600 meters to VO2max: Predicting Cardiorespiratory Fitness with an Uphill Run

Abstract

Introduction Cardiorespiratory fitness (CRF) is used to assess the body's ability to perform daily activities and is a criterion of overall health besides endurance performance. There is an inverse relationship between CRF and heart diseases or mortality due to any potential cause in healthy individuals [1]. Therefore, an estimation of CRF can yield relevant information for healthy people to assess their risk of developing coronary diseases, as well as for athletes to design tailored training programs to optimize their performance. The gold standard for assessing CRF is a direct measurement of maximal oxygen uptake (VO2max) in laboratory settings, which quantifies oxygen consumption mostly during graded maximal exercise tests [2]. In spite of the high accuracy provided by direct measurements, they are time-consuming, costly, and require expertise. At the same time, multiple indirect methods have been developed to estimate VO2max, most of which are field tests. While these tests are more accessible, they often still require technical knowledge or specialized equipment, as most of them are designed primarily for athletes and administered by experts. Therefore, performing these tests and interpreting the results can be challenging for recreational individuals. As technology has been advancing, wearable devices are increasingly used for physical activity monitoring with proper accuracy [3, 4]. This advancement further provided access to data for specialists, enabling the development and comparison of field tests and predictive models. Subsequently, various studies have shown that deep learning models outperform linear regression models in predicting VO2max values [5]. Previous studies also showed that, to accurately predict VO2max in a laboratory or through indirect tests, individuals need to reach their maximal effort rather than performing submaximal exercises [6]. In this study, we present an approach to predict VO2max from data collected during a maximal effort test on a 600-meter uphill course, using both linear regression and machine learning models.

Methods The subjects consisted of 10 men and 8 women between the ages of 20 and 39 years. All participants provided informed consent before participating in the study and they were given detailed information about the objectives of the study, procedures, and potential risks. Then, a maximal incremental treadmill test was completed by the 18 participants to assess VO2max values. The treadmill test began at an initial speed of 7 km·h-1 with a 7% incline and the speed was increased by 0.5 km·h-1 every 30 seconds until volitional exhaustion [7]. Gas analysis was recorded using the mixing chamber module of the Quark cardiopulmonary exercise testing system (COSMED Srl, Rome, Italy). After a rest period of 3 to 7 days, participants performed an all-out test on a 600-meter uphill track with a mean gradient of 9.7%. Data were collected during this field test with participants wearing a Polar Vantage V2 watch and a Polar H10 chest belt (Polar Electro Oy, Finland). In addition to the user’s anthropometric data (age, gender, body mass index) and time, we also used the speed to heart rate (HR) ratio as a predictor variable, since a previous study has shown its importance [8]. Multiple linear regression (MLR), XGboost, and a multilayer perceptron (MLP) were used to develop models to predict VO2max. The data set was randomly split into 80% and 20% as training and test set, respectively. For the purpose of overcoming multicollinearity among the predictor variables speed to HR ratio, time, and gender, principal component analysis with two components was applied before we fed the data into the multiple linear regression model.

Results The mean measured VO2max scores were 44.1 ± 4.7 mL·kg-1·min-1 and 55.8 ± 5.8 mL·kg-1·min-1 for women and men, respectively. The total distance was covered in an average time of 3:08 ± 0:41 minutes, with an average speed of 11.7 ± 2.5 km·h-1. Regarding the performance of VO2max estimation, the MLR model achieved an R2 of 0.77 with a standard error of the estimate (SEE) of 3.4 mL·kg-1·min-1, the XGBoost model an R2 of 0.82 with a SEE of 3.0 mL·kg-1·min-1, and the MLP model an R2 of 0.94 with a SEE of 1.6 mL·kg-1·min-1.

Discussion/Conclusion These results suggest that our short, high-intensity field test, when combined with a neural network model, can provide accurate predictions of VO2max. This method can be applied by healthy individuals without requiring assistance from an expert since it is simple and inexpensive. Therefore, our approach to estimate CRF level may be adopted by a wide range of people. Within our data, we have uncovered a non-linear, complex relationship between VO2max and the predictor variables that a feed-forward neural network with one hidden layer can reliably approximate. This is consistent with the findings of the previous research that emphasizes the power of deep learning models to predict VO2max accurately [5]. One potential limitation of this study is the size of the data, for having more data could unveil other relationships between VO2max and the predictor variables. Furthermore, the performance of the models may also suffer from overfitting due to the small and homogeneous sample population and potential discrepancies between the distributions of the training and the test data. Thus, future work should include increasing the sample size and exploring probabilistic models for VO2max prediction.

References

Kodama, S., Saito, K., Tanaka, S., Maki, M., Yachi, Y., Asumi, M., ... & Sone, H. (2009). Cardiorespiratory fitness as a quantitative predictor of all-cause mortality and cardiovascular events in healthy men and women: A meta-analysis. JAMA, 301(19), 2024–2035.

Beltz, N. M., Gibson, A. L., Janot, J. M., Kravitz, L., Mermier, C. M., & Dalleck, L. C. (2016). Graded exercise testing protocols for the determination of VO₂max: Historical perspectives, progress, and future considerations. Journal of Sports Medicine, 2016(1), 3968393.

Tang, M. S. S., Moore, K., McGavigan, A., Clark, R. A., & Ganesan, A. N. (2020). Effectiveness of wearable trackers on physical activity in healthy adults: Systematic review and meta-analysis of randomized controlled trials. JMIR mHealth and uHealth, 8(7), e15576.

Nuuttila, O. P., Korhonen, E., Laukkanen, J., & Kyröläinen, H. (2021). Validity of the wrist-worn polar vantage V2 to measure heart rate and heart rate variability at rest. Sensors, 22(1), 137.

Ashfaq, A., Cronin, N., & Müller, P. (2022). Recent advances in machine learning for maximal oxygen uptake (VO₂ max) prediction: A review. Informatics in Medicine Unlocked, 28, 100863.

Buckley, D. J., & Rowe Jr, J. R. (2018). Actual versus predicted VO₂max: A comparison of 4 different methods. International Journal of Exercise Science: Conference Proceedings, 2(10), 41.

Maier, T., Gross, M., Trösch, S., Steiner, T., Müller, B., Bourban, P., ... & Seidel, R. (2016). Manual Leistungsdiagnostik. Swiss Olympic.

Altini, M., Van Hoof, C., & Amft, O. (2017, February). Relation between estimated cardiorespiratory fitness and running performance in free-living: An analysis of HRV4Training data. In 2017 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI) (pp. 249–252). IEEE.