Related Experiment Videos
Transformer-based models for predicting cardiovascular risk in Chinese adults: development and validation
Qiuyu Cao1,2, Xingkun Xu2, Hong Lin1,2
1Department of Endocrine and Metabolic Diseases, Shanghai Institute of Endocrine and Metabolic Diseases, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, 197 Rui Jin 2nd Road, Shanghai 200025, China.
European Heart Journal
|July 9, 2026
Summary
New transformer-based deep learning models (China-AIHeart) show improved cardiovascular disease (CVD) risk prediction in Chinese adults compared to traditional methods. These models offer a practical tool for better risk stratification.
Area of Science:
- Artificial Intelligence in Medicine
- Cardiovascular Disease Epidemiology
- Machine Learning for Healthcare
Background:
- Traditional Cox models perform suboptimally for cardiovascular disease (CVD) risk prediction in Chinese populations.
- Deep learning, particularly transformer models, shows promise for clinical risk prediction.
- There is a need for improved CVD risk assessment tools in China.
Purpose of the Study:
- To develop and validate sex-specific transformer-based models (China-AIHeart) for 10-year CVD risk prediction in Chinese adults.
- To compare the performance of China-AIHeart models against traditional Cox models and existing risk scores.
- To assess the clinical utility of transformer-based models for CVD risk stratification.
Main Methods:
- Development of sex-specific transformer-based time-to-event prediction models (China-AIHeart) using a large derivation cohort (156,790 participants).
- External validation in two independent Chinese cohorts (Xinjiang and CHARLS).
- Comparison of China-AIHeart models with Cox models and established risk scores (China-PAR, PREVENT-ASCVD, SCORE2 Asia-Pacific).
Main Results:
- China-AIHeart demonstrated good discrimination (C-statistics: .767 men, .780 women) and calibration.
- Models showed improved discrimination (ΔC-statistic: .027 men, .031 women) and reclassification compared to Cox models.
- China-AIHeart outperformed existing risk scores (China-PAR, PREVENT-ASCVD, SCORE2 Asia-Pacific) in external validation.
Conclusions:
- Transformer-based China-AIHeart models accurately predict 10-year CVD risk in Chinese adults.
- These models significantly outperform traditional Cox-based approaches.
- China-AIHeart provides a practical and effective tool for CVD risk stratification in the Chinese population.
