0312 The HRV Of The Ring - Comparison Of Nocturnal HR And HRV Between A Commercially Available Wearable Ring And ECG

Abstract

Results: The convolutional neural network classifier for apnea events demonstrated an AUC of 0.93 and accuracy of 97%. The convolutional neural network classifier for hypopnea events demonstrated an AUC of 0.80 and accuracy of 80%. The mean squared error for the model's predicted AHI for the patients in the hold-out set was 3.541. Conclusion: The deep learning sleep classifiers for apneas and hypo-pneas demonstrated excellent accuracy for apnea detection and fair accuracy for hypopnea detection. They achieve comparable results to studies performed in literature for automated sleep event detection with models that do not require hand-engineered features. Introduction: The beat of a healthy heart shows significant variation among the time intervals between heartbeats. Heart rate variability (HRV) consists of periodic and aperiodic changes in the duration of the cardiac "QRS-wave" cycle. Greater nocturnal HRV has been linked with better sleep quality in healthy and clinical populations. Measurement of HRV during sleep might therefore have applications for sleep related maladies, but to collect long term data there is a strong need for comfortable measurement devices that do not disturb rest. Here we assess the Oura ring's (Ring) ability to fill this need. Methods: We measured nocturnal photopletysmogram (PPG) based inter-beat interval (IBI) data with the Ring (Oura Health Ltd, Oulu, Finland) and simultaneous R-R interval data with Faros 360 elec-trocardiogram (ECG) device (Mega Electronics, Kuopio, Finland) in 10 healthy individuals (3 female, 7 male). All subjects had the Ring on both hands, resulting in 20 nightly recordings for analysis. Root mean square of successive differences (rMSSD) was used as the HRV measurement. We determined heart rate HR and HRV as nightly averages for the Ring using only the normal IBI values (HR Ring and rMSSD Ring) and for ECG using Kubios software with automatic filter having medium setting (HR ECG and rMSSD ECG). The agreement between the methods was assessed by correlation analysis using Matlab software. Results: High correlation was observed between HR Ring and HR ECG (r 2 = .998) with a bias of-0.53 bpm (p 0.5 and 0.8, respectively), accurately capturing within-and between-subject variations in performance across the 62 h of TSD. Sleep loss had a large effect (Cohen's d > 1.0) on all 2B-Alert PVT statistics. A comparison of the 48-h-ahead real-time predictions against the measured mean RT data revealed that the app's prediction accuracy increased with the number of PVT measurements available for tool customization. The app learned each individual's sleep-loss phenotype within 12 PVT measurements over the first 36 h of TSD, yielding an average error of less than 10 ms when compared to the UMP customized using all 20 PVT measurements. Conclusion: The 2B-Alert App offers a practical means for personal fatigue management by allowing for real-time individualized performance assessment and prediction in a smartphone.