The aim of this study was to develop regression models for the quantification of parkinsonian bradykinesia. Forty patients with Parkinson's disease participated in this study. Angular velocity was measured using gyro sensor during finger tapping, forearm-rotation, and toe tapping tasks and the severity of bradykinesia was rated by two independent neurologists. Various characteristic variables were derived from the sensor signal. Stepwise multiple linear regression analysis was performed to develop models predicting the bradykinesia score with the characteristic variables as input. To evaluate the ability of the regression models to discriminate different bradykinesia scores, ANOVA and post hoc test were performed. Major determinants of the bradykinesia score differed among clinical tasks and between raters. The regression models were better than any single characteristic variable in terms of the ability to differentiate bradykinesia scores. Specifically, the regression models could differentiate all pairs of the bradykinesia scores (p<0.05) except for one pair in the finger tapping task and one pair in the toe tapping task. In contrast, any single characteristic variable was found not sensitive enough to discriminate many of the pairs, especially in case of the toe tapping task. The results suggest that the multiple regression models reflecting these differences would be beneficial for the quantification of bradykinesia because the cardinal features included in the determination of bradykinesia score differ among tasks as well as among the raters.
Bibliographical noteFunding Information:
This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2014R1A1A2057508).
© 2015 - IOS Press and the authors.
- Parkinson's disease
- clinical rating
- stepwise multiple regression
ASJC Scopus subject areas
- Biomedical Engineering