用心电图和临床数据联合预测糖尿病，准确率显著提升
Improving Early Prediction of Type 2 Diabetes Mellitus with ECG-DiaNet: A Multimodal Neural Network Leveraging Electrocardiogram and Clinical Risk Factors
- 融合心电图与临床风险因子的多模态深度学习模型
- 预测准确率AUROC达0.845，优于单一数据源模型
- 适合关注糖尿病早期筛查及精准预防的研究者
2型糖尿病（T2DM）仍是全球健康挑战，亟需早期精准风险预测。本研究提出ECG-DiaNet，一种整合心电图（ECG）特征与临床风险因素（CRFs）的多模态深度学习模型，以提升T2DM发病预测能力。基于卡塔尔生物银行（QBB）数据，在开发队列（n=2043）上训练并验证模型，于纵向测试集（n=395，五年随访）上评估性能。相比仅使用ECG或仅使用CRFs的单模态模型，ECG-DiaNet表现更优，其AUROC为0.845，高于仅用CRFs模型的0.8217（DeLong检验，p<0.001）。再分类指标进一步验证：净重新分类改善（NRI=0.0153），综合区分度改进（IDI=0.0482）。将风险分为低、中、高三个等级后，高风险人群的阳性预测值（PPV）显著提高。该模型依赖无创且普及的心电图信号，具备在临床与社区卫生场景中的可行性。通过结合心脏电生理与系统性风险特征，ECG-DiaNet回应了T2DM的多因素本质，助力精准预防。研究结果凸显多模态AI在提升T2DM早期检测与防控策略中的价值，尤其适用于中东地区等代表性不足人群。
原文摘要 · Abstract (English)
Type 2 Diabetes Mellitus (T2DM) remains a global health challenge, underscoring the need for early and accurate risk prediction. This study presents ECG-DiaNet, a multimodal deep learning model that integrates electrocardiogram (ECG) features with clinical risk factors (CRFs) to enhance T2DM onset prediction. Using data from Qatar Biobank (QBB), we trained and validated models on a development cohort (n=2043) and evaluated performance on a longitudinal test set (n=395) with five-year follow-up. ECG-DiaNet outperformed unimodal ECG-only and CRF-only models, achieving a higher AUROC (0.845 vs 0.8217) than the CRF-only model, with statistical significance (DeLong p<0.001). Reclassification metrics further confirmed improvements: Net Reclassification Improvement (NRI=0.0153) and Integrated Discrimination Improvement (IDI=0.0482). Risk stratification into low-, medium-, and high-risk groups showed ECG-DiaNet achieved superior positive predictive value (PPV) in high-risk individuals. The model's reliance on non-invasive and widely available ECG signals supports its feasibility in clinical and community health settings. By combining cardiac electrophysiology and systemic risk profiles, ECG-DiaNet addresses the multifactorial nature of T2DM and supports precision prevention. These findings highlight the value of multimodal AI in advancing early detection and prevention strategies for T2DM, particularly in underrepresented Middle Eastern populations.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。