Transformer-based models for predicting cardiovascular risk in Chinese adults: development and validation
Authors' conclusion
The transformer-based China-AIHeart models predicted 10-year CVD risk and outperformed traditional Cox-based approaches, providing a practical tool for risk stratification in Chinese adults.
Study context
- Population
- Chinese adults without cardiovascular disease (CVD) (China Cardiometabolic Disease and Cancer Cohort; external: Xinjiang and CHARLS cohorts)derivation cohort
- Participants
- 156,790 adults
- Comparison
- China-AIHeart vs Cox models
- Outcome
- 10-year CVD risk prediction
- Follow-up
- 10 years
Methodological detailssex-specific models, external validation in two independent cohorts
- Sex-specific models
- External validation in two independent cohorts
Performance
As reported in the abstract
Discrimination (C-statistic)
China-AIHeart
- C-statistic, men
- .767 [.754-.779]
- C-statistic, women
- .780 [.769-.791]
- Calibration χ2, men
- 14.806
- Calibration χ2, women
- 9.326
- Brier score, men
- .104
- Brier score, women
- .077
China-AIHeart
C-statistic, men.767 [.754-.779]
C-statistic, women.780 [.769-.791]
Calibration χ2, men14.806
Calibration χ2, women9.326
Brier score, men.104
Brier score, women.077
- Men
- ΔC-statistic .027 (95% CI .025-.028)
- Women
- ΔC-statistic .031 (95% CI .029-.033)
- Men
- net reclassification index .478 (95% CI .466-.492)
- Women
- net reclassification index .560 (95% CI .551-.572)
Further findings
External validation C-statistics (men/women)
- Full model, Xinjiang
- .781/.825
- Simplified model, Xinjiang
- .748/.820
- Simplified model, CHARLS
- .740/.771
Original publication
Transformer-based models for predicting cardiovascular risk in Chinese adults: development and validation. Eur Heart J. Published 9 July 2026. doi:10.
Summarised from the published abstract; every number is checked against it