Aims: The widespread adoption of consumer-grade wearable sensors has made both electrocardiogram (ECG) and photoplethysmogram (PPG) signals increasingly available. While the utility of ECG-derived markers for cardiovascular risk assessment is well established, the combined use of ECG and PPG features remains underexplored. This study aimed to determine whether integrating ECG and PPG features improves prediction of major adverse cardiovascular and cerebrovascular events (MACCE) compared with either modality alone. Methods: A total of 45,776 UK Biobank participants without prior MACCE and with same-day (but not simultaneous) ECG and PPG recordings were included. Signal-averaged single-beat PPG waveforms and 15-second resting ECG recordings were analysed, and standard features were extracted, including PPG morphological markers, resting heart rate, heart rate variability, and repolarization variability. Survival analysis was performed using Cox proportional hazards model regression and XGBoost with a Cox objective. Model performance was evaluated in a held-out test set (80:20 train:test split). Models incorporated demographic variables (age, sex, and BMI), ECG-derived features, PPG-derived features, and their combinations.
Results: During a median follow-up of 12.4 years (IQR 5.0–12.6), MACCE occurred in 3.5% of participants. Models based on XGBoost consistently outperformed those using the Cox proportional hazards model. As shown in Table 1, the demographic model achieved a C-index of 0.70 (95% CI 0.69–0.71). The addition of PPG features improved performance to 0.72 (95% CI 0.71–0.73), while ECG features yielded a C-index of 0.73 (95% CI 0.72–0.74). The combined model incorporating demographic, ECG-, and PPG-derived features demonstrated the highest performance (C-index 0.76; 95% CI 0.75–0.77).
Conclusions: The combination of PPG- and ECG-derived features yielded greater predictive for cardiovascular risk compared with either modality alone.