AI-Powered ECG Model Predicts Postoperative Atrial Fibrillation (AF) After Cardiac Surgery
New research confirms that an artificial intelligence (AI) model trained on millions of electrocardiograms (ECGs) serves as a novel, robust, and independent predictor of postoperative atrial fibrillation (AF) following cardiac surgery. It provides additive or synergistic predictive value when integrated with existing postoperative AF prediction tools or other risk factors. The AI-enhanced model significantly improves the accuracy of existing prediction tools, acting as a noninvasive biomarker for preoperative risk stratification for postoperative AF prediction in cardiac surgery patients, potentially enabling healthcare providers to initiate tailored prophylactic therapy and implement targeted monitoring during the perioperative period.
These findings were published in the November issue of the Journal of Medical Internet Research (J Med Internet Res).
Postoperative atrial fibrillation (AF) is widely recognized as the most frequently occurring complication following cardiac surgery, with reported incidence rates sometimes reaching as high as 50%. The development of postoperative AF is linked to substantial clinical and economic repercussions, including heightened rates of morbidity, an increased risk of mortality, extended hospital stays, and significantly elevated health care costs. Despite overall improvements in cardiac surgery outcomes, the incidence of postoperative AF has remained largely unchanged over the past several decades. Existing strategies aimed at preventing postoperative AF are often suboptimal, and traditional risk scores have proven statistically weak and inconclusive. Therefore, the identification of patients with a high likelihood of developing this condition is crucial for initiating tailored prophylactic therapy and targeted monitoring during the perioperative period.
To address this unmet need, researchers conducted a single-center retrospective cohort study involving 2266 adult patients who underwent cardiac surgery at a tertiary hospital in South Korea. The final analysis included 5204 standard 12-lead ECGs. The central component of the study was the AI-ECG-AF model, which was trained on an extensive database of 4.05 million non-AF standard 12-lead ECGs from 1.13 million patients using a 1D EfficientNet-B0 architecture. Postoperative AF was defined as AF documented by ECG within 30 days after surgery. The researchers assessed the AI-ECG-AF model score using multivariable logistic regression, adjusting for clinical variables, and evaluated its additive value by combining it with the existing postoperative AF prediction tool, known as the POAF score.
The AI-ECG-AF model demonstrated strong performance in its development, achieving an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.901 in its test set for identifying the electrocardiographic signature of AF in non-AF ECGs. Crucially, the model was validated as a powerful independent risk factor for postoperative AF in the cardiac surgery cohort. After adjusting for clinical variables, a 10% absolute increase in the AI-ECG-AF model score was associated with a 1.197- to 1.209-fold increase in the odds of developing postoperative AF. For instance, combining the AI-ECG-AF model with variables from the POAF score increased the odds of developing postoperative AF by 19.7% for a 10% absolute increase in the score. Furthermore, the AI model significantly enhanced existing prediction capabilities: the AUROC of the standard POAF score alone was 0.643; adding the AI-ECG-AF model score increased the AUROC to 0.680 (P<.001), and combining the AI-ECG-AF score with other risk factors raised it to 0.710 (P<.001).
In conclusion, the AI-ECG-AF model provides additive or synergistic predictive value when integrated with existing postoperative AF prediction tools or other risk factors. By capturing atrial electrophysiological vulnerability not reflected in conventional clinical scores, the AI-ECG-AF model may function as a noninvasive biomarker for preoperative risk stratification in cardiac surgery patients. The identification of these underlying predispositions allows for the recognition of individuals at heightened risk. This predictive capability has significant clinical implications, potentially enabling health care providers to initiate tailored prophylactic therapy and implement closer rhythm monitoring during the perioperative and long-term follow-up periods for high-risk patients.
Reference: Han C, Soh S, Park JW, Pak HN, Yoon D. Artificial Intelligence-Based Electrocardiogram Model as a Predictor of Postoperative Atrial Fibrillation Following Cardiac Surgery: Retrospective Cohort Study. J Med Internet Res. 2025 Nov 10;27:e77164. doi: 10.2196/77164. PMID: 41213128; PMCID: PMC12603327.
