Abstract
Introduction Common, consumer-grade biosensors mounted on fitness trackers and smartwatches can measure an array of biometrics that have potential utility in post-discharge medical monitoring, especially in remote/rural communities. The feasibility characteristics for wrist-worn biosensors are poorly described for post-COVID conditions and rural populations.
Methods We prospectively recruited patients in rural communities who were enrolled in an at-home rehabilitation program for post-COVID conditions. They were asked to wear a FitBit Charge 2 device and biosensor parameters were analyzed (e.g. heart rate, sleep, and activity). Electronic patient reported outcome measures (E-PROMS) for mental (bi-weekly) and physical (daily) symptoms were collected using SMS text or email (per patient preference). Exit surveys and interviews evaluated the patient experience.
Results Ten patients were observed for an average of 58 days and half (N=5) were monitored for 8 weeks or more. Five patients (50%) had been hospitalized with COVID (mean stay = 41 days) and 4 (36%) had required mechanical ventilation. As baseline, patients had moderate to severe levels of anxiety, depression, and stress; fatigue and shortness of breath were the most prevalent physical symptoms. Four patients (40%) already owned a smartwatch. In total, 575 patient days of patient monitoring occurred across 10 patients. Biosensor data was usable for 91.3% of study hours and surveys were completed 82.1% and 78.7% of the time for physical and mental symptoms, respectively. Positive correlations were observed between stress and resting heart rate (r=0.360, p<0.01), stress and daily steps (r=0.335, p<0.01), and anxiety and daily steps (r=0.289, p<0.01). There was a trend toward negative correlation between sleep time and physical symptom burden (r=-0.211, p=0.05). Patients reported an overall positive experience and identified the potential for wearable devices to improve medical safety and access to care. Concerns around data privacy/security were infrequent.
Conclusions We report excellent feasibility characteristics for wrist-worn biosensors and e-PROMS as a possible substrate for multi-modal disease tracking in post-COVID conditions. Adapting consumer-grade wearables for medical use and scalable remote patient monitoring holds great potential.
Introduction
Recovery from COVID can be long and unpredictable. Some patients return to premorbid health quickly while others endure a debilitating and poorly understood recovery path lasting months to years, termed ‘long-COVID’, comprising both mental and physical symptoms (reviewed in Davis et al., 2023 and Koc et al., 2022). Traditional, clinic-based methods of disease assessment are episodic, limiting their utility in a complex and undulating disease course such as post-COVID. New approaches are needed that can track longitudinal physiologic changes at the individual patient level.
Wearable devices may be ideal tools for studying complex diseases and post-COVID conditions (reviewed in Smuck et al., 2021). Biosensor metrics have been shown able to predict health outcomes for a range of chronic diseases including congestive heart failure, COPD, hypertension, diabetes, and more (Channa et al., 2021; Khondakar & Kaushik, 2022; Rodriguez-León et al., 2021; Singhal & Cowie, 2020). More recently, smartwatches have been used to detect pre-symptomatic COVID infections (Mishra et al., 2020). Common and affordable smartwatches and fitness trackers might support remote patient monitoring (RPM) if the emergent data is high-enough quality to inform safe decision making (Iqbal et al., 2021, Kwok et al., 2021; Lu et al., 2020, Kang & Exworthy, 2022). Remote patient monitoring is particularly important for patients living in remote and rural areas where in-person care can be limited (Fraser et al., 2022; Liao et al., 2019; Canali et al., 2022).
Feasibility data for wearable devices varies with population, technology, and monitoring protocol and little has been published for post-COVID conditions, especially in rural settings. Our study describes the feasibility of using common, consumer-grade wrist-worn biosensors (e.g., FitBit fitness trackers) for disease parameter tracking in post-COVID conditions in a rural community, including both technical aspects (e.g. data quality/completeness) and the patient experience.
Materials and Methods
Patient Cohort
Patients were recruited from an early supported discharge (ESD) program in Camrose, Alberta, Canada, which has a population of just under 21,000 residents. The ESD program was originally designed to support patients recovering from acute stroke (Chouliara et al., 2023); however, during the COVID-19 pandemic, it was adapted for COVID-19. Patients were referred to the program either directly from hospital (at discharge) or through community clinics. The ESD team is a rehabilitation team that delivers intensive rehabilitation programs in the home using both in-person visits and telemedicine (video or phone). The team typically includes occupational therapy, physical therapy, speech-language pathology, social work, nursing, therapy assistants, recreation therapy and a psychologist. Patients were also recruited through a virtual post-covid pilot program and were located throughout Alberta, primarily in rural settings. Patients were referred to this program through the Rehabilitation Advice Line. Rehabilitation in this program was provided solely via telemedicine.
All ESD patients were screened for eligibility at program start. Eligible patients were ambulatory, with mild to moderate functional impairment and sufficient cognitive capacity to participate in the program. Excluded patients were those without internet connectivity, who were unable to complete surveys (ex. language and speech deficits, non-English-speaking), or for whom device retrieval was deemed to be unlikely by the study investigators.
Data Sources
Data sources included: 1) wearable devices, 2) health records, 3) digital surveys, and 4) patient interviews. Health record data comprised diagnostic, medication, and procedure codes occurring in the year before enrollment and during the observation period.
Devices
FitBit Charge 2 devices were provided to each patient at enrollment. The FitBit Charge 2 is a wrist-worn fitness tracker that includes a triaxial accelerometer, an altimeter, and an optical heart rate tracker. Parameters collected included activity (steps per day), heart rate (beats per minute), and duration (minutes) and quality (% deep) of sleep. Data was sent via Bluetooth to a patient-owned smartphone, uploaded to the FitBit cloud database and then extracted through the third-party platform (Fitabase; https://www.fitabase.com/).
Monitoring Protocol
Patients received an orientation to the device on study enrollment and were instructed to wear it as much as possible, up to 24 hrs a day (except when charging). There were no specific instructions on how to use device data and patients were free to share data with the clinical team if desired; however, data was not directly transmitted to the clinical team. If the patient had questions about the device or technical issues, a research nurse was available for support by phone during daytime hours.
Surveys
Patients received digital surveys via SMS text or email (per patient preference) using the RedCap (Harris et al., 2008) survey platform. Baseline health status surveys included measures of mental and physical health and a technology experience survey (Comprising two sections: Tech comfort (7-items, avg score [0-5]), Health Literacy (4-items, avg score [0-5]) adapted from the Telehealth Usability Questionnaire (TUQ) (Parmanto et al., 2016). During the follow-up period physical and mental symptom surveys were delivered daily and bi-weekly, respectively. English versions of validated self-reported screening scales were used to measure severity of stress, anxiety, and depression symptoms, including the Perceived Stress Scale (PSS), a 10-item questionnaire with a Cronbach’s alpha of >0.70 used to assess level of stress in the previous month (PSS; PSS score ≥14 indicates moderate or high stress) (Cohen et al., 1983); the Generalized Anxiety Disorder 7-item (GAD-7) scale, a 7-item questionnaire with a Cronbach’s alpha of 0.92 and used to assess the self-reported levels of anxiety in respondents in the two weeks prior to assessment (GAD-7 score ≥10 indicates likely generalized anxiety disorder [GAD]) (Spitzer et al., 2006); and the Patient Health Questionnaire-9 (PHQ-9) a 9-item questionnaire with a Cronbach’s alpha of 0.89 and used to assess the severity of depression symptoms (for PHQ-9; a score ≥10 indicates likely major depressive disorder (MDD) (Kroenke et al., 2001).
Detailed description of physical symptom surveys and results can be found in the supplementary materials, Table S1. In short, 18 total symptom scales (6-point scale, ‘absent’ [1] to ‘very severe’ [6]) were grouped by body system (constitutional, gastrointestinal, neurologic, respiratory, cardiovascular). At discharge, patients received an exit survey that included feedback on the devices, also adapted from the TUQ.
Interviews
At the conclusion of the study patients were approached for interview. Questions focused on technology usability, acceptance, and perceived barriers to use, such as data privacy/security. Author MW completed interviews virtually using Zoom and curated auto-generated transcriptions.
Study Outcomes
Primary Outcome: Protocol Adherence
Our primary outcome was protocol adherence, defined as device wear time (% of minutes with analyzable heart rate data) and survey completion rates. Patient interviews and survey responses added context for this outcome, exploring patient perceptions and experience.
Secondary Outcomes: Biosensor Parameter Correlations
Secondary outcomes were associations between 1) the primary outcome (adherence) and patient characteristics (e.g., age, sex, symptom severity, technology readiness measures, and time under observation), and 2) device biometrics (activity/HR/sleep) and disease outcomes (physical and mental symptoms).
Statistical Analysis
Descriptive statistics (mean and standard deviation [SD]) were calculated for primary and secondary outcomes. We used Pearson correlation coefficients and weekly aggregated data to explore secondary outcomes. We performed a within-subject longitudinal analysis and paired t-tests to evaluate for changes in symptom severity over time, comparing weeks 1-4 and weeks 8+ for those subjects observed for more than 8 weeks (N=5).
Qualitative Analysis
We conducted thematic analysis of interview transcripts using a combined deductive and inductive approach, aided with NVIVO software. Two team members (MW and JL) reviewed original transcripts to identify themes of interest and compared results. Where there was disagreement, consensus was obtained through discussion. Smaller themes were sequentially grouped into larger categories until core themes were identified.
Funding and Ethics
This study was registered and approved by the University of Alberta Research Ethics Board (Pro00113943).
Results
Participant characteristics
Between November 2021 and May 2022, 18 patients were approached for enrollment; three declined and one patient was ineligible due to lack of wireless connectivity. After enrollment, 3 of 14 patients withdrew consent, and 1 was removed due to technical issues with their device, leaving 10 patients for the final analysis. Seven patients consented to an interview. The overall recruitment rate was therefore 61.1%.
Table 1 shows baseline patient characteristics. Mean age was 53 years (SD 15.0) and 80% (8/10) of patients were female. Five (50%) had been hospitalized with COVID (average length of stay = 41 days) and four (40%) had required ventilatory support in the ICU. The remaining 50% were referred from community clinics. On average, patients had fewer than one documented medical condition prior to their COVID-19 diagnosis and half (5/10) owned their own smartwatch. The most common physical symptoms were ‘constitutional’ (ex. fatigue, poor sleep), followed by ‘respiratory’ and ‘cardiovascular’. Average mental health symptoms scores were GAD-7 = 15.7 (severe anxiety), PHQ-9 = 14.8 (moderate depression) and PSS = 22.7 (moderate stress). Patients had moderate to high levels of familiarity with technology (ex. comfort with technology = 3.6/5; health literacy = 3.7/5). Detailed patient characteristics, including comorbidities and medications, are shown in Supplementary Table S1 and Figures S1 and S2.
Descriptive analysis
Table 2 summarizes descriptive statistics for primary adherence outcomes (group aggregate). Overall, 575 patient days of patient monitoring occurred across 10 patients, patients were observed for 58 days on average and half (N=5) were monitored for 8 weeks or more. For the primary outcome (adherence), heart rate data was available for 91.3% of study hours and physical and mental symptom surveys were completed 82.1% and 78.7% of the time, respectively. Overall physical symptom burden was generally mild with an average overall rating of 0.8/4; constitutional and respiratory symptoms being the most frequent. Average group activity, sleep duration and quality (proportion time in deep sleep) were highest in weeks 4-8. Average resting heart rate was relatively constant over time. To test for within-subject effects, we compared weeks 1-4 and weeks 8+ for those patients staying longer than 8 weeks (Table S2). Except for respiratory symptoms (mean difference –0.23, p=0.03), temporal changes did not reach statistical significance.
Group-level associations
Survey response rates were correlated with device wear time (r=0.67, p=0.03), however there were no statistically significant associations between wear-time and other patient characteristics (e.g. demographics, symptoms, previous technology experience or time under observation; Table S3). Table 3 shows group-level associations between biosensor metrics (weekly aggregate). Resting heart rate was positively correlated with stress (r=0.360, p<0.01); step count was correlated with anxiety (r=0.289, p<0.01) and stress (r=0.335, p<0.01). Correlations for sleep time were not statistically significant, however trends suggested negative correlations with physical symptoms (r=-0.211, p=0.05)) and anxiety (r=-0.186, p=0.09)) and a positive correlation with stress (r=0.208, p0.06)).
Individual sensor data: case examples
Figure 2 shows data for two patients, illustrating the complexity and heterogeneity of individual data. At enrollment, Patient A experienced marked physical symptoms, a high resting heart rate (79-83 bpm) and low daily step counts (approx. 1km). Over time, their physical symptoms improved, resting heart rate decreased, and daily activity increased. In contrast, Patient B’s physical symptoms remained prominent throughout observation. Their resting heart rate was lower than Patient A (46-54 bpm) and didn’t significantly change over time. Their activity levels didn’t show a clear temporal change. Mental health symptoms were prominent in both patients.
Flow diagram of study recruitment.
Sample longitudinal data for Patient A (left) and B (right), including results from symptom surveys, heart rate and total daily number of steps.
Surveys results
Figures 2 and 3 display survey results; additional survey data is presented in supplementary materials (Figure S1). Proportions herein represent overall levels of agreement (agree or strongly agree) vs. disagreement (disagree or strongly disagree). Of those who responded, the majority liked using the FitBit system (5/8 [63%]), found it easy to learn (6/8 [86%]) and simple to use (7/7 [100%]); no patients found the device interfered with their lifestyle and only one (13%) found the device uncomfortable. Three-quarters (6/8 [75%]) of the patients found that wearing a device made them feel safer and 4/8 (50%) felt it helped them to better understand their disease. A minority (2/8 [25%]) used the device to help decide when to seek medical care and no patients (0/8) reported that the devices caused anxiety.
Survey results for questions pertaining to device acceptability with answers ranging from strongly disagree to strongly agree.
Survey Results for questions pertaining to psychological impact and clinical use with answers ranging from strongly disagree to strongly agree.
Theme analysis of interview transcripts
Two male and five female patients completed exit interviews. Key themes are summarized below, and representative quotes are presented in Table 4.
Support in isolated environments
Most felt that wearable devices are likely to improve medical safety for rural communities in the future, possibly through symptom, activity, and vital sign data transmission to clinical teams, family members and caregivers.
Gaining disease insights
Some patients spontaneously use device data (mostly heart rate) to help them interpret physical symptoms, even without medical guidance. The impacts of biosensor data on mental health were varied. Some found that normal heart rate readings reassured them while others found that elevated heart rates worsened anxiety about their disease.
Concerns around privacy
Few were concerned data privacy and security. One participant was hesitant to share their data with for-profit device companies, specifically.
Potential for research
Almost universally, participants felt wearable devices held great promise as tools for clinical research and expressed a willingness to contribute their data to the research community. Participants were motivated by a desire to see advancements in understanding and treatment of post-COVID conditions.
Discussion
We report excellent feasibility characteristics for wrist-worn fitness trackers in post-COVID disease monitoring in rural areas, including hypothesis generating associations suggesting that biosensor data can deliver disease insights at the individual level. Biosensor data was complex and heterogenous, highlighting the need for machine learning techniques to separate signal from noise. Our qualitative work illustrated a unique lived experience for rural patients with technology and unmet needs for remote monitoring tools that integrate with existing clinical care models to ease transitions from hospital to home.
Compared to existing studies, we observed high levels of protocol adherence. For common disease targets like COPD and CHF, RPM protocol adherence is typically low and decreases over time (Singhal & Cowie, 2020; Stehlik et al., 2020; Wu et al., 2021). Our post-COVID population was younger than most existing RPM cohorts, had relatively few comorbidities and were generally comfortable with technology, perhaps explaining our favorable result. Also, most existing studies RPM protocols use medical-grade devices, which are often unfamiliar to patients and challenging to operate. Consumer choice in technology is key driver of success for RPM and consumer-grade devices that offer non-medical applications present an opportunistic substrate for building scalable RPM platforms (Curry, 2023). The drawback of consumer-grade biosensors is that sensor accuracy is often unknown and cannot be assumed clinical grade (Bent et al., 2020). However, as sensor accuracy and signal detection algorithms improve for longitudinal patient data, we will be able to better compensate for imperfection in underlying discrete data. Learning to process data from consumer-grade biosensors is foundational for just-in-time medical decision making, especially in relatively healthy patient populations who haven’t been prescribed a medical grade device (Takahashi et al., 2022). Our study underscores the need to develop patient-centered technologies built with the non-medical consumer in mind.
Our group level associations and individual case examples highlight the potential for smartwatches to deliver novel disease insights in scientific studies. In our post-COVID patients, anxiety and stress correlated with activity and resting heart rate while physical symptom correlated with sleep. This granular, continuous data paints a rich picture of disease at the individual level, forming a foundation for precision medicine and improving on traditional forms of episodic measurement. Given our small sample size, our observed group-level associations are hypothesis generating only. Subjectively, the heterogeneity seen across individual data supports the hypothesis that post-COVID conditions comprise a range of distinct phenotypes rather than a single entity (Lusczek et al., 2021; Osuchowski et al., 2021), a possibility that will hopefully be further elucidated in ongoing clinical trials that utilize wearable devices (Moore Vogel, 2023). As more studies using wearable biosensors emerge it will be important to create and adhere to standards in metric derivation and validation so that results are clinically applicable.
Our patient interviews provided a rich context for interpreting quantitative data and brought to the fore a range of patient experiences that are key for understanding the future impacts of wearable devices in medicine. Continuous physiologic data can amplify or dampen anxiety around disease depending on the individual. In some instances, normal device data helped to reassure when symptoms were fluctuating unpredictably. For others, abnormal heart rate data worsened anxiety around mild or absent physical sensations. The potential impacts of personalized biodata on mental health are poorly described in existing literature and for some it could contribute to excess healthcare utilization (Rosman et al., 2020). Our result highlights the need to better understand these impacts through further research and to improve medical support patients who choose to wear devices. Interestingly, concerns about data-sharing and privacy were infrequently raised by our patients but are known to be major determinants of success for wearable devices in healthcare, overall (Banerjee et al., 2018; Cilliers, 2020).
Patients unfamiliar with wrist-worn devices in our study required a significant amount of time and support during device set-up. For those who owned a device already, changing to a new one was unpopular, and some patients refused enrollment or withdrew consent as a result. Human factors like this are critical determinants of success for RPM platforms. Shin et al. (2019) found that wearables are best accepted by patients who are motivated to monitor a chronic condition. Yin et al. (2022) reported that convenience, social influence, and expectation of improved health are important for uptake, while cost and perceived risk are less so. In rural settings, such as ours, perceived benefits for wearable technologies might be elevated as in-person services are often inaccessible (reviewed in Brahmbhatt et al., 2022). In future work using smartwatches rather than fitness trackers would likely yield even stronger feasibility data given the attractiveness of their non-medical applications like text messaging, music, and social networking, all of which encourage wear-time. Indeed, many patients in our study were particularly reluctant to use a fitness tracker if they already owned a smartwatch.
Limitations
Our study has several limitations. Most notably, our sample size was small, and it is unclear if our findings can be replicated a larger population. As conditions changed during our study (vaccines emerged and prevalent viral strains shifted) we observed that the post-COVID phenotype also changed. In the early stages of our work, we observed more severe respiratory symptoms and most patients had been hospitalized, often recovering from a period of mechanical ventilation. In the later stages, patients were often referred from community clinics with more mild respiratory disease but more severe constitutional symptoms (ex. fatigue or poor sleep). The shifting disease phenotype makes it challenging to infer similar results for future endemic states for COVID. Importantly, in this setting our patients were visited near-daily in their homes or via telemedicine by therapists, which could have artificially increased protocol adherence, and critically, a relatively low recruitment rate (60 percent) suggests that population-level impacts for wearable technologies remains uncertain.
Conclusions
We find promising feasible characteristics for wrist-worn devices in remote disease tracking for post-COVID conditions in rural communities. Our data are foundational for future testing of consumer-grade devices in the medical sphere, highlighting the need for co-created RPM platforms designed with patient/consumer technology preference in mind. Such innovations have great potential to improve healthcare access and safety for otherwise isolated populations.
Supplementary Materials
Baseline technology comfort survey responses (N=10).
Baseline health literacy survey responses (N=10)
Acknowledgements
We acknowledge the extensive efforts of the Camrose Early Supported Discharge team (Dana Norton, Alyssa Rose, Sharene Lamson) in providing clinical and administrative support. We also acknowledge the contributions of the Glenrose research team (Elton Lam, Tod Vandenberg, Stephan Pham) for their provision of analytic support. Funding for this work was generously provided by Covenant Health and Alberta Health Services. No direct funding was received from FitBit.
References
Subject Area
Reviews and Context
0
Comment
0
TRIP Peer Reviews
0
Community Reviews
0
Automated Services
0
Blogs/Media
Author Videos