Author(s): SABIONI, M., WILLÉN, J., JAKOBSSON, M., DUAL, S., Institution: KTH ROYAL INSTITUTE OF TECHNOLOGY, Country: SWEDEN, Abstract-ID: 2448

INTRODUCTION:
The quality of RR-intervals (RR) and heart rate (HR) measured by wearable heart rate monitors has been assessed extensively [1-3]; however, most of the validation protocols include only long steady-state acquisition periods, neglecting the dynamic responses and delaying effects of filters applied by manufacturers. While irrelevant in endurance sports, in applications such as high-intensity interval training, those characteristics become important; still, they are largely undocumented. Therefore, this study aims to quantify, evaluate, and compare the dynamic response of RR and HR measurements of commercially available chest-worn wearable monitors during induced sharp changes in HR.
METHODS:
A cheap, simple, and highly reproducible strategy was adopted, where a waveform generator was used to create ECG signals simulating the heart activity. RR and HR were recorded using the standard Bluetooth heart rate service for four ECG-based wearable monitors: Garmin HRM-Dual (G), Movesense Active (M), Polar H10 (P), and Wahoo TICKR (W). To simulate sharp changes in HR, four step functions were used (60-120 bpm, 120-60 bpm, 120-180 bpm, and 180-120 bpm), where each test was repeated ten times for each device. Dynamic response was quantified by resampling the signals to 1 Hz and time-aligning to the start of the test. Evaluation and comparison were based on latency, computed as RR latency (time elapsed from the step signal until the sensor RR response) and HR latency (time elapsed from the change in RR until the HR response was within ±3 bpm of the reference).
RESULTS:
The RR measurements of all devices responded nearly immediately to changes on the reference signal. RR latency was 1.7±0.2 s (G), 1.6±0.2 s (M), 3.2±0.7 s (P), 2.0±0.5 s (W) (mean±SD) across all step tests. Mean absolute error of RR measurements pre and post step (constant signal) was below 3 ms for all but one device (W: 21 ms). HR response was significantly delayed, and latency was different between devices and step tests. The longest HR latency was observed on the 120-60 bpm test: 38.9±1.2 s (G), 24.0±0.0 s (M), 23.0±0.3 s (P), 24.3±5.6 s (W). The shortest HR latency was observed on the 120-180 bpm test: 12.6±0.3 s (G), 8.3±0.1 s (M), 4.7±0.7 s (P), 5.2±0.8 s (W).
CONCLUSION:
HR measurements were significantly different between the four devices, where all presented some level of latency, indicating that manufacturers implemented different digital filters and thresholds to compute the HR values. Such filtering strategies had great impact on the dynamic response of the sensors, suggesting that those characteristics could be relevant in applications where sharp changes in HR are present. Open documentation of the processing steps is necessary, and future research involving sharp HR changes should be based on RR measurements rather than HR measurements.
1. Schaffarczyk et al. (2022) Sensors
2. Gilgen-Ammann et al. (2019) European Journal of Applied Physiology
3. Rogers et al. (2022) Sensors