Improving age prediction using ECG signals: Insights into lifestyle impacts

The discrepancy between cardiac age, estimated from electrocardiograms (ECGs), and chronological age can serve as an indicator of cardiovascular health. A significant difference may reflect vascular aging and elevated risk, but improving the precision and interpretability of these predictions remains challenging.

This study proposes novel preprocessing techniques to improve prediction accuracy and investigates the relationship between lifestyle factors and cardiac aging. ECG signals were processed using a Butterworth bandpass filter combined with a Hyperbolic Tangent transformation. A loss-tolerant training mechanism was introduced to reduce the impact of label noise. The model was trained on the PTB-XL dataset, and lifestyle-related analyses were conducted using the PLHDB and AISDB databases, which include both ECG recordings and behavioural information.

The proposed model achieved a mean absolute error (MAE) of 6.76 years on the PTB-XL dataset, outperforming existing approaches under comparable conditions. Lifestyle analysis revealed that regular exercise was associated with a 1.31-year reduction in predicted cardiac age, while smoking and excessive alcohol intake were associated with increases of 3.67 and 1.22 years, respectively.

These findings demonstrate the utility of ECG-based deep learning models not only for precise cardiac age estimation but also for elucidating the effects of lifestyle behaviours on cardiovascular health.