Toward objective monitoring of Parkinson's disease motor symptoms using a wearable device: wearability and performance evaluation of PDMonitor®

Abstract

Parkinson's disease (PD) is characterized by a variety of motor and non-motor symptoms. As disease progresses, fluctuations in the response to levodopa treatment may develop, along with emergence of freezing of gait (FoG) and levodopa induced dyskinesia (LiD). The optimal management of the motor symptoms and their complications, depends, principally, on the consistent detection of their course, leading to improved treatment decisions. During the last few years, wearable devices have started to be used in the clinical practice for monitoring patients' PD-related motor symptoms, during their daily activities. This work describes the results of 2 multi-site clinical studies (PDNST001 and PDNST002) designed to validate the performance and the wearability of a new wearable monitoring device, the PDMonitor^®, in the detection of PD-related motor symptoms. For the studies, 65 patients with Parkinson's disease and 28 healthy individuals (controls) were recruited. Specifically, during the Phase I of the first study, participants used the monitoring device for 2-6 h in a clinic while neurologists assessed the exhibited parkinsonian symptoms every half hour using the Unified Parkinson's Disease Rating Scale (UPDRS) Part III, as well as the Abnormal Involuntary Movement Scale (AIMS) for dyskinesia severity assessment. The goal of Phase I was data gathering. On the other hand, during the Phase II of the first study, as well as during the second study (PDNST002), day-to-day variability was evaluated, with patients in the former and with control subjects in the latter. In both cases, the device was used for a number of days, with the subjects being unsupervised and free to perform any kind of daily activities. The monitoring device produced estimations of the severity of the majority of PD-related motor symptoms and their fluctuations. Statistical analysis demonstrated that the accuracy in the detection of symptoms and the correlation between their severity and the expert evaluations were high. As a result, the studies confirmed the effectiveness of the system as a continuous telemonitoring solution, easy to be used to facilitate decision-making for the treatment of patients with Parkinson's disease.

Keywords: Parkinson's disease; automatic ambulatory monitoring; digital health; inertial measurement unit sensors; telemonitoring; wearable devices.

Figure 1

The PDMonitor^® ecosystem.

Figure 2

(A) PDMonitor^® OFF/Dyskinesia chart for a patient with clear fluctuations. (B) PDMonitor^® OFF/Dyskinesia chart for a patient with significant dyskinesia. In this report the different areas of interest have been marked. Specifically, the area 1 illustrates the severity of a symptom for a 30-min interval, including medication and nutrition information, the area 2 presents a chart with the average symptom intensity for any time of the day, while the area 3 lists the medication the patient receives.

Figure 3

(A) The PDMonitor^® box, docking station, monitoring devices and accessories. (B) The PDMonitor^® monitoring devices' placement on the wrists, torso and ankles. In the middle image, the waist sensor has been placed to the waist with a velcro band and a StrapFrame, but there is also the option to be mounted on a belt using a ClipFrame accessory. (C) The placement of all monitoring devices on the appropriate body position at the same time.

Figure 4

General pipeline used by the PDMonitor^® algorithms. The raw IMU signals are used for motion feature extraction, which are, in turn, utilized for symptom evaluation. The evaluation is generated every 30 min and the symptoms are presented in relevant clinical scales.

Figure 5

Clinical trials (Pilot, PDNST001, and PDNST002) that took place for the generation of datasets used for the development/verification and validation of the algorithms of the PDMonitor^® device. The PERFORM project, used for the initial development of the algorithms, is not related to the studies described in the body of this manuscript.

Figure 6

(A) Boxplot representing the PDMonitor^® bradykinesia score distribution for the different subgroups based on expert UPDRS bradykinesia evaluations. (B) Boxplot of PDMonitor^® gait score distribution for the different subgroups based on expert UPDRS gait evaluation. (C) Boxplot of PDMonitor^® dyskinesia score for the different subgroups based on expert AIMS dyskinesia evaluation. (D) Boxplot of PDMonitor^® tremor (wrist) score distribution for the different subgroups based on expert UPDRS tremor evaluation. The dots represent outliers in the dataset, while the asterisks represent statistical significance. On top of each box there is the number of data points contained within each group. Every data point represents an estimation of the respective symptom for a 30-min window. Regarding the underlying rules for the generation of the box plots, the “whiskers” extend to all points that belong within 1.5 IQR (interquartile range). The rest of the points, lying outside this range, are considered as outliers and are depicted as dots. The asterisks (*) that are drawn on top of the box plots, denote statistical significance and correspond to p-values' ranges. Specifically, 4 asterisks would denote p ≤ 0.0001, 3 asterisks p ≤ 0.001, 2 asterisks p ≤ 0.01, 1 asterisks p ≤ 0.05 while ns denotes p>0.05.

Figure 7

(A) PDMonitor^® output importance for OFF detection (per group defined in Table 1). (B) UPDRS item importance for OFF detection (per group in Table 1).

Figure 8

(A) PDMonitor^® OFF estimation and Bland Altman plot for the patient diaries. (B) Correlation and Bland-Altman plots for day-to-day agreement of PDMonitor^® estimated time percentage, where (left leg) bradykinesia score was more than 1 (UPDRS).