PUSHING THE BOUNDARIES OF CONSUMER GRADE WEARABLE DEVICES IN HEALTH CARE FOR OLDER ADULTS

Background: The proliferation of wearable and mobile devices in recent years has led to the generation of unprecedented amounts of health-related data. Together with the growing population of older adults in Canada, the increasing adoption of these technologies created a momentous opportunity to improve the way we deliver, access, and interact with the health care system. Many have recognized the opportunity, yet there is a lack of evidence on how these devices and the growing size of health data can be used to transform health care and benefit us. In Chapter 2, a review of the literature was presented to identify the current evidence of wearable technology and gaps that exist in aging research. Based on the literature review, one promising way to use wearable devices is to assess frailty, which can contribute to improving care and enhancing aging-in-place. Chapter 3 summarizes key concepts related to wearable devices including mobile health, patient-generated health data, big data, predictive algorithms, machine learning, and artificial intelligence. While in-depth mathematical representation of these big data analytics is outside the scope of this dissertation, this chapter provides foundational information along with examples found in health care settings. Objective: The overall aim of this dissertation was to investigate possible use of consumer-grade wearable devices and the patient-generated health data to improve the health of older adults. Methods: This thesis is presented as three individual studies included in Chapters 4 to 6. Study 1 aimed to investigate use of wearable devices to predict and find associations with frailty for community-dwelling older adults receiving home care service. Participants were asked to wear wearable device for 8 days in their home environment and no supervision was provided. Frailty level was assessed using the Fried Frailty Index. Other variables were collected including Charlson Comorbidity Index, independence using the Katz Index, and home care service utilization level. A sequential stepwise feature selection method was used to determine variables that are fitted in multiple variable logistic regression model to predict frailty. Study 2 extended the investigation of possible use of wearable devices for understanding frailty by examining the relationship between wearable device data and frailty progression among critical illness survivors from an intensive care unit at Kingston General Hospital. Participants were assessed for frailty using the Clinical Frailty Scale three times; at admission, at hospital discharge, and at 4-weeks post-hospital discharge. The changes in frailty level between the three time points were used to identify association with wearable device data that was collected for 4 weeks post-hospital discharge. Demonstrating evidence for wearable devices and patient-generated health data in research does not guarantee its use in real life. In Study 3, a mixed method study was conducted to explore clinicians’ and older adults’ perceptions of patient-generated health data. Focus group interviews were conducted with older adults and health care providers from the Greater Toronto Area and the Kitchener-Waterloo region. A questionnaire that aimed to explore perceived usefulness of a range of different patient-generated health data was embedded in the study design. Focus group interviews were transcribed verbatim. Line by line coding was conducted on all interviews followed by thematic analysis. Results: Results from Study 1 indicate data generated from wearable devices are closely linked to frailty level. Results showed a significant difference between frail and non-frail participants in age (p<0.01), home care service utilization (p=0.012), daily step count (p=0.04), total sleep time (p=0.010), and deep sleep time (p<0.01). Total sleep time (r=0.41, p=0.012) and deep sleep time (r=0.53, p<0.01) were associated with frailty level. A receiver operating characteristics area under the curve of 0.90 was achieved using deep sleep time, sleep quality, age, and education level (Hosmer-Lemeshow p=0.88), demonstrating that data from wearable devices can augment the demographic and conventional clinical data in predicting frailty status. Results from Study 2 demonstrated that frailty level increases significantly following a critical illness (p=0.02). Frail survivors had significantly lower daily step counts (p=0.02). Daily step count (r=-0.72, p=0.04) and mean heart rate (r=-0.72, p=0.046) were strongly correlated with frailty level at admission and discharge. Mean standard deviation of heart rate was correlated with the change in frailty status from admission to 4-week follow-up (r=0.78, p