Dual-modal AI wearable system for geriatric AF-HF-SAS comorbidity management: a national database-backed translational study

Background The comorbid triad of atrial fibrillation (AF), heart failure (HF), and sleep apnea syndrome (SAS) forms a fatal vicious cycle in elderly populations, driving accelerated disease progression and poor clinical outcomes. Conventional optical wearables have critical limitations of signal drift and low accuracy in elderly patients with skin aging and motion artifacts, and there remains an unmet need for clinical-grade tools to unmask the mechanistic links of this triad and enable early risk stratification. Purpose This study aims to translate our proprietary dual-modal sensing technology into a clinical-grade wearable system, validate its diagnostic performance for AF and SAS in elderly patients, and develop a risk prediction model for HF decompensation using our team-led national database. Methods We leveraged the team-spearheaded MIIT AI Standardized Database for Heart Failure, extracting a geriatric comorbidity cohort from 1.05 million patients in the China Heart Failure Alliance to build a multimodal risk prediction model. We deployed a self-developed fingertip bio-inspired flexible wearable system with dual PENG-PPG sensing. The SAS detection algorithm, based on a Vision Transformer architecture, was pre-trained on 7,300 PSG-annotated 2-minute data segments, with further optimization for geriatric populations. A diagnostic trial was conducted in 80 elderly arrhythmia patients, with simultaneous Holter monitoring as the gold standard. Results The system achieved 91% diagnostic accuracy for AF in the elderly cohort, overcoming signal loss limitations of conventional optical wearables. It maintained a SAS detection rate of ≥85%, and multimodal analysis revealed a strong temporal correlation between high nocturnal SAS burden and elevated AF burden the following day. The AI model trained on the national database demonstrated that combined monitoring of SAS burden and AF rhythm significantly improved HF decompensation risk prediction, outperforming single-marker monitoring. Conclusion This national database-backed dual-modal AI wearable system delivers medical-grade AF diagnostic accuracy in elderly patients, uncovers the mechanistic link of the AF-HF-SAS vicious cycle, and provides a mechanism-based early warning tool for HF risk stratification, with significant value for geriatric cardiovascular comorbidity management.