Performing a short sway to distinguish Parkinsonisms

The objective of the present study was to analyze postural stability of patients with different Parkinsonisms by verifying the ability of a short sway could distinguish Progressive Supranuclear Palsy (PSP), atypical parkinsonism, from the typical Parkinson’s disease (PD). Postural stability was investigated by using a stabilometric analysis system during quiet stance with eyes open in a trial of 5/6s. The study population comprised 30 participants (15 PSP patients and 15 patients with recent diagnosis of PD (De Novo PD)). Univariate statistical analysis was used to compare PSP patients and De Novo PD patients. Findings indicated that balance and postural stability were poorer in PSP patients than De Novo PD. PSP patients exhibited increased measures of medio-lateral (M-L) instability, as attested by augmented M-L sway, M-L range and radius. Then, sway variables were given as input to machine learning algorithms: Decision Tree (DT), Support Vector Machine (SVM) and Naïve-Bayes (NB). Overall, machine learning classifiers showed evaluation metrics about the 70%. DT achieved the highest accuracy (73.0%) and the highest AUCROC (75.0%). SVM achieved the best sensitivity (67.0%). Application of predictive models to sway data revealed that machine learning analysis was able to classify patients with different Parkinsonism. The severity of PSP seems to be particularly associated with postural sway.