| Title: | Multimodal artificial intelligence for cross-population prediction of major adverse cardiovascular events: A multi-cohort external validation study |
| Journal: | - |
| Published: | 1 Jun 2026 |
| DOI: | https://doi.org/10.1016/j.ijci.2026.100003 |
Publication 19853
WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
Abstract
Early identification of individuals at risk for Major Adverse Cardiovascular Events (MACE) remains a major challenge in preventive cardiology. Conventional risk scores incompletely integrate multimodal cardiovascular data and often demonstrate limited generalizability across heterogeneous populations. We developed and externally validated a multimodal artificial intelligence model integrating echocardiographic, electrocardiographic, and circulating biomarker data for cross-population cardiovascular risk stratification. A total of 12,845 individuals from five independent cohorts - UK Biobank, MIMIC-IV Waveform Database, PTB-XL ECG Dataset, Framingham Heart Study, and a multicenter echocardiography registry (2012-2024) - were included. While multimodal data were available in the majority of cohorts, PTB-XL and MIMIC-IV datasets had partial modality availability, which was addressed using modality-aware modeling. The primary endpoint was 3-year MACE. The proposed framework was compared with the Framingham Risk Score, ASCVD Risk Estimator, logistic regression, Convolutional Neural Network (CNN), and transformer-based models. Model discrimination, calibration, and clinical utility were evaluated using area under the receiver operating characteristic curve (AUC), Brier score, calibration slope and intercept, and decision curve analysis. During follow-up, 1624 patients (12.6%) experienced MACE. The multimodal model achieved external validation AUCs of 0.89-0.92, significantly outperforming baseline models (p < 0.001). Calibration remained stable across cohorts, with improved net reclassification and consistent performance across age and sex subgroups, supporting clinical integration.</p>
6 Authors
- I. Govindharaj
- G. Michael
- P. Manju Bala
- S. Balamurugan
- R. Sathishkumar
- E. Bharath
Enabling scientific discoveries that improve human health