多因素分析危重患者新发房颤人工智能预警模型的构建

Objective To construct an artificial intelligence(AI) early warning model for new-onset atrial fibrillation(AF) in critically ill patients, providing a reliable tool for early identification of high-risk individuals in clinical practice. Methods A retrospective study was conducted, involving 317 inpatients from the critical care unit of our hospital from September 2024 to June 2025. Patients were divided into the AF group(76 cases) and the non-AF group(241 cases) based on whether they developed new-onset AF during hospitalization. Multi-dimensional information including time-series data, imaging data, text data, discrete data, and clinical medication use was collected. Features of AF were screened, and an AI early warning model was constructed using support vector machine(SVM) and generative adversarial network(GAN). Data statistics and model validation were performed using SPSS 26.0, Python 3.8, and Matlab 2021b. Results After feature screening, 12 key predictive features were identified, with the optimal correlation threshold set at 0.6(at which the model accuracy reached 89.2%). The GAN-SVM fusion model performed exceptionally well in the test set: accuracy 87.6%, sensitivity 85.3%, specificity 88.4%, and the area under the receiver operating characteristic curve(AUC) was 0.901, significantly outperforming the single SVM model(AUC=0.845, P<0.05). Conclusion The AI early warning model for new-onset AF can efficiently predict the risk of new-onset AF in critically ill patients. The GAN technology adopted in the model effectively addresses the issue of imbalanced clinical samples, and the selected key features provide a direction for precise clinical prevention and control.