Machine learning predicts peak oxygen uptake and peak power output for customizing cardiopulmonary exercise testing...

Area of Science:

• Exercise Physiology
• Machine Learning in Sports Science
• Biomedical Data Analysis

Background:

• Cardiopulmonary exercise testing (CPET) is the gold standard for assessing cardiorespiratory fitness.
• Individualized CPET protocols are crucial for accurate assessment.
• Current methods for protocol individualization require further optimization.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting peak oxygen uptake ($$ O_2peak) and peak power output (PPO).
• To individualize CPET ramp protocols using non-exercise features.
• To compare the predictive accuracy of ML models against multiple linear regression (MLR).

Main Methods:

• A cross-sectional study involving 274 participants undergoing CPET on a cycle ergometer.
• Application of various ML models (e.g., random forest, gradient boosting) and MLR to predict $$ O_2peak and PPO.
• Utilized Shapley additive explanation (SHAP) to identify key predictive features.

Main Results:

• Random forest and gradient boosting models demonstrated superior accuracy in predicting $$ O_2peak and PPO, respectively.
• ML models reduced root mean square error (RMSE) by up to 28% for $$ O_2peak and 22% for PPO compared to MLR.
• Body composition features, including skeletal muscle mass and extracellular water, were identified as the most impactful predictors.

Conclusions:

• Machine learning models offer a more accurate approach to predicting $$ O_2peak and PPO than traditional MLR.
• These ML models can effectively individualize CPET ramp protocols.
• Body composition data significantly enhances the accuracy of predicting CPET outcomes.