC Scale and 21 of 28 for the Mini-BESTest. ConclusionSway did not improve the accuracy of predicting future fallers beyond common clinical measures and fall history.