Metabolic and Fitness Correlates of Heart Rate Variability in Middle-Aged Endurance Athletes: A Pilot Study Using Continuous Glucose Monitoring and Machine Learning

Background: HRV is a key biomarker of autonomic nervous system function and cardiovascular health. Emerging evidence suggests that metabolic regulation, particularly glucose variability, may influence autonomic balance even in non-diabetic populations. With the increasing availability of CGM and wearable sensors, the integration of metabolic and cardiovascular signals enables novel data-driven approaches for personalized health monitoring. Objectives: This pilot study aimed to investigate the associations between glucose dynamics, fitness indicators, and HRV in middle-aged amateur endurance athletes, and to evaluate the feasibility of predicting HRV using machine-learning models based on CGM and wearable-derived features. Methods: Ten male endurance athletes (age 39–50 years) were monitored over a two-month period using CGM devices and advanced smartwatches. Daily metrics included average glucose, glucose standard deviation and coefficient of variation, resting heart rate (RHR), nocturnal HRV (RMSSD), blood oxygen saturation (SpO₂), and estimated VO₂max. Anthropometric and biochemical markers (BMI, fat mass, skinfolds, HbA1c, ApoB, vitamin D) were also collected. Random Forest, XGBoost, and LightGBM regression models were trained to predict HRV. Model performance was evaluated using cross-validated R² (R² ranged from 0.12 to 0.28) and normalized mean absolute error (nMAE). Results: Correlation analysis revealed that elevated ApoB, increased fat mass, and higher RHR were strongly associated with lower HRV. Glucose variability showed weaker associations in this cohort. Machine-learning models demonstrated limited predictive accuracy for HRV (R² < 0.30 across models), suggesting that while physiological links exist, the selected features alone are insufficient to fully explain HRV variability in a pilot cohort. Conclusions: In middle-aged endurance athletes, glucose regulation and cardiovascular fitness markers show meaningful associations with autonomic function, but the feasibility of accurately predicting HRV from CGM-derived metrics remains limited in this pilot dataset. These findings support the physiological link between metabolic stability and autonomic balance, while highlighting the need for larger longitudinal datasets and time-series modeling to capture the complex dynamics between glucose fluctuations and cardiac autonomic regulation.