Abstract 4362776: Multimodal and Serial Electrocardiogram and Chest X-ray Age Have Prognostic Significance as Biomarkers for Cardiovascular Health

作者
Veer Sangha,Xiao Gu,Rohan Khera,David Andrew Clifton
出处
期刊:Circulation [Lippincott Williams & Wilkins]
卷期号:152 (Suppl_3)
标识
DOI:10.1161/circ.152.suppl_3.4362776
摘要

Background: Estimating biological age from diagnostic tests such as electrocardiograms (ECGs) and chest X-rays (CXRs) via artificial intelligence (AI) is an emerging strategy for cardiovascular (CVD) risk stratification. We investigated whether AI-predicted age from ECG and CXR images can serve as multimodal and dynamic biomarkers of CVD risk. Methods: EfficientNet-B3 convolutional neural networks were trained to predict chronological age from 12-lead ECG and frontal CXR images using data from Yale and the CheXpert dataset, respectively. Models were validated externally using the MIMIC-IV cohort. The difference between predicted and actual age (delta-age) was assessed for its association with major adverse CVD events (MACE) and mortality in a cohort of patients with both ECGs and CXR using age- and sex-adjusted multivariable Cox models. Additional analyses evaluated the prognostic significance of rate of biological aging, defined as the slope from a linear regression of AI-predicted age versus time in patients with multiple recordings. Results: On internal validation, the AI-ECG and AI-CXR models achieved mean absolute errors of 7.5 and 5.4 years, respectively. In MIMIC-IV, analysis of 16,294 patients with both ECG and CXR confirmed that these biomarkers provided additive prognostic information. Patients with high delta-age ECGs (delta > 5.5 years, top quartile) had higher risk of mortality (HR: 1.17) and higher risk of MACE (HR: 1.32), while a high delta-age CXR (delta > 4.8 years, top quartile) also portended higher mortality (HR 1.53) and MACE risk (HR 1.36). Patients with high delta-ages in both modalities had the highest hazard in cumulative hazard plots. In longitudinal analyses, in a cohort of 49,702 with multiple ECGs, both accelerated (top quartile) and decelerated (bottom quartile) rates of AI-predicted aging had higher risk of mortality (HR 4.16 and 4.08 respectively) and MACE (HR 3.07 and 2.65 respectively) even after accounting for delta-age from the initial recording. A cohort of 14,748 patients with multiple CXRs saw similar patterns for accelerated and decelerated agers (HR 3.48 and 2.66 respectively for mortality, and HR 2.13 and 1.75 respectively for MACE). Conclusion: AI-derived age from ECG and CXR images are robust biomarkers of CVD risk. These biomarkers offer additive and dynamic prognostic value when used in combination or over time, supporting a shift toward personalized, multimodal, and temporally adaptive risk assessment.

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
wangchaofk发布了新的文献求助10
刚刚
圆1223完成签到 ,获得积分10
1秒前
1秒前
LEE发布了新的文献求助10
2秒前
xiayut发布了新的文献求助10
3秒前
3秒前
顺利完成签到,获得积分10
4秒前
Ryun完成签到,获得积分10
4秒前
5秒前
灰太狼发布了新的文献求助10
5秒前
内向的小凡完成签到,获得积分0
6秒前
科研通AI2S的应助被伶俐浩轩采纳,获得10
6秒前
大力思雁完成签到,获得积分10
7秒前
8秒前
shan发布了新的文献求助10
9秒前
小杰完成签到,获得积分10
9秒前
wangchaofk完成签到,获得积分10
9秒前
852的应助被新月采纳,获得10
9秒前
密密麻麻的完成签到 ,获得积分10
9秒前
10秒前
10秒前
wenz01完成签到,获得积分10
10秒前
shirleen完成签到,获得积分10
10秒前
11秒前
null的应助被Ryun采纳,获得10
11秒前
852的应助被ewmmel采纳,获得10
11秒前
yx完成签到,获得积分10
11秒前
我是老大的应助被JLW采纳,获得10
11秒前
LEE完成签到,获得积分10
11秒前
塔罗完成签到,获得积分10
13秒前
求田问舍发布了新的文献求助10
13秒前
13秒前
科研通AI6.4的应助被玖熙采纳,获得10
14秒前
禾几一鸣发布了新的文献求助10
15秒前
15秒前
打打的应助被shirleen采纳,获得10
15秒前
16秒前
沃德天sama完成签到,获得积分10
16秒前
17秒前
taotao完成签到,获得积分10
18秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1500
John McDowells Bildungstheorie 800
Analyses de reconnaissances de dettes passées devant les échevins d'Ypres (1249-1291) 600
Handbook of Pharmaceutical Excipients. 9th ed 400
Synthesis, Application and Future Perspectives of Smart Nano-materials - Part 1 400
Breaking the World 300
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8153710
求助须知:如何正确求助?哪些是违规求助? 9550475
关于积分的说明 21698625
捐赠科研通 7742521
什么是DOI,文献DOI怎么找? 3425571
关于科研通互助平台的介绍 2550735
邀请新用户注册赠送积分活动 2469211