Abstract
INTRODUCTION
Therapy for Parkinson’s disease (PD) typicallybegins with dopaminergic medication (e.g., levodopa); however, dopaminergic medication management is complicated by disease progression and side effects such as dyskinesias (uncontrolled, irregular movements) and unpredictable On/Off fluctuations can occur [1]. Clinicians may attempt to manage fluctuations and dyskinesias by adjusting levodopa dose and frequency or adding a dopamine agonist, COMT inhibitor or MAO-B inhibitor [2]. When fluctuations and side-effects can no longer be adequately controlled, advanced therapies such as subcutaneous apomorphine infusion, deep brain stimulation (DBS), and levodopa-carbidopa intestinal gel (LCIG) are often recommended [3, 4].
Two challenges associated with advanced therapy include patient access and cost [5]. More specifically, it is important to identify and provide accessibility for patients who may qualify for advanced therapies to improve quality of life, and of equal importance filter patients who are not appropriate to minimize risk and costs [6]. These therapies are invasive, costly, and not without risk, making it critical to accurately and efficiently identify those patients who would benefit most from advanced therapy. There is a significant emotional burden on patients over many years, especially if the anticipated therapeutic improvement is never achieved [7, 8].
There are currently no uniform criteria for identifying appropriate advanced therapy candidates [9–11]. The levodopa response is typically a strong indicator of advanced therapy outcome [4, 12–15]; however, methods for accurate and efficient evaluation are severely lacking. The rating scales used in clinical assessment, most commonly the Unified Parkinson’s Disease Rating Scale (UPDRS) [16], only provide a snapshot of the patient and can suffer from poor inter- and intra-rater reliability, particularly in the bradykinesia domain [17–21]. Additionally, the pattern and severity of motor symptoms and dyskinesias vary greatly throughout the day; yet, in-clinic testing over multiple medication doses is not practical. To address the above concerns regarding in-clinic assessment, subjective screening questionnaires and software apps have been developed to determine appropriate candidates based on a levodopa response (e.g., motor symptoms improve by greater than 25%), the presence of motor fluctuations and dyskinesias, and minimal to no cognitive decline [10, 23]. However, home diaries rely entirely on the patient’s perception and have poor patient compliance [24, 25]. Patients may also be unreliable at self-assessment and have difficulty distinguishing between motor symptoms, dyskinesias, and voluntary movements, making evaluating medication responses particularly challenging [26–28].
In addition to the above concerns regarding a lack of adequate screening tools, access to movement disorder specialists can be financially burdensome for geographically disparate patients and those unable to travel. The geographic location of movement disorder centers can limit patient access to well-trained clinicians and effective symptom management [29]. Recently, web-based assessments have been shown to offer a feasible format for assessing PD-related impairment from home [30]. Motion sensor-based systems can also be used to objectively quantify tremor, bradykinesia, and dyskinesia unsupervised at home with high compliance to track motor fluctuations and medication responses throughout the day [20, 31–35]. The quantitative features extracted from motion sensors highly correlate with clinical tremor, bradykinesia, and dyskinesia scores [20, 32] and demonstrate improved responsiveness for measuring changes ON and OFF medication and DBS compared to clinical ratings [19, 21]. The objective was to determine the impact of remote motion sensor-based monitoring on advanced therapy referral rate in patients with advanced PD and if the data captured in patients’ homes differed in patients who were referred and those who were not.
METHODS
The methods described herein represent a retrospective, exploratory, secondary analysis of data collected in a single-center, randomized, prospective study examining the cost-effectiveness of remote monitoring in advanced PD using objective sensors [36]. In that study, outpatients from a tertiary regional movement disorders clinic with a Mini-Mental scale [37] score >24 and diagnosed with idiopathic, advanced PD were invited to participate. Advanced PD was defined as having a UPDRS motor complication score ≥4. The protocol was reviewed and approved by the Complejo Asistencial Universitario of Burgos Institutional Review Board and all participants signed an informed consent before being enrolled in the study. Participants were randomized to either home-based motor monitoring with wireless motion sensor technology (Kinesiatrademark, Great Lakes NeuroTechnologies, Cleveland, OH) or to traditional office-based management (Control) using random digit tables. Twenty patients began the study in each group, with 17 patients in the Kinesia group and 18 patients in the control group completing the study. Both groups were largely comparable in their clinical and demographic variables at the start of the study, except there were more patients already on DBS or LCIG in the Kinesia group compared to the control group (6 versus 1,
All patients were assessed during the “on” motor status in the clinic every four months over the course of a year by the same movement disorders neurologist (EC) and PD medications and other treatments could be changed at any time based on patients’ clinical needs. However, in the Kinesia group only, PD motor symptoms were monitored at home one day per month using the Kinesia system. For these patients, the neurologist reviewed Kinesia reports detailing motor symptoms and dyskinesia throughout the day to aid in disease management decisions.
The Kinesia system includes a tablet software app, wireless finger-worn motion sensor unit, and automated web-based symptom reporting. On assessment days, the patient wore the motion sensor on the index finger of his/her more affected hand while the software app guided him/her through automated motor assessments approximately ten times per day at approximately one-hour intervals (Fig. 1). The assessments consisted of five motor tasks performed for 15 seconds each. The tasks included arms resting to evaluate rest tremor and dyskinesia, arms extended to evaluate postural tremor and dyskinesia, and repetitive finger-tapping, hand opening-closing, and pronation-supination to evaluate bradykinesia. Data from the motion sensor was automatically uploaded via mobile broadband to a secure cloud server for processing into 0–4 scores previously shown to be highly correlated with clinician ratings [20, 32]. Reports detailing tremor, bradykinesia, and dyskinesia severities throughout the day as well as daily averages and standard deviations were made available on a secure webserver for the neurologist to review to aid in therapy adjustment for the Kinesia group.
Throughout the study, the neurologist recommended various therapy changes based on standard clinical practices and the information provided in the Kinesia reports (Kinesia group only). This included the option to recommend advanced therapy (e.g., DBS or LCIG), which was based the clinical judgment of the neurologist and standard criteria (e.g., motor and non-motor symptoms, motor complications, cognitive function, age, caregiver support, etc.) [4]. We performed a
In addition to examining referral rates, within the Kinesia group, we also investigated if various kinematic features could be extracted from the Kinesia reports to objectively distinguish patients who were recommended by a clinician to receive advanced therapy from those who were not. These features included resting tremor, bradykinesia, and dyskinesia severities, which were calculated as the intra-day averages of all resting tremor, bradykinesia, and dyskinesias scores, respectively. Additional features included tremor and bradykinesia fluctuations, which were calculated as the standard deviation of the tremor and bradykinesia scores each day. These features were compared between patients recommended and not recommended for advanced therapy using a Wilcoxon rank sum test.
RESULTS
Baseline patient characteristics for participants included in the analyses are listed in the top section of Table 1. The bottom of Table 1 summarizes the advanced therapy status of patients in each group during the one-year study. Out of the 11 patients in the Kinesia group, 7 were recommended by the clinician to receive advanced therapy, for a referral rate of 63.6%. However, only 2 of the 17 patients in the control group were recommended by a clinician to receive advanced therapy, for a referral rate of 11.8%. Out of those recommended for advanced therapy, 4 patients in the Kinesia group and 0 patients in the control group actually received the advanced therapy during or immediately following the one-year study.
Figure 2 shows reports for one day of Kinesia monitoring from two representative patients in the Kinesia group (one who was recommended for advanced therapy and one who was not). Discussions with the neurologist revealed several key features in the reports that helped inform her decision on whether or not to recommend advanced therapy. Specifically, the report for the patient who was recommended for advanced therapy includes a large number of scores greater than 2 (color-coded orange and red) indicating severe bradykinesia at many time points throughout the day. The report also shows that symptom scores varied greatly throughout the day indicating motor fluctuations. Finally, the report shows this patient exhibited dyskinesias of varying severity throughout the day. Conversely, for the patient not recommended for advanced therapy, the report is mostly green with low severity scores indicating symptoms did not fluctuate and were very mild throughout theday.
The observations described above provided a basis for identifying potential features that could objectively discriminate patients who might be and not be candidates for advanced therapy. Kinesia-measured average bradykinesia severity, bradykinesia fluctuation, and average dyskinesia severity were significantly different between patients recommended and not recommended for advanced therapy by the clinician; however, tremor severity and fluctuation did not differ significantly (Fig. 3).
DISCUSSION
The results suggest that objective, wearable sensors can help identify patients who may be candidates for advanced therapies. The Kinesia group had significantly greater advanced therapy referral rates than the control group (63.6% vs. 11.8%). As the only difference in procedures between groups was the use of Kinesia home-based motor monitoring, the increased referral rate can likely be attributed to the neurologist’s improved knowledge of patients’ motor fluctuations and symptom severities. This suggests that home-based monitoring could lead to increased advanced therapy referrals. A clinician may be more comfortable referring a patient when symptom severities and fluctuations are confirmed by objective home monitoring with sensors. Not only was the referral rate increased for the Kinesia group, but also the actual initiation of advanced therapy (36.4% v. 0%), indicating that home-based monitoring can lead to more than simply advanced therapy consideration. While a number of factors can influence the ultimate decision on whether or not a referred patient decides to avail advanced therapy (e.g., age, cognition, resources, cost), a patient may be more likely to agree to have a procedure performed after seeing objective evidence that the therapy is likely to work. It is also important to note that in addition to increased advanced therapy initiation, there were also more patients recommended for advanced therapy in the Kinesia group who, for a variety of reasons, did not end up receiving advanced therapy. Future work will be necessary to determine how best to optimize the sensitivity and specificity in identifying advanced therapy candidates using sensor-based remote monitoring.
While reports such as those shown in Fig. 2 can help a clinician visualize symptom severities and fluctuations throughout the day, clinical interpretation is still required. Automated processing of motion sensor data without requiring expert clinical interpretation could improve the efficiency of home-based monitoring for identifying candidates for advanced therapy. Demonstrating that objective measures of bradykinesia, fluctuation, and dyskinesia captured by Kinesia differed significantly between patients who were and were not recommended for advanced therapy based on clinical judgment (Fig. 3) suggests that these features may be able to objectively classify patients who would and would not be potential candidates for advanced therapy. The lack of statistically significant differences for tremor is likely due to the small sample size and absence of tremor in the majority of the study population. Therefore, tremor severity and fluctuation should likely still be considered when investigating advanced therapy referral prediction algorithms. One limitation of this paper is that the data was collected during a larger protocol in which the clinician was not blinded to the sensor data. Therefore, the sensor data and the clinical decision to recommend advanced therapy were not independent. Future studies involving a large number of patients are planned to develop and evaluate algorithms for automatically screening patients for advanced therapy referral. These studies will include blinding the clinician to the sensor data and confirm selection of appropriate candidates by monitoring them after the advanced therapy is applied to make sure there is an actual benefit. This will improve advanced therapy candidate selection as post-surgery outcomes should be the ultimate gold standard.
In this study we did not differentiate between DBS and LCIG referrals as currently there is no comparative data to support the use of one advanced therapy over another and the decision to use a specific therapy depends largely on patient and clinician preference [3, 15]. However, as advanced therapies become increasingly widespread, remote monitoring could provide comparative data on who responds best to various advanced therapies. This could lead to differentiation in advanced therapy referrals and provide the data necessary to develop models for automatically identifying candidates for specific therapies.
Sensor-based monitoring will never replace the judgment of a clinical team as there are many variables that go into the decision to recommend advanced therapy. However, the ability to remotely screen potential candidates has major implications for healthcare. Remote screening using wearable sensors could serve as a first step to automatically flag patients and let clinicians know it might be time to pursue advanced therapy. This type of screening might also help identify candidates that would not otherwise consider advanced therapy. New research suggests DBS may be beneficial in patients with early motor complications that emerge before the appearance of severe disabling motor complications [38]. Identifying potential candidates as early as possible could help these patients receive therapy sooner and improve their quality of life.
While academic medical centers may appropriately identify advanced therapy candidates, remote screening of patients who live far from expert centers could greatly improve patient selection and increase access to advanced therapy. This would also reduce disparities by allowing PD patients to be pre-screened from their homes and eliminate the need for travel to a PD specialty center until after being identified as a potential candidate for advanced therapy. Remote monitoring could also benefit patients after they receive advanced therapy by determining if a patient is responding well to the therapy or is in need of a therapy adjustment. Remote, sensor-based monitoring will likely play an important role in informing therapy decisions and help identify candidates for advanced therapy.
CONFLICT OF INTEREST
This study was funded by Great Lakes NeuroTechnologies Inc. Dr. Heldman and Dr. Giuffrida have received compensation from Great Lakes NeuroTechnologies Inc. for employment. Dr. Cubo received travel funding from AbbVie, Allergan, and UCB Pharmaceuticals, and received research support from the International Parkinson’s Disease and Movement Disorder Society, World Federation of Neurology, Junta de Castilla y León, and Great Lakes NeuroTechnologies Inc.
Footnotes
ACKNOWLEDGMENTS
The authors would like to thank Maureen Phillips for assisting in data management and Chris Pulliam and Jerry Vitek for providing thoughtful comments on the manuscript.