Development, validation, and transportability of several machine-learned, non-exercise-based VO_2max prediction models...

Area of Science:

• Gerontology
• Cardiorespiratory Fitness
• Machine Learning Applications

Background:

• Maximal oxygen uptake (VO2max) measurement is often impractical in large-scale studies.
• Existing non-exercise prediction equations for VO2max are limited, especially for older adults.
• Machine learning (ML) offers potential for improved VO2max prediction.

Purpose of the Study:

• To develop and validate ML algorithms for predicting VO2max in older adults.
• To compare the performance of different ML models.
• To assess the transportability of these algorithms and their association with mortality.

Main Methods:

• Utilized data from the Baltimore Longitudinal Study of Aging (BLSA) with 1080 participants.
• Trained various ML algorithms including LASSO, XGBoost, Random Forest, and SVM.
• Developed models using all available variables and a subset common in aging cohorts.

Main Results:

• ML models achieved low root mean squared errors, indicating good prediction accuracy (e.g., LASSO and XGBoost at 3.4 mL/kg/min).
• Measured VO2max demonstrated a clear inverse relationship with mortality risk.
• Predicted VO2max showed similar trends with mortality but lacked robustness upon adjustment.

Conclusions:

• ML enhances VO2max prediction accuracy compared to simpler methods.
• Measured VO2max is a significant predictor of mortality in older adults.
• Further research is needed to validate ML-based VO2max predictions for promoting healthy aging.