AI Predicts VO2 max from Exercise-Free Heart Recordings

Cardiovascular Deconditioning Challenge

Cardiovascular fitness levels are widely known to improve through training at increasing aerobic volume, high intensity interval training (HIIT) and threshold exercise. Vice versa, a decline can occur when the body experiences prolonged reduction in physical activity or unloading, which includes periods of rehabilitation or sedentary lifestyle. The most striking manifestation of cardiovascular deconditioning appears in measurable reductions of VO₂ max—the gold standard metric for aerobic fitness that represents the maximum amount of oxygen your body can utilize during intense exercise. This parameter serves as a crucial indicator of cardiovascular health and exercise capacity, making its decline a serious concern.

Interestingly, cardiovascular deconditioning is not only a concern for humans on Earth, but also for austronauts on space flights. Cardiovascular deconditioning occurs rapidly in microgravity environments, where the absence of gravitational stress fundamentally alters how the heart and blood vessels function. Without the constant downward pull of gravity, blood distribution shifts dramatically throughout the body, reducing the workload on the cardiovascular system and triggering a cascade of adaptive responses that ultimately compromise fitness levels. Previous shuttle and Spacelab missions documented VO₂ max reductions ranging from 10.4% to 22% during 9-17 day flights [2]. However, while measuring VO₂ max on earth is typically done though exercise at varying intensity using a treadmill or cycle ergometer, in space this is a particularly difficult and expensive procedure. Solbiati et al. have therefore developed a novel approach, using artificial intelligence (AI) to predict VO₂ max from a 24-hour recording of a small, portable Holter electrocardiogram (ECG) [1].

AI Prediction Methodology

Solbiati et al. used data from six comprehensive bed rest campaigns conducted by the European Space Agency (ESA) between 2010 and 2018, that recruited 148 healthy male volunteers aged 20-45 years. These participants were exposed to different durations of bed rest, ranging from 5 to 60 days, and underwent rigorous VO₂ max testing using incremental cycle ergometer protocols both before and after. Simultaneously, researchers collected continuous 24-hour recordings with a Holter ECG, capturing detailed cardiac electrical activity patterns. This bed rest analog proved remarkably effective at reproducing space-induced cardiovascular changes. The research team then utilized sophisticated machine learning algorithms to predict cardiovascular fitness decline from these continuous heart monitoring data collected during simulated space conditions.

Five distinct regression models we evaluated to identify the most accurate predictor of VO₂ max changes. After comprehensive testing, the AdaBoost Regressor (ABR) emerged as the superior performer, demonstrating the lowest prediction errors across multiple metrics. This ensemble method combines multiple "weak" learners to create a robust predictive model, achieving remarkable accuracy in forecasting aerobic capacity decline: a median Root Mean Squared Error (RMSE) of just 4.87 mL/kg/min. This is notably more accurate than consumer wearables that rely on resting heart rate variability (HRV) features, having errors up to 30.5 mL/kg/min [3].

The machine learning model achieved particularly impressive results for shorter duration studies. In 5-day bed rest campaigns, prediction accuracy improved significantly with RMSE values reaching down to 3.34-4.23 mL/kg/min. The prediction accuracy was found to worsen with longer bed rest durations: 5.37-7.57 for 60-day bed rest campaigns. This pattern likely reflects the complex physiological adaptations occurring during extended periods of cardiovascular deconditioning, where multiple systems beyond cardiac electrical activity contribute to fitness decline.

Healthcare Implementation Potential

The impressive findings from this AI-powered cardiovascular monitoring research open exciting possibilities for both space exploration and terrestrial medicine. For space applications, this technology could revolutionize the current Fitness Evaluation (PFE) protocol used on the International Space Station. Currently, astronauts must perform time-consuming submaximal cycling tests every 30 days, which consume valuable crew time and resources. The AI model's high precision in predicting VO₂ max suggests that astronauts showing favorable predictions could skip routine testing entirely, while those flagged for potential decline would receive targeted assessment.

The technology's ability to work with standard 24-hour Holter ECG recordings makes implementation feasible across diverse medical settings. Future developments integrating this AI approach with wearable ECG devices could enable real-time fitness monitoring, transforming preventive cardiology and personalized medicine while ensuring astronaut safety during humanity's expansion into deep space.

References

1. Solbiati S, Fiorentino MC, Bendandi R, Moccia S, Caiani EG. "AI-based prediction of VO2 max from 24-h Holter ECG recording", NPJ Microgravity 11 (2025) 89. https://doi.org/10.1038/s41526-025-00542-4
2. Trappe T, Trappe S, Lee G, Widrick J, Fitts R, Costill D. "Cardiorespiratory responses to physical work during and following 17 days of bed rest and spaceflight", Journal of Applied Physiology 100 (2006) 951-957. https://doi.org/10.1152/japplphysiol.01083.2005
3. Molina-Garcia, P. et al. "Validity of Estimating the Maximal Oxygen Consumption by Consumer Wearables: A Systematic Review with Meta-analysis and Expert Statement of the INTERLIVE Network", Sports Med 52 (2022) 1577–1597. https://doi.org/10.1007/s40279-021-01639-y