Remote prediction of cardiorespiratory fitness in a preoperative cohort: exploring short and long-term heart rate variability

Background Wearable sensors offer a scalable alternative to cardiopulmonary exercise testing for assessing cardiorespiratory fitness, and there is growing evidence to support their use for remote VO2max estimation. This study investigated whether heart rate variability (HRV) measures derived from wearable ECG sensors improve VO2max estimations in a preoperative cohort and compared the relative contributions of short- and long-term HRV features. ECG and accelerometer data from 198 participants scheduled for major abdominal surgery (REMOTES study, ClinicalTrials.gov: ID NCT06042023) were collected over 72 h. Measures including physical activity, steps, heart rate, and HRV were extracted. Short-term (5-minutes) and long-term (24-hour) heart rate variability features were extracted from free-living ECG data. Two LASSO regression models with five-fold cross-validation were developed: a baseline model (excluding HRV) and a HRV model.

Results After exclusions, 163 participants were included in analyses. The HRV model outperformed the baseline across all metrics, achieving a higher R2 (0.47 ± 0.12 vs. 0.42 ± 0.13) and lower mean absolute error (2.63 ± 0.34 vs. 2.77 ± 0.38 ml/kg/min), root mean square error (3.38 ± 0.53 vs. 3.54 ± 0.57 ml/kg/min) and absolute percentage error (15.55 ± 2.19% vs. 16.22 ± 2.45%). Analysis of feature contributions identified long-term HRV (SDANNHR24), age, gender, and step-counts as key contributors to model performance.

Conclusion HRV features from wearable data, especially long-term measures, can improve remote VO2max predictions in a clinical cohort. While performance gains were small, these findings support the integration of HRV features into remote monitoring systems in real-world settings. Long-term HRV measures derived from heart rate signals offer a practical option for cardiorespiratory fitness assessment, requiring minimal additional processing.

Trail registration This study was registered at ClinicalTrials.gov (Clinical trial number: NCT06042023) and was registered retrospectively on 11/09/2023.