Data were prospectively collected from 30 volunteers recruited through advertisements at the University of Warwick. Data were recorded for a 24-hour free-living period of continuous monitoring, including electrocardiogram (ECG) (from a wearable sensor), interstitial glucose (from a clinically used device), and ambulatory blood pressure (from a clinically used device). To allow evaluation of the relationship between ECG morphology and other parameters, ECG morphology features were generated utilising a published deep learning-based ECG segmentor tool.

Participants

A target of 30 participants was set for this study, and 30 were recruited. Pregnant individuals were the only group excluded from participation. On enrollment, participants’ information was collected, including age, gender, ethnicity, height, weight, general health status, use of medications, and frequency of physical activity. Participants were asked to complete a daily food and activity diary, as well as the Consensus Sleep Diary (CSD) ¹. The CSD is a standardised assessment developed by expert consensus and tested in patient-focused groups, which allows the collection of self-reported information, including sleep duration, disturbance, and timing.

Ethical approval for the study was obtained from the Biomedical and Scientific Research Ethics Committee of the University of Warwick (Ref: BSREC 53/21-22 AM01). All participants were provided with comprehensive study details and instructions and gave written informed consent prior to participation.

Protocol

Participants attended an initial session, where they received a briefing on the study protocol and filled in a questionnaire for demographic and physical health information. Participants were then fitted with wearable sensors (detailed below) and subsequently underwent a 24-hour period of monitoring. Participants wore the sensors from the initial briefing, overnight and returned the devices and diaries the following day. Participants were asked ahead of the initial session to avoid bathing, showering, or swimming and to not remove the devices unless due to irritation/discomfort or a wish to end participation in the study. Otherwise, participants were encouraged to maintain normal behaviour during the monitoring time, including sleep schedule and other habits.

Monitoring devices

Three monitors were used, recording continuous ECG, ambulatory blood pressure, and interstitial glucose.

ECG was recorded using the CE marked device, Zephyr BioHarness™ 3.0 (Medtronic, Inc., Annapolis, MD, USA), that has been previously validated ^2,3 and used in several previous scientific studies ^4-6. This is a chest-worn wearable sensor that records ECG across a single lead, with amplitude range between 0.25 and 15 mV, and sampling frequency 250 Hz. The sensor was fitted either using electrodes or chest strap provided. The Zephyr BioHarness also contains a tri-axial trunk accelerometer (sampling rate of 100Hz), and this is used to compute, via proprietary methods, parameters for the activity level and posture of the wearer. Activity was recorded, calculated in units of g using averages of the three axial acceleration magnitudes (x, y, z).

As such, activity levels of approximately 0.2 are generated during walking, and 0.8 during jogging ⁷. Posture was calculated from the angle of deviation from the vertical axis (reported in degrees). Both activity and posture are automatically calculated by the device and reported at one sample per second. All ECG and other metrics were stored internally on the device and subsequently downloaded for processing.

Ambulatory Blood Pressure Monitoring (ABPM) is the gold standard for measuring BP over a 24-hour period. The SpaceLabs OnTrak Ambulatory Blood Pressure monitor was used, which has been comprehensively tested and validated – passing three recognized international protocols (AAMI/ANSI/ISO 81060, British Hypertension Society, European Hypertension Society ⁸) and used in over 500 clinical trials⁹. The monitor was placed on the non-dominant arm of the participant and set to automatically take readings at half hour intervals during the day (7:00-22:00) and hourly intervals during the night (22:00 – 7:00). Two cuff sizes were available for this study (medium and large), selected based on the manufacturer guidelines for upper arm circumference of the participant. The monitor records blood pressure (BP) readings of systolic BP (SBP), diastolic BP (DBP), partial pressure (PP = SP – DP), and mean arterial pressure (MAP = DBP + 1/3(SBP – DBP)).

The Abbott FreeStyle Libre 2 was used to measure glucose in the interstitial fluid, reporting measurements every 15 minutes. These sensors were placed on the back of the upper arm (dominant arm).

ECG segmentation was performed to extract individual ECG beats with key fiducial points labelled, and to calculate a comprehensive set of time-based ECG features. A deep learning-based ECG segmentor tool, developed by researchers in the Applied Biomedical Signal Processing and Intelligent eHealth Lab of the University of Warwick ¹⁰, was used to process the ECG signals. This involves filtering the signal followed by identification of ECG fiducial points (P, QRS, and T), as shown in Figure 3A.

This tool was based on Attention based Conv-BiLSTM network which was trained on PhysioNet’s QT database ¹¹ via three block procedure.

Secondly, the ECG beat morphology extractor was deployed, which detected the fiducial points (end and peak points of the waves present in the ECG beat, Figure 3A). The morphological extractor detected five fiducial points namely P, Q, R, S and T. The calculation of ECG beat morphology was based on calculation of i) intervals and ii) slopes among all the five fiducial points.

1.             Carney CE, Buysse DJ, Ancoli-Israel S, et al. The Consensus Sleep Diary: Standardizing Prospective Sleep Self-Monitoring. *Sleep. *2012;35(2):287-302. DOI: 10.5665/sleep.1642

2.             Johnstone JA, Ford PA, Hughes G, Watson T, Garrett AT. Bioharness(™) multivariable monitoring device: part. I: validity. *J Sports Sci Med. *2012;11(3):400-408. DOI. Published 2012/01/01.

3.             Johnstone JA, Ford PA, Hughes G, Watson T, Garrett AT. Bioharness(™) Multivariable Monitoring Device: Part. II: Reliability. *J Sports Sci Med. *2012;11(3):409-417. DOI. Published 2012/01/01.

4.             Montesinos L, Castaldo R, Cappuccio FP, Pecchia L. Day-to-day variations in sleep quality affect standing balance in healthy adults. *Scientific Reports. *2018;8(1):17504. DOI: 10.1038/s41598-018-36053-4

5.             Castaldo R, Chappell MJ, Byrne H, et al. Detection of melatonin-onset in real settings via wearable sensors and artificial intelligence. A pilot study. *Biomedical Signal Processing and Control. *2021;65:102386. DOI: https://doi.org/10.1016/j.bspc.2020.102386

6.             Simjanoska M, Gjoreski M, Gams M, Madevska Bogdanova A. Non-Invasive Blood Pressure Estimation from ECG Using Machine Learning Techniques. *Sensors (Basel). *2018;18(4). DOI: 10.3390/s18041160

7.             Zephyr Technology: Bioharness log data descriptions. https://www.zephyranywhere.com/media/download/bioharness-log-data-descriptions-07-apr-2016.pdf. Accessed 18 Jan, 2024.

8.             de Greeff A, Shennan AH. Validation of the Spacelabs 90227 OnTrak device according to the European and British Hypertension Societies as well as the American protocols. *Blood Press Monit. *2020;25(2):110-114. DOI: 10.1097/mbp.0000000000000424

9.             ABP Monitoring OnTrak. https://spacelabshealthcare.com/wp-content/uploads/2023/08/030-2193-00-Rev-B-Eng-OnTrak.pdf. Accessed 18 Jan, 2024.

10.          Haleem MS, Pecchia L. A Deep Learning Based ECG Segmentation Tool for Detection of ECG Beat Parameters. Paper presented at: 2022 IEEE Symposium on Computers and Communications (ISCC); 30 June-3 July 2022.

11.          Laguna P, Mark RG, Goldberg A, Moody GB. A database for evaluation of algorithms for measurement of QT and other waveform intervals in the ECG. Paper presented at: Computers in Cardiology 1997; 7-10 Sept. 1997.