Estimation of cardiorespiratory fitness using heart rate and step count data

Predicting cardiorespiratory fitness levels can be useful for measuring progress in an exercise program as well as for stratifying cardiovascular risk in asymptomatic adults. This study proposes a model to predict fitness level in terms of maximal oxygen uptake using anthropometric, heart rate, and step count data. The model was trained on a diverse cohort of 3115 healthy subjects (1035 women and 2080 men) aged 42 +/- 10.6 years and tested on a cohort of 779 healthy subjects (260 women and 519 men) aged 42 +/- 10.18 years. The developed model is capable of making accurate and reliable predictions with the average test set error of 3.946 ml/kg/min. The maximal oxygen uptake labels were obtained using wearable devices (Apple Watch and Garmin) during recorded workout sessions. Additionally, the model was validated on a sample of 10 subjects with maximal oxygen uptake determined directly using a treadmill protocol in a laboratory setting and showed an error of 4.982 ml/kg/min. Unlike most other models, which use accelerometer readings as additional input data, the proposed model relies solely on heart rate and step counts-data readily available on the majority of fitness trackers. The proposed model provides a point estimation and a probabilistic prediction of cardiorespiratory fitness level, thus it can estimate the prediction's uncertainty and construct confidence intervals.

Automated summary

This study proposes a model for predicting cardiorespiratory fitness levels using anthropometric, heart rate, and step count data. The model shows accurate predictions on a diverse cohort and provides estimations of uncertainty for the predictions.