Introduction

It is increasingly recognized that rising global temperatures due to climate change present a threat to human health, wellbeing, and performance1,2,3. Periods of extreme heat are associated with increased ambulance calls4,5 and hospital admissions6,7,8 due to respiratory, cardiovascular, or renal diseases, mental health disorders, dehydration, and heat stroke9,10,11. Extreme heat events are the leading cause of death among all weather-related factors12,13,14,15, and are predicted to be more intense, more frequent, and longer in the 21st century16. Populations such as the elderly, individuals living alone, and those with chronic diseases are at greater risk17,18. Heat-related mortality among individuals over 65 years of age was recently estimated to have increased by 68% in the last 2 decades1.

Epidemiological evidence has demonstrated that older individuals are at an increased risk of heat-related mortality within their residence17,18. Considering individuals, especially the elderly, spend more than 80% of their time indoors19, there is an urgent need to better understand how individuals respond and react in their homes during periods of extreme heat to develop appropriate evidence-based heat protection strategies. Our current understanding of the human physiological and behavioral responses during extreme heat exposure are limited to lab-based simulations that cannot fully reflect responses in the natural indoor environment. Laboratory-based studies have typically employed short (<3 h) exposures20,21,22,23, with limited data on longer (up to 9 h) exposures24,25. A surge in heat-related morbidity and mortality typically occurs 1-2 days after peak air temperatures26,27,28, highlighting that acute exposures may not capture the compounding physiological strain that leads to compromised health and well-being. While more recent efforts have been made to simulate 3-day extreme heat events29, they require continuous oversight by multiple research personnel and advanced infrastructure. Moreover, these multi-day laboratory-based simulations have not fully replicated natural diurnal fluctuations in ambient conditions observed during extreme heat events26,30. Laboratory-based studies also impose strict pre and/or per exposure instructions that are not representative of an individual’s behaviour (non-structured physical activity, hydration, etc.) during extreme heat events. Collectively, current laboratory-based studies lack the ecological validity required to examine the synergistic effects of environmental, physiological, and/or behavioural factors that contribute to heat-related health risks.

Recent advancements in wearable technology may offer an opportunity for researchers and clinicians to better understand how various populations respond in-situ during extreme heat events. The current potential to remotely monitor key physiological and behavioural variables alongside the environmental conditions experienced by the individual requires aggregation of various off the shelf components from different manufacturers. While it may still be possible for researchers to collect the data by aggregating various commercially available devices, each device may have its own proprietary third-party software or mobile application that it must communicate with. This necessary step with some devices for data synchronization may increase participant burden, especially if multiple independent applications require regular participant interaction. Although some devices have automatic synchronization with their manufacturers cloud infrastructure, researchers would still be tasked with aggregating the data, often through an application programming interface (API) which may be a technical barrier requiring expertise. Furthermore, some devices and software may collect participant-generated data that is unrelated to the primary research question31, and may be redistributed to other third-party organizations32,33, compromising data-governance. Researchers have made enhancements to current wearables, including innovative applications to add surveys to smartwatches from Apple and Fitbit (e.g., Cozie App34). However, these wearable devices, and others, do not permit researchers to have direct access to the processing techniques of onboard sensors that derive other key parameters of interest creating replication issues. Furthermore, they are limited by the availability of other devices within their respective ecosystem. Lastly, when these devices are updated with new algorithms or sensors, or reach the end of life, their previous internal hardware and algorithms may not be made publicly available, further obfuscating how the original data was derived.

To address the technological gaps limiting a comprehensive monitoring of physiological and behavioural responses during exposure to environmental stressors in the built environment, we developed HeatSuite, a low-cost, scalable, customizable, and fully data-governed remote data acquisition system that can monitor the local environmental conditions and record physiological and behavioural responses of free-living individuals in their natural environment. HeatSuite is a multimodal sensor platform that minimizes user interactions, including automatic ingestion of participant data, that is built upon open-source hardware and developer-friendly software that enables continued expansion and modularity to meet the current and future needs of researchers and clinicians (see Fig. 1 for a simplified network overview). The purpose of the current study was to demonstrate the feasibility of deploying HeatSuite in populations at risk of heat-related complications. The systems functionality will be described, and a longitudinal compliance assessment will be presented using the physiological and survey data collected over a 28-day deployment period amongst older adults living in social housing in Montreal, Canada.

Fig. 1: Graphical depiction of HeatSuite.
figure 1

The figure demonstrates how data is collected from the participant, stored, and visualized by HeatSuite within its own ecosystem eliminating third-party data sharing. (1) In addition to the onboard sensors that measure heart rate via photoplethysmography, near wrist temperature, and accelerometry, the smartwatch (Bangle.js2) operates as the primary data acquisition device with i) additional wireless devices, including, but not limited to, the blood pressure monitor, oral temperature, and chest based heart rate sensors; and ii) interactive on-device data collection such as ecological momentary assessment questionnaires and micturition frequency recording. (2) The environmental node monitors the local environmental conditions (see text for a more detailed description), and monitors for smartwatches in proximity for localization monitoring, and automatically downloads data collected and stored on the smartwatch. (3) If the environmental node is connected to the internet via a Wi-Fi or cellular network and associated with a HeatSuite Cloud Software instance (code base written in PHP and JavaScript), raw data will be automatically uploaded to the online cloud, processed, and collated programmatically into predefined tables in a password protected My Structured Query Language (MySQL) database. (4) The HeatSuite Cloud Software provides a password protected web frontend for visualization of data to monitor end-user responses, and device connectivity and integrity.

Full size image

Results

Participant characteristics

A total of 21 participants were enrolled in the study, and their characteristics are presented in Table 1.

Table 1 Sex segregated participant characteristics (n = 21) on the first day of observation
Full size table

Smartwatch

On average, participants wore the smartwatch for 18.5 ± 2.8 h (or 77 ± 12%) per day demonstrating low moderate compliance (Fig. 2). Out of 588 total days, a high moderate compliance was observed for valid daytime wear (n = 528, 90%), and low compliance for valid full-day wear (n = 289, 49%). No main effect of week was observed (p = 0.10) in daily wear time (week 1: 19.1 ± 0.4 h; week 2: 18.5 ± 1.2 h; week 3: 18.6 ± 0.4 h; week 4: 18.5 ± 0.5 h).

Fig. 2: Smartwatch wear patterns over the 28-day observation period.
figure 2

Percent compliance per day for users wearing the smartwatch for daytime (>12 h), or full day (>20 h) wear (top), and individual wear time (white circles) and mean wear time (orange line) for each day during the observation period (28 days, bottom).

Full size image

Body mass

One participant was unable to complete the mass task due to physical restrictions (wheelchair bound). Therefore, body mass measurements from 20/21 participants were available for analysis. On average, participants demonstrated a near optimal compliance rate for daily mass measurements (94 ± 5%). All mass measurements taken were obtained 58 ± 41 min from the programmatic nudge (Fig. 3). No main effect of week was observed for compliance (p = 0.10, Fig. 3). Of a possible 560 mass measurements over the observation period, missing data included user error (n = 29), and technical errors out of the participant’s control (n = 4). As a secondary analysis, participants demonstrated low compliance with taking their mass barefooted to permit bioelectrical impedance analysis for body composition (59 ± 29%).

Fig. 3: Time of day and compliance analysis for completing study tasks using HeatSuite.
figure 3

Individual task completion time-of-day during the observation period (left) with corresponding mean compliance binned by week and time-of-day (right) for (top to bottom): body mass, blood pressure, oral temperature, and ecological momentary assessment (EMA) questionnaires. The solid lines represent the time of day at which a programmatic nudge was delivered by the watch for each of the respective measures. αmain effect of time of day (p < 0.05).

Full size image

Blood pressure

Participants demonstrated a high moderate compliance (90 ± 9%) for measuring blood pressure 3 times a day. On average, blood pressure measures were taken 29 ± 41 min, 60 ± 33 min, and 47 ± 42 min from programmatic nudges received at 08h30, 13h30, and 19h30, respectively (Fig. 3). There was no main effect of time of day (p = 0.68), or week (p = 0.64) for compliance with measuring blood pressure (Fig. 3). Missing data were attributed to either technical issues outside participant control (n = 118), personal reasons (n = 4), or unknown (n = 172).

Oral temperature

Participants demonstrated a high moderate compliance (89 ± 9%) for measuring oral temperature 3 times a day. Relative to the programmatic nudges at 08h30, 13h30, and 19h30, participants submitted oral temperature data within 35 ± 34 min, 66 ± 36 min, and 52 ± 51 min, respectively (Fig. 3). There was no main effect of time of day (p = 0.69), or week (p = 0.57) for compliance with measuring oral temperature over the observation period (Fig. 3). Missing data were attributed to either technical issues outside user control (n = 22), or unknown (n = 193).

Surveys

Participants demonstrated a low moderate compliance (83 ± 20%) for completing all custom EMA questionnaires requested 4 times per day. Participants completed the EMA questionnaires within 18 ± 25 min, 33 ± 16 min, 33 ± 21 min, and 0 ± 45 min of the programmatic nudges at 9h00, 12h00, 15h00, and 19h00, respectively (Fig. 3). A main effect of time of day was observed (p = 0.04), without a main effect of week (p = 0.22, Fig. 3). EMA questionnaires at 19h00 demonstrated a high moderate compliance (89 ± 4%) that was significantly greater than 15h00 (80 ± 2%, low moderate compliance, p = 0.01) but not 9h00 (80 ± 4%, low moderate compliance) or 12h00 (84 ± 3%, low moderate compliance). Missing data included technical issues (n = 11) and unknown (n = 324).

Discussion

Extreme heat events are expected to rise in frequency, duration, and intensity over the 21st century16, posing a potentially fatal health threat to all individuals, particularly at-home for the most at-risk17. We demonstrate the feasibility and capacity of deploying HeatSuite, a fully-data governed remote data collection platform, to aggregate and monitor pertinent patient vitals in some of the most at-risk populations. Further, HeatSuite passively monitors the end-user’s local environment, which may be necessary information for researchers and clinicians in the decision-making process for mitigating heat-related health risks and prescribing appropriate intervention strategies35,36. Herein, we contrast the current HeatSuite platform relative to pre-existing technology, detail a health and wellness check that was conducted on an 84-y old individual during an extreme heat event due to data aggregated by HeatSuite (see Fig. 4), and propose future applications for using the technology to capture in-situ physiological and behavioral responses.

Fig. 4: Example application of participant health real-time remote monitoring and personalized environmental exposure using HeatSuite.
figure 4

Indoor temperatures (lines) measured in the main room (light red) and bedroom (blue) of a 1-bedroom apartment (8th floor, malfunctioning air conditioner) during a 4-day extreme heat event (June 17–20th) and the 3 proceeding days (June 21–23rd), with 20-minute resolution presence detection of the 84-y old resident (solid circles, colour matched). Participant monitoring with HeatSuite revealed: i) ~48 h of exposure to indoor temperatures exceeding potentially hazardous threshold (>25.9 °C57); and ii) hypertension on wake up (161/91 at 8:02 AM), prompting a researcher (KL) to contact the participant to assess wellbeing at ~10:00 AM on June 20th, 2024.

Full size image

Other attempts have been made to create a multimodal sensor system to monitor patients, particularly elderly individuals, in their homes37,38. For example, SMARTA is a personal health monitoring system developed to permit home telemonitoring of health and interaction with their environment38. In initial pilot testing, they observed a 65-70% reliability rate in operation of their system by potential end-users, despite perfect user interaction, and the authors attributed these to technological issues with sensor integration. However, the exact technical error was not disclosed. In contrast, we observed a high moderate compliance (>85%) in users submitting valid data ignoring technical errors that were outside the users’ control. HeatSuite, much like any device or platform, is not immune to technological errors. Upon secondary analysis of our technical issues, we only observed them in the same 25% of all participants with blood pressure and oral temperature devices, and the issues were resolved with device replacement. Further, even including technical errors in our analysis retained compliance within a low moderate threshold (>80%), comparable to other at-home health monitoring technologies37. Future work is needed to determine the long-term compliance of using the HeatSuite platform.

It is certainly possible, as demonstrated by others37,39,40, to collect pertinent physiological, behavioural, and/or environmental data, within a person’s home using separate devices. For example, Tartarini et al.40 employed various offline and internet connected devices to monitor the personal comfort of 20 young healthy participants over a 6-month period in Singapore. However, the aggregation of data and time synchronization from the independent third-party platforms that originally stored the collected data likely requires technical expertise. This necessary final step for data analysis may be limiting to some, secondary to the requirement of sharing end-user data with device manufacturers who host and maintain the databases. HeatSuite simplifies this process by integrating all devices within one unified platform ensuring complete data-governance (see Methods and Fig. 1). Next, commercial wearable devices often require their own specific user account and/or head-unit (e.g., smartphone) for data synchronization. In contrast, once a smartwatch is affiliated with the primary ‘Group ID’ for the HeatSuite instance, it is capable of broadcasting and synchronizing its data with any environmental node in proximity attached to that Group ID, without sharing personally identifiable information. This enables monitoring multiple users simultaneously (e.g., a family), without additional devices (aside from smartwatches) or user accounts, alongside capacity to log user proximity to environmental nodes for more representative environmental exposure analysis. Furthermore, all secondary wearables currently integrated into HeatSuite (e.g., body mass, blood pressure, and oral temperature) interface directly and wirelessly with the smartwatch for measurement logging and synchronization with the environmental node occurs during charging. While this method for physiological and behavioural monitoring may limit true real-time responses of participants, particularly when away from their primary environmental nodes (see Fig. 1), local sensor derived data from the smartwatch such as heart rate, near wrist temperature, and movement trends, are transmitted wirelessly at regular intervals to environmental nodes within communication range (~10 m). Future upgrades to the HeatSuite platform can include high priority transmission of key health metrics aggregated from current or additional devices that may inform intervention decisions from monitoring personnel35,36. Figure 4 highlights how this was implemented during current development and deployment when an extreme heat event was observed, which is suspected to have resulted in 14 heat-related fatalities in Montreal41.

HeatSuite is focused on advancing our understanding of how extreme heat impacts the health and wellbeing of at-risk individuals in their natural environment. Future work includes the capacity of leveraging this technology to empirically evaluate the effectiveness of various heat management solutions in more ecologically valid settings to compliment lab-based evidence20,22,42. The negative consequences of extreme heat are not exclusive to a person’s residence. Rising temperatures due to climate change will negatively impact the workforce2, with greater risk of heat-related injuries among occupations requiring outdoor work and high levels of physical exertion, such as construction and agriculture industries43,44,45. HeatSuite can facilitate real-time worker monitoring and data collection efforts in these outdoor occupational sectors where internet connectivity or power access may be sparce or unavailable to characterize the multifaceted heat stress response, and evaluate the effectiveness of occupational interventions to reduce heat-related illness46. Lastly, ongoing and future work could support development of primary prevention strategies through early detection of physiological strain, or combination of factors, which could precipitate heat-related illness and provide the end-user personalized heat resilience strategies in advance of a period of extreme heat.

Hot days followed by hot nights have been shown to increase the risk of mortality among heat vulnerable individuals47, suggesting that overnight heat stress may increase physiological strain and compromise sleep quality, negatively impacting health and wellbeing. In this compliance study, 12/21 participants wore the smartwatch consistently overnight to permit continuous measurements of heart rate and calculating sleep indices using an open source algorithm48. Notable reasons for not wearing the watch included habitual removal of all watch and/or jewelry at bedtime, concerns of long-term exposure to wireless radio frequencies, and strap discomfort. Substituting the supplied watchband with a more comfortable strap may be a solution to increase overnight compliance to obtain more objective sleep and heart rate data. If problems with overnight wear persist, prospective users could quantify sleep duration by instructing participants to place the smartwatch on the charger immediately before bedtime and resume wearing on wake. This method has demonstrated reductions in sleep duration with increasing overnight temperatures49. The HeatSuite Cloud sleep analysis will automatically identify this extended charging time as the likely sleep period – although it will be appropriately flagged as a non-wearing time. Irrespective of objective sleep metrics, researchers and clinicians may include EMA question(s) about sleep quality to aggregate individualized subjective data to complement other physiological and/or environmental data collected.

Lastly, the Bangle.js2 smartwatch used in HeatSuite includes an onboard temperature and pressure sensor which records one-minute averages. While the housing of the Bangle.js2 is in direct contact with the wrist, the sensor is located below the display and on the side of the circuit board opposing the skin. As such, this sensor is not in direct contact with the skin and therefore is likely influenced by the surrounding environment40,50. Nonetheless, the sensor proximity to the skin in the Bangle.js2 is similar to that of the one-wire iButton Thermocrons which are commonly used to measure skin temperature51. Future work is needed to validate the agreement between the Bangle.js2’s onboard sensor and true wrist surface temperature under various ambient conditions.

In conclusion, HeatSuite demonstrated moderate to optimal compliance for all physiological parameters, and moderate compliance for EMA questionnaires among a cohort of older adults who presented with various health conditions. The programmatic nudging for scheduled tasks in combination with automatic data syncing may minimize researcher or clinician interaction ultimately reducing patient burden and enhance ecological validity of data. With its offline-first data collection approach, it has the capacity for long-term use in rural areas where internet connectivity may be limited, or to maintain data collection during power disruptions which are likely during natural disasters such as extreme heat events. Lastly, the continued expandability of the HeatSuite platform enables its growth with current and future wearable technology to address future lines of inquiry to better understand and monitor the physiological and behavioural responses of humans in their built environment.

Methods

Ethics and participant recruitment

The study protocol was approved by the Lakehead University Research Ethics Board prior to participant recruitment (#1469635). Verbal and written consent, and confirmation of eligibility to participate, was obtained from all participants in advance of enrollment in the study. With support from the Office Municipal d’Habitation de Montréal (OMHM), recruitment was done through word of mouth and snowball sampling amongst residents of low-income housing complexes. Eligible participants included any male or female above 50 years of age who resided in a single or double occupant apartment managed by the OMHM. Recruitment and deployment of HeatSuite kits was initiated at the end of May through early July 2024. The study follows the STROBE reporting guidelines.

Technical description of HeatSuite

The HeatSuite kit as deployed in this current project and how data are processed and managed are graphically depicted in Fig. 1. Each kit contained one main environmental node (custom), a satellite environmental node (custom), an open-source smartwatch (Bangle.js2), a blood pressure cuff (A&D Medical UA-651BLE), a YSI 400 Series thermistor probe (±0.1°C) and 12-bit Bluetooth capable dongle (custom), and a body composition scale (±50 grams, Mi Scale 2, Xiaomi). All custom hardware, firmware and/or software used by the HeatSuite system was written and developed by NR and are described below.

Each main environmental node was assembled in-house with 3D printed components (printed with UV resistant Acrylonitrile Styrene Acrylate thermoplastic, total print time of all components is <24 h), which contained a custom printed circuit board that interfaced the microcontroller with all environmental sensors and a small 30 mm fan for forced aspiration of the sensors. Onboard sensors monitored air temperature and relative humidity (Sensiron SHT85 [±0.1°C & ±1.5%RH] with SHT41 [±0.2°C & ±2.0%RH] as backup, Sensiron SEN54 [± 0.45°C & ±4.5%RH], and Bosch BME680 [±0.5°C & ±3%RH]), carbon dioxide levels (Winsen MH-Z1311A, ±30 ppm), particulate matter 2.5–10 ppm (Sensiron SEN54 [0–100 μg/m3: PM1 & PM2.5: ±5 μg/m3; PM4 & PM10 ± 25 μg/m3]), black globe temperature using a thermistor (Maxim Integrated DS18B20 [±0.5°C]) centered within a matte black 38 mm plastic sphere, and air velocity measured using an ultrasonic anemometer (HongYuv HY-WDC2E [± 0.1m/s]). All sensors interfaced with a third-party Espressif ESP32 microcontroller development board (LilyGo T-SIM7000G) which added additional onboard battery backup, expandable memory with a micro-SD card, cellular network connectivity, and global positioning system (GPS) capability. Data from all sensors were logged at one-minute intervals using custom firmware. Satellite environmental nodes monitored air temperature and humidity (Bosch BME680) and transmitted environmental data at 10-minute intervals to the main environmental node over wireless transmission from the onboard microcontroller (ESP32) with custom firmware. Both the main and satellite environmental nodes monitored participant proximity (e.g. presence detection) at predefined intervals of 5- and 10-minutes, respectively. If the main environmental node is connected to the internet (either by a Wi-Fi connection and/or cellular network), local data stored in memory is uploaded to a private server running the HeatSuite Cloud Software (explained below) over secure communication protocols hosted by the research team.

The Bangle.js2 is an open-source smartwatch with onboard 3-axis accelerometer (Kionix KX023), optical heart rate monitor (Vcare VC31B), and pressure & temperature sensor (Bosch BMP280 [±0.12 hPa & ±1.0°C]). Previous work has demonstrated that the onboard 3-axis accelerometer demonstrates good agreement with the ActiGraph GT9X52. The HeatSuite application was sideloaded onto each smartwatch using a locally hosted version of the Espruino App Loader53. The companion application allows the research team to preprogram scheduled study tasks and notifications (e.g., nudges), and upload custom ecological momentary assessment (EMA) questionnaires54, in multiple languages, for participants to complete by interacting with the smartwatch touchscreen with the results saved to the onboard memory. Total steps, heart rate, raw accelerometry data, and near wrist temperature are continuously recorded at one-minute intervals to onboard storage on the smartwatch. All wireless devices (e.g., blood pressure monitor, oral temperature monitor, body composition scale) transmit results over Bluetooth to the smartwatch for local storage of data. A chest-based heart rate monitor (e.g. Polar H10 or equivalent) may be connected to the smartwatch over Bluetooth to record min-averaged heart rate based on a single lead electrocardiogram, and log all R-R intervals for future analysis. When the smartwatch is charging, the main environmental node will download all data stored programmatically. When the smartwatch is out of range of any environmental node that is affiliated with the study, the onboard GPS receiver (AT6558) can be programmatically engaged at predefined intervals (researcher determined) to monitor user location for future analysis (e.g., thermal strain prediction using outdoor weather data).

The HeatSuite Cloud Software, automatically deployed on servers hosted and maintained by the research team using a single installation command, provides a web portal to monitor the status of all devices affiliated with the study (e.g., environmental nodes and smartwatches), and visualizes incoming data. It stores all incoming data from devices into a password protected My Structured Query Language (MySQL) database that monitoring personnel have direct access to and creates a raw copy of all ingested data in the event of database corruption. No personally identifiable information is stored on the server or MySQL database. The HeatSuite Cloud Software includes postprocessing of smartwatch data such as, but not limited to, total steps in 24 hours, average heart rate at rest and during movement, and sleep detection analysis using an open-source algorithm48. Lastly, the HeatSuite Cloud Software includes a graphical user interface for developing custom EMA surveys with options including i) time-dependent presentation, ii) once daily, and iii) randomizing order. Figure 5 illustrates the multiple physiological and perceptual endpoints integrated into the HeatSuite system at present and the typical raw and analyzed data provided by HeatSuite.

Fig. 5: Example 7-day physiological and perceptual dataset using HeatSuite.
figure 5

Overview of sample physiological and perceptual data collected from one participant (female, 60-y) over 7 days using the HeatSuite platform in the current deployment. Left most side depicts the data retrieved from the smartwatches onboard sensors each minute every day, with sleep detection conducted on the HeatSuite Cloud Software using an open-source algorithm48. Middle graphs demonstrate the data collected from external devices that synchronize with the watch. Right most table visually demonstrates the responses to ecological momentary assessment questions at each of the 4 time points, each day (note that white boxes represent no response received). BPM Beats per minute, Temp Temperature, MAP Mean Arterial Pressure.

Full size image

Study protocol

Following informed and signed consent, a member of the research team (KL or AM) visited each participant’s residence to set up the environmental nodes contained within the HeatSuite kit and provided a brief demonstration of how to independently complete each measurement and record results with the smartwatch. Participants were also left with a standard operating procedure manual developed by KL which visually describes how each task is completed. Following this meeting, participants were asked to wear the smartwatch continuously and to only remove during bathing or charging, and to conduct the following measurements, preprogrammed in the smartwatch application at predefined times, each day for 28 days: Body mass (08h00), blood pressure and oral temperature measures (08h30, 13h30, and 19h30), surveys (09h00, 12 h00, 15h00, and 19h00). Participants were also asked to monitor their micturition frequency as an index of hydration55 using a specific on-screen button on the smartwatch, however these data are not presented herein. Ongoing device deployment monitoring and technical troubleshooting was conducted by only one research team member at a given time (KL or AM).

Protocol compliance

The sample size was determined based on the logistical, financial and time constraints required to complete data collection. Compliance in using the HeatSuite platform was assessed by calculating the individual successful data reporting rates for each day. These rates were calculated for each unique parameter (body mass, blood pressure, oral temperature, and surveys) separately for each day. The average net difference in expected versus actual measurement reporting time was computed for each parameter. Compliance to wearing the smartwatch was calculated as the number of wear-minutes each day, and as the number of valid days over the 28-day observation, with a valid day defined by daytime (≥12h) and full-day (≥20h) of wear over a 24 h period. Daytime wear was included as discussions with prospective participants expressed concerns of potential discomfort wearing the smartwatch overnight. To further confirm a valid wear day, a lower bound movement threshold was implemented using the raw accelerometer data collected for each smartwatch during the charge state. Change in compliance for each of the specific parameters was conducted by binning compliance data in 4 blocks of 7-day intervals and analyzed using a one-way ANOVA (smartwatch wear and mass), or two-way ANOVA (blood pressure, oral temperature, surveys) with the repeated factor of time of day (minimum 3 levels). Missing data was categorized as i) technical issues out of participants’ control, ii) personal reasons noted to researchers, or iii) unknown. Missing data due to technical errors outside of participants’ control was excluded from compliance analysis. Compliance was defined using a modified version of previously reported thresholds in clinical populations56 with either optimal (>95%), high moderate (>85–95%), low moderate (>75–85%), and low (<75%).