Sleep analysis via wearable sensors in people with Parkinson’s disease

Abstract

Parkinson disease (PD), a well-known illness of motor dysfunction, is characterized by a high prevalence of sleep problems due to degenerative brain changes or comorbid conditions [1]. Wearable devices, in the form of actigraphy, have been shown to also be appropriate for monitoring sleep variables in PD patients [2,3] despite reports that current actigraphy algorithms may misinterpret dysfunctional motor activity, such as tremors, bradykinesia, dyskinesia, and limited arm movement while walking, as well as drug-induced hypermotility, thus making their use problematic in people with PD (PwPD) [4]. The ActiGraph GT3X (Pensacola, FL, USA) accelerometer is capable of recording accelerometry measurements for multiple days at 100 Hz, and has been adopted for massive population-level data collections [5]. In the last few years, Van Hees et al. have developed and made freely available open-source software to estimate sleep variables using data collected from similar off-the-shelf wearable inertial sensors [6]. The goal of this study is to investigate if the ActiGraph data, in combination with Van Hees et al.’s heuristic algorithm Distribution of Change in Z-Angle (HDCZA), can correctly estimate sleep variables in PD patients. To the best of the authors’ knowledge, it is the first study that adopts ActiGraph sensors and this methodology for sleep analysis in PwPD. For further comparison, a custom hardware prototype device named WESAA (Wearable Enabled Symptom Assessment Algorithms) developed at the Tyndall National Institute [8] and with the same capabilities as an ActiGraph device was adopted for additional analysis. Nineteen PD subjects took part in a data collection where participants wore the ActiGraph on their most affected wrist for a minimum of 24 hours and simultaneously filled out a sleep diary. Accelerometer data was collected at 100 Hz. Additionally, six subjects repeated the same data collection protocol while wearing the WESAA system. The heuristic algorithm described in [7] was implemented to detect periods of sleep and compared against the participant diaries. Results are shown in Table I and Figure I in the picture below. Accuracy reported on the subjects using the Actigraph was appropriate with an average 77.8±13.6%, even though results were quite variable across patients (between 31.6% and 91.2%). Less variability is shown with the WESAA device, even though only 6 subjects have carried out this data collection, with an average accuracy of 81.9±6.2% (71.8%-90.2%).Download : Download high-res image (157KB)Download : Download full-size image The present investigation shows that ActiGraph accelerometry data collected over 24 hours, in conjunction with the heuristic algorithm HDCZA for the detection of sleep periods, is an appropriate approach to estimate sleep duration even in PwPD. The same algorithm adopted on the WESAA hardware device shows even more promising results but further investigations with a larger sample size are required to confirm this. Funding: This work was supported in part by Enterprise Ireland (EI) and Abbvie, Inc. under agreement IP 2017 0625; and in part by the Science Foundation Ireland which are Co-Funded through the European Regional Development Fund under Grant 12/RC/2289-P2-INSIGHT..