A lightweight and explainable cardiac signal framework for screening-oriented cardiometabolic risk assessment

Early identification of cardiometabolic and autonomic dysfunction using electrocardiogram (ECG) signals is essential for preventive cardiovascular screening, especially in resource-constrained settings. This paper presents a lightweight and interpretable ECG analytics pipeline for (i) diabetes screening, (ii) physiology-driven autonomic dysfunction screening (ADS) as a proxy task, and (iii) heart disease risk stratification. Handcrafted ECG statistics and heart rate variability (HRV) features were retrieved and combined at the subject level to prevent data leakage across repeated recordings. A novel Heart Disease Risk Index (HDRI) is introduced by combining normalized HRV suppression, heart-rate increase, and signal instability, enabling risk stratification without deep learning (DL). With strict subject-wise GroupKFold validation on multi-record ECG data (29 subjects, 131 recordings), the combined statistical + HRV feature set shows strong screening discrimination, with mean AUCs of 0.87 for diabetes screening, 0.95 for autonomic dysfunction screening proxy, and 0.99 for high-versus low-risk HDRI-based stratification. This is better than individual feature groups. Statistical analysis shows that diabetic participants have significantly higher HDRI values (p < 0.05), supporting the physiological plausibility of this finding. Feature-to-physiology mapping indicates that reduced RMSSD, along with an increased heart rate, signifies autonomic imbalance, aligning with recognized cardiometabolic risk factors. The suggested framework is meant to be a decision-support tool that is easy to use and is based on screening, not a diagnostic system. The framework demonstrates methodological robustness and physiological coherence under limited-cohort evaluation.