Parkinson's disease (PD) has been reported to involve postganglionic sympathetic failure and, in 25% of patients, autonomic failure. In this work we investigate autonomic dynamics in PD using a novel methodology able to provide instantaneous estimates of the Lyapunov spectrum within a point process framework. Physiological signals were recorded from 10 healthy subjects and 9 cognitively preserved PD patients. We computed conventional heart rate variability (HRV) features as well as the full

more »

... unov spectrum over 600s recordings at rest, and tested for significant group effects using a general linear model taking into account age and gender as nuisance covariates. Additionally, the discriminatory power of all features was tested by training a Support Vector Machine (SVM) classifier combined with recursive feature elimination (RFE) with a variable number of target features. The first and second Lyapunov exponents were significantly higher (p<0.05) in PD patients vs. controls. No other HRV measure differed significantly between groups. The best classification performance (75% sensitivity, 80% specificity, area under ROC curve 0.8) was attained when instructing RFE to retain 2 features, where the algorithm selected the first and second Lyapunov exponents. Our results suggest that autonomic control in PD entails a preponderance of nonlinear, more unstable heartbeat dynamics with respect to controls. This could point to possible autonomic dysfunction which cannot be detected by conventional HRV measures.