Calculation of approximate heart rate variability indicators based on low-resolution heart rate data provided by widely used commercially available wearable devices

Abstract

Heart rate variability (HRV) assessment using wearable technology is a valuable tool for monitoring physical and emotional health. However, many widely used wearable devices, such as those from Apple and Fitbit, do not provide high-resolution heart rate (HR) data (i.e., data for every heartbeat) but instead report low-resolution data, typically as average HR values over fixed intervals (e.g., every 5 s). In this study, we developed algorithms to estimate HRV indicators from such low-resolution HR data and evaluated their reliability and accuracy. High-resolution HR data were collected over one week from 154 pregnant women (aged 25–44 years, 23–32 weeks gestation) using a chest-worn portable HR monitor. The average HR over each 5-second interval was calculated to match Fitbit’s data format. HRV indicators were computed from the reconstructed low-resolution data and compared with those from the original high-resolution data using two one-sided tests of equivalence (TOST), correlation analysis, and principal component analysis (PCA). Additional validation using Bland–Altman plots and bootstrap-derived confidence intervals assessed estimation stability. All analyses indicated high similarity between estimated and reference HRV values. TOST confirmed statistical equivalence (p < 0.05) with negligible effect sizes (Cohen’s d < 0.1). Correlation coefficients ranged from 0.714 to 0.921, and PCA yielded a similarity index of 0.95. The algorithms demonstrated robustness through equivalence testing, distributional similarity, error stability, and cross-cohort generalizability. Further validation using both high- and low-resolution HR datasets from publicly available databases supported these findings. These results suggest that HRV indicators derived from low-resolution HR data may be sufficiently accurate for clinical and everyday health monitoring.