Introduction

Parkinson’s Disease (PD), a common neurodegenerative disease, affects about 10 million people worldwide1. Deep Brain Stimulation (DBS), which involves the implantation of electrodes in specific areas of the brain to modulate neural activity and alleviate symptoms, is an established neuromodulation treatment for PD and other movement disorders, such as Essential Tremor (ET). DBS offers patients with neurodegenerative diseases flexible and customizable therapy intended to improve quality of life. However, DBS therapy requires multiple programming appointments and complex assessments, which can be burdensome for patients and their families.

Among the gaps facing the field of neuromodulation therapy are: (1) burden of therapy maintenance, (2) therapy access, and (3) discrete assessments in controlled environments2,3. To address these challenges, digital health tools have emerged as a promising solution for optimizing DBS therapy. Digital health is a rapidly evolving field that is transforming healthcare delivery. Access to ongoing care from DBS specialists can be cumbersome as it may require patients to take time off from work, commute long distances, arrange for travel, or work around a caregiver’s schedule. Conventional DBS therapy maintenance restricts patient independence as it is not optimized to patient goals or patient-clinician interactions. The challenges of patient independence and care access burden are further exacerbated due to limited specialists, racial and socioeconomic disparities, and an overall lack of consensus on DBS referral and timing4,5. While there was accelerated adoption of remote care during the COVID-19 pandemic to reduce risk of in-person visits for patients suffering from neurodegenerative conditions, it also highlighted the benefits of telemedicine that continue to be relevant. Specifically, digital solutions can mitigate existing barriers to care access and offer a cost-effective alternative to traditional care. Additionally, discrete assessments, often conducted in controlled clinical settings, offer an inaccurate snapshot of the patient’s condition.

Integration of digital health solutions with neuromodulation care has the potential to significantly improve patient outcomes and experience. Tools, such as wearable technology and mobile health platforms, may offer personalized and efficient solutions for accessing patient care, improving clinical assessments, and enhancing neuromodulation therapy while limiting patient travel, time, and cost6,7,8. Additionally, by collecting and analyzing large amounts of data from patients undergoing DBS, scientists and healthcare providers can gain insights into how DBS works and develop more effective therapies9. Routine integration of digital platforms and large-scale data infrastructure can provide healthcare providers and translational scientists with low-burden remote monitoring capabilities, allowing for real-time virtual programming and optimization of patients’ therapy.

The foundation for an optimized digital health platform, that combines clinical assessment and neuromodulation therapy delivery, was set with the recent introduction of the NeuroSphere(TM) Virtual Clinic platform. Virtual Clinic enables remote neurostimulation programming and telehealth services directly from the patient’s or clinician’s therapy controller app. In addition to modifying programs remotely, its key features include secure in-app audio/video conferencing, an encrypted backend (cloud-based server enabling communication between clinician and patient devices), the ability to remotely interrogate the system, and the protected recovery program (PRP) capability, a safety feature that allows the user to define a default program in the event of a network interruption10.

A digital health platform that can extend these initial capabilities with integrated remote monitoring is critical for optimizing DBS therapy. Remote monitoring enables healthcare providers to remotely monitor patients’ health status including vital signs, using digital tools such as wearable devices, sensors, and mobile apps. In the context of neuromodulation therapy, it will need to facilitate access to real-time clinical outcomes, including symptom severity and side-effects, for health care providers and empower patients by providing them with a better understanding of their progress. In addition, these technologies should help identify digital biomarkers critical for understanding the onset, symptoms, medication-related fluctuations and side-effects, and correlation to DBS treatment2,11.

Digital biomarkers for cardinal motor symptoms of PD (ex: tremor and bradykinesia) have been quantified in recent studies using non-invasive sensors that include inertial measuring unit (IMU) sensors, such as, accelerometers, gyroscopes, and magnetometers12,13. The results from these studies demonstrate that signals from wearable devices integrated with IMU sensors correlate with clinical assessment of PD motor symptoms and provide a promising avenue for automating assessments and reducing dependency on in-person clinical evaluation14,15,16. Additionally, increased acceptability of body-worn sensors by patients makes them a practical option for remote monitoring of ongoing therapeutic optimization17,18.

Through our investigational remote monitoring application (RM App) integrating patient-reported outcomes (PROs) and objective data using wearables, we aimed to develop an accessible, data-driven digital tool to remotely monitor patient symptoms and deliver low-burden and easy-to-access individualized DBS therapy.

Results

Sixty-seven subjects (n = 67), who received a home-monitoring kit in the feasibility study, were evaluated for compliance analysis. Table 1 summarizes the baseline characteristics of the cohort included in our feasibility analysis. For all subjects, the mean duration for passive data collection was 6.8 days during the baseline period and 55 days for the post-implant period.  Compliance in this paper is defined as adherence to the RM app and Apple watch(R) usage. During the remote monitoring phase, subjects complied with IMU data collection for an average of 22.6 days; this required active start on the watch for high-fidelity signals. Compliance rate was defined as the number of times a survey was completed divided by the number of times it was administered. For example, for a subject enrolled in the remote monitoring phase for 100 days, completion of 70 daily check-ins would imply 70% compliance rate. In the context of PDQ-39, the threshold for compliance was referenced to 3 survey administrations. We observed an average compliance rate of 61.2% for daily check-ins (PGIS) and 83.6% for monthly surveys (PDQ-39) (Fig. 1).

Table 1 Baseline characteristics of study participants.
Full size table
Fig. 1
figure 1

Compliance results for daily and monthly surveys along with passive data collection (Healthkit: Apple Watch). Median compliance rates (orange) for the subjects were 100% for PDQ-39 (mean = 85.5%), 67.9% for PGIS (mean = 63.3%) and 71.8% for passive data (mean = 61.9%).

Full size image

A group analysis of all subjects demonstrated that on average, subjects wore their watch 61.9% of the time with an average of 55 days of post-implant data. Forty-three out of 67 subjects had over 50% compliance with completion of daily check-ins and passive data collection. HealthKit data collected from the watch included activity metrics (step count, stride length, active minutes, etc.) and physiological measures (heart rate, heart rate variability, SpO2, etc.), while core-motion and IMU data included high-resolution accelerometer and gyroscope signals. On average, patients improved on the PGIS over the course of 3 months. At baseline, mean PGIS score was 4.4 ± 0.65, at 1-month it was 3.6 ± 0.09 and at 3-months it was 3.2 ± 0.08. Statistical significance (p < 0.05) for PGIS post-implant was compared to the baseline period and determined according to the paired Wilcoxon test with a one-sided alternative hypothesis on median values (Fig. 2).

Fig. 2
figure 2

Daily check-in scores (PGIS) tracked from baseline to 3 months post implant. On average, patients improved on the PGIS over the course of 3 months. At baseline, mean PGIS score was 4.4 ± 0.65, at 1-month it was 3.6 ± 0.09 and at 3-months it was 3.2 ± 0.08. Statistical significance (p < 0.05*) of PGIS post-implant compared to the baseline period was determined according to the paired Wilcoxon test with a one-sided alternative hypothesis on median values; multiple comparisons correction was applied.

Full size image

Discussion

Here, we summarized the development of a remote monitoring platform that enables us to collect PROs and objective data via consumer-grade wearables in a customizable fashion. We tested this platform for usability and feasibility in a large-scale clinical study focused on remote DBS programming for PD (NCT05269862). The remote monitoring platform, currently available on iOS, was developed to integrate passive and objective data to support patient-feedback and clinical decision-making. Results from the study demonstrate that the platform has potential for real-world adoption.

Continuous, real-world data are essential for optimizing neuromodulation therapy; technological progress in neuromodulation has thus far depended on hardware-based improvements such as implantable device miniaturization and improved geometry of electrodes. However, currently, there is a need for digital solutions that can extend care beyond the clinic to address increasing care-access burden and to evaluate patient concerns in real-world scenarios. Advancements in wearable sensors, algorithm development and the Internet of Things (IoT) platform have enabled development of remote monitoring systems for movement disorders2,19,20,21. Wearable sensors, such as smart watches embedded with accelerometers and gyroscopes, can capture disease traits, including gait patterns22,23, tremor episodes, and activity levels; these sensors connect to the internet through data aggregators and transform signals into clinically relevant knowledge.

One example of a remote monitoring platform for movement disorders is the “MyParkinsoncoach” system, which demonstrated efficiency in the comprehensive management of PD. This platform used a combination of wearables and smartphone technology to collect patient data and provide personalized feedback24. Another platform, the iHandU system25, measures bradykinesia, rigidity, and tremor to track PD. Other similar systems have also been developed for epilepsy management26.

We built the RM app to combine passive monitoring through consumer-grade wearables and disease-specific patient reported outcomes (PROs) to leverage ubiquitous platforms such as Apple Watch, Fitbit, and Oura ring. Objective data collected through this system includes validated algorithms27, and high-resolution data to bridge the gap where classifications are not yet defined. Collecting data passively and creating a low-burden experience improves compliance, which is essential for collecting patient-centric data to build a more holistic profile for the patient and improve clinical decisions. To note, in the RM app iteration presented as part of the feasibility study, patients received pre-configured iPhones and Apple Watches to better streamline data collection. Though the RM app is compatible with Fitbit and Oura, those wearables were not used in the study. The vision for the platform is to be wearable-agnostic and customizable in the future. In subsequent iterations, the RM app may be downloaded from the marketplace (ex: Apple app store) and patients could configure the devices with the help of their clinician tailored to specific needs. For example, if the platform is leveraged for a study-specific use, the clinician may provide study-specific identifiers to enable the correct version. Alternatively, if the app-use is for clinical decision-making, prescriptive identifier may be applied that may activate features with a finite timeline and provide options for reminders, etc. The flexible infrastructure would allow for customizations as required. Objective data can also help identify kinematic biomarkers. Though the DBS app is currently separate from the RM app, intermittent learnings could help optimize clinical outcomes by providing more guidance for clinicians with data-driven personalized insights. Eventually, if the biomarkers are robust and detectable at the required time-resolutions, automated implementations could be explored. The frequency of PROs can provide important insights into the effectiveness of therapy for movement disorders. Hence, the RM app was built to integrate PROs at variable frequencies; PGIS was recorded daily to account for normal fluctuations, while qoL metrics and other surveys were delivered at a lower cadence (monthly or when clinically needed). Also, survey completion was coupled with data upload so a daily cadence for PGIS ensured fewer data gaps. PROs can be valuable in assessing treatment effectiveness and patient-centered outcomes in a variety of movement disorders, including PD, ET, and dystonia. A systematic review focused on PD highlighted the importance of selecting appropriate PROs for different measures and the need for consistent and frequent monitoring of patients to ensure optimal outcomes28. A study29 found that frequent PRO assessments may be valuable in identifying changes in non-motor symptoms, such as cognitive function and mood. The study found that patients’ PRO scores on these non-motor symptoms were highly variable and could change rapidly over time, highlighting the importance of frequent monitoring to ensure that these symptoms are effectively managed. In the feasibility study presented here, the PROs were determined based on the primary goals of the clinical study and pre-configured in the app. Future iterations may allow clinicians to choose which PROs they might be interested in based on research versus clinical needs.

In addition to PROs and wearables, remote monitoring systems for movement disorders should also include video and voice analytics, and other sensor-based technologies. These systems allow for real-time monitoring and can provide valuable insights into patient symptoms and behavior. Future iterations of the RM app will need to leverage artificial intelligence (AI) driven features to further optimize experience and clinical insights. Specifically, by analyzing real-time physiological data from body-worn sensors and implantable devices, AI algorithms can suggest optimized DBS settings to improve motor function and reduce side-effects30,31,32. These insights can be integrated into acute and longitudinal clinical decision-making, paving the way for adaptive and closed-loop systems. However, as data becomes increasingly informative, careful consideration of privacy and data security is necessary to ensure that patient data is protected and used responsibly33.

With these AI driven features, the proposed remote monitoring system, RM, can be extended to other chronic conditions treated with neuromodulation such as depression and pain. While depression is traditionally diagnosed through face-to-face or virtual encounters with a clinician, research into virtual technologies has revealed that computer applications can detect depression based on facial expressions and eye movements34,35. Additionally, a study investigating a text-messaging based social support intervention for chronic pain patients attending pain clinics in New York City demonstrated reduced scores on pain perception, reduced pain interference, and improved rating of affect over 4 weeks36. Research into video analytics and text-based digital health applications demonstrate the potential for these technologies to improve therapy for chronic illnesses in addition to movement disorders.

The remote monitoring of patients with chronic illness is efficient37 and cost-effective for healthcare utilization7,38,39 especially in patients with implanted devices40,41,42. Recent modifications in reimbursement policies are aligned with the use of remote monitoring; CPT codes around remote patient (RPM) and remote therapeutic (RTM) monitoring were first introduced in 2018. The latest guidelines around these codes are based on at least 50% compliance by patients (ex: 16 days of data transmitted within a 30-day period). Results from our clinical feasibility study support adherence to these guidelines, as we observed compliance above these thresholds from patients representing various geographies (US and Europe). A major limitation for patients to adhere to the required compliance is reliable internet access. Additionally, we observed that compliance decreased over time implying a general lack of motivation. For the feasibility study, patients did not receive specific insights about their data. However, future iterations could be aimed at further improving compliance by personalizing the app experience based on insights that are useful for patients to monitor their well-being; these features could include elements of gamification and clear instructions with finite requirements that provide short-terms goals to increase motivation.

Lastly, this platform is a technology for optimizing study execution. By leveraging decentralized monitoring, large-scale clinical studies can be designed to directly engage patients. Increased frequency of data collection, which is challenging with conventional research methods, allows for comprehensive evaluation of patients. The option of a low-cost, efficient method of research leads to high quality data leading to results with adequately powered studies. As we continue to develop this platform to leverage these features, resources will be expanded to include patient educational material and digestible feedback. We also plan to focus on customization of the user-interface specific to the patient accounting for their goals and specific symptoms.

In conclusion, the RM app is a remote monitoring platform that can help address the growing need for passive and objective data to support patient feedback and clinical decision-making for movement disorders. Its compatibility with several wearable systems, patient-centric data collection, and low-burden experience improves compliance, enabling improved patient profiles and clinical decisions. The use of IoT and wearable technologies in healthcare can further transform data into clinically relevant knowledge, benefiting the efforts of caregivers in delivering more efficient and comprehensive healthcare.

Methods

The RM app is an iOS-based remote monitoring application that integrates wearable data capture and PROs via surveys in a patient-centric manner. Patient preferences are integrated into the interface enabling a flexible platform that is designed for use in a decentralized setting, such that patients may have greater access to care, as services are provided in more convenient and accessible ways. The infrastructure is developed with a plug and play (PnP) vision. PnP refers to the ability of this platform to automatically detect and configure hardware when connected. The platform, as defined here, includes the RM app and the wearables; it is not currently interfacing with the app that enables adjustment of DBS parameter. In its current iteration, it is developed as a data collection tool. Specifically, it is designed to be compatible with several consumer-grade wearables and allows for passive data collection for a low-burden experience. Figure 3 illustrates the overarching vision of the Remote Therapeutic Monitoring Platform, where the backend can accept multimodal data as objective input from wearables (skin adhesives, smart ring, or smart watch) and patient-reported outcomes surveys, diaries, etc.). This data can be analyzed for specific insights using validated algorithms that are available through the backend. In the depicted example, we are showing how insights about severity and presence of cardinal symptoms for PD can be extracted using this framework. Note, additional data is required to develop validated algorithms for the symptoms described. Development of this platform is an effort to further streamline collection of the necessary data (Fig. 3).

Fig. 3
figure 3

Conceptual framework of the Integrated platform for the RM app leveraging the cybersecurity infrastructure from commercially available Neurosphere platform with compatibility to consumer-grade wearables and validated algorithms to determine presence and severity of symptoms. The goal of the platform is to improve patient experience and optimize neuromodulation therapy. This figure illustrates the overarching vision of the Remote Therapeutic Monitoring Platform, where the backend can accept multimodal data as objective input from wearables (skin adhesives, smart ring, or smart watch) and patient-reported outcomes surveys, diaries, etc.). This data (raw traces) can be analyzed for specific insights using validated algorithms that are available through the backend. In this example, we are showing how insights about severity and presence of cardinal symptoms for PD can be extracted using this framework.

Full size image

Back end

The RM app backend is generalized to support multiple implementations of the app. In terms of cyber-security, the cloud infrastructure is built on the commercial NeuroSphere platform and leverages best-in-class protection ensuring patient data privacy and state-of-the-art mechanisms to authenticate and establish a secure session.

The cloud services are hosted on an ISO 27001 certified infrastructure43. All data within the platform (cloud system, devices, and mobile app) are encrypted in transit and at rest. The communication channels for this data transfer are protected using TLS1.244. The data ingestion to the cloud is protected by Web Application Firewall (WAF) policies that intercept and inspect every request for any known attacks45. The cloud resources’ access points are controlled at a granular level and protected at a network level to isolate them from open internet access. The infrastructure further uses certificate pinning technique to avoid Man-In-The-Middle (MITM) attacks46.

Lastly, the backend is designed to be configurable enabling dynamic application updates with minimal interruption to users. The mobile app establishes a certificate-based authentication with the cloud services to upload data collected without additional user intervention for repeated login to the app. Data from sensors are uploaded via (1) device-to-cloud transfer (ex: Apple Watch) or (2) cloud-to-cloud transfer (ex: Fitbit or Oura). The approach is contingent on options available by the wearable manufacturer. For the current iteration, the platform leverages an SDK to interface with Apple devices and API to connect to Fitbit or Oura. Data permissions for specific metrics are obtained during configuration.

User Interface (UI)

As noted, the app was designed to deliver a personalized experience (Fig. 4). Hence, it was essential for the UI to be customizable in a study-specific manner. Patients respond to clinically validated surveys and connect to specific wearables based on study requirements. Features, such as, language, resources, data-feedback, etc. can be enabled, disabled, or modified to match patient requirements. Sensor data packets are uploaded periodically to the secure backend via automated queries. Also, when a survey is completed, any PRO data along with sensor data is uploaded; this is an additional, fail-safe mechanism with minimum user-burden for data upload in case periodic queries are interrupted.

Fig. 4
figure 4

Easy-to-navigate user interface of the RM application. The interface includes (1) Daily check-in using validated surveys (Patient Global Impression of Severity (PGIS)), (2) Timed surveys including validated quality of life metrics (Parkinson’s Disease Questionnaire 39 (PDQ-39)) and (Patient Global Impression of Change (PGIC), (3) Continuous objective sensor data collected through ubiquitous wearables (Apple Watch).

Full size image

On the study monitoring side, the clinician has access to the app or web-portal to enter therapy-related information. For the feasibility study described below, survey questions for the study were determined based on the protocol. To improve compliance, surveys are delivered in synchrony with relevant clinical events (ex: programming sessions); Associated notifications are also time specific. An additional dashboard is available to the research team to monitor survey completion and wearable usage, and to visualize real-time data analytics (Fig. 5).

Fig. 5
figure 5

Web Portal to monitor patient compliance for survey completion and wearable use. The profile tab allows the user to enter study duration and upcoming programming dates; survey data and device data tabs provide real-time data about patient compliance.

Full size image

Feasibility study

We tested the feasibility of this investigational tool by implementing it in a large-scale DBS study (NCT05269862). This was a multi-site randomized control study with ethics approval from the central institutional review board (IRB), WCG IRB. When necessary, approval from additional site-specific IRBs in the US or independent ethics committees (EC) in Europe were obtained. All research in this study was performed in accordance with the Declaration of Helsinki. Subjects with PD, who were candidates for DBS, provided written informed consent to enroll in the study. The primary goal of this study is to compare clinical outcomes in a cohort that has access to clinical intervention in-clinic and through the NeuroSphere Virtual Clinic platform versus one that is constrained to in-clinic visits. Both cohorts received a home-monitoring kit consisting of an iPhone and Apple watch, pre-configured with the RM app to enable remote data collection. For this study, we implemented a daily check-in survey based on the validated Patient Global Impression of Severity (PGIS). Additionally, monthly quality of life (qoL) measure, Parkinson’s Disease Questionnaire-39 (PDQ-39) and 3-month teleprogramming survey are included referenced to the patient’s initial programming session. Lastly, a symptom improvement survey is captured by collecting PGIS and PGIC after a titration programming session, where iterative therapy adjustments are made. Through the apple ecosystem, we collect activity metrics from HealthKit, high-resolution IMU data from watch (accelerometer and gyroscope), and core motion data to determine tremor and dyskinesia metrics. The watch interface is designed to be a single-tap engagement to account for a population that is specifically suffering from a movement disorder. Additional details about the study will be presented in a separate publication focused on the results of primary endpoints.

Statistical analysis

The SciPy python package was used for all statistical analysis (SciPy Software, scipy.org). Statistical significance was determined according to the paired Wilcoxon test with a one-sided alternative hypothesis on median values; multiple comparisons correction was applied. The format for boxplots shown in this manuscript consists of the median as a center line surrounded by a box indicating the 25th and 75th percentiles. Whiskers indicate the outer-most datapoint that is not considered to be an outlier, where outliers are datapoints that are more than a distance of 1.5 times the inter-quartile range from the nearest quartile. Outliers are indicated as single datapoints.