Objective: Utilize passive, continuous data from mobile and wearable devices to estimate on-off states in Parkinson’s patients.

Background: The Hauser motor diary is often used to quantify on-off motor fluctuations in Parkinson’s. For every 30 minute interval within a 24 hour day, patients self-report whether they feel off, on, or on with dyskinesia. Clinical trials have leveraged home-based motor diaries as primary endpoints, and several FDA-approved Parkinson’s drugs demonstrated increased on time and/or decreased off time via patient self report. Though motor diaries are accepted endpoints in gauging therapeutic efficacy, each completed diary only characterizes a single day, which may not be representative of daily motor states beyond the examined period. Furthermore, motor diaries are associated with significant patient burden, and patient compliance may decrease as the duration of monitoring increases. We aim to generate passive, continuous estimates on-off states using data from wearable sensors on widely-available, consumer devices.

Method: Twenty subjects who experience and can differentiate their on-off fluctuations are enrolled into this feasibility study. They are instructed to carry a mobile device and wear a smart watch for at least 10 days. Subjects self-report Hauser on-off states every 30 minutes during days 1-3, and approximately 4 times per day during days 4-10. From the phone and watch, we passively capture various metrics surrounding mobility, activity, vitals, sleep, and motor symptoms (tremor and dyskinesia). These passive wearable metrics are used to develop models of on-off classification based on patient self-reported labels.

Results: Our on-off classification models utilize input features that are based on statistical properties of sensor data, windowed around patient-reported events every 30 minutes. We utilize a lagged classification model and a recurrent neural network for posterior smoothing on the lagged model. The model output is a continuous estimation of off, on, or on with dyskinesia, at a resolution of up to one classification per minute.

Conclusion: This study highlights the feasibility of using wearable sensors, in combination with a short period of patient self-reporting, to build continuous estimators of on-off states in Parkinson’s disease. Further studies are needed to 1) validate modeling in expanded cohorts and 2) continuously update algorithms throughout the patient journey.

« Back to 2023 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/continuous-estimation-of-parkinsons-on-off-states-using-wearable-devices/