Meta-prediction of coronary artery disease risk.

Summary

Using UK Biobank for development and All of Us for external validation, the authors built a 10-year incident CAD risk model that integrates genetic and clinical data into 15 meta-features and achieved AUC 0.84 (external 0.81). The framework also estimates individualized benefits of standard interventions, enabling tailored prevention strategies based on genetic and phenotypic profiles.

Key Findings

• A 10-year CAD risk model using 15 derived meta-features achieved AUC 0.84 in UK Biobank and 0.81 in All of Us, outperforming standard clinical scores.
• The framework integrates multiple polygenic risk scores with demographics, labs, vitals, medications, and diagnoses (~2,000 candidate features).
• It quantifies individualized benefits of standard interventions, showing genetic risk modulates the magnitude of risk reduction.
• The approach provides a generalizable meta-prediction pipeline for precision risk estimation across cohorts.

Clinical Implications

Clinicians could use this tool to refine CAD risk stratification beyond traditional scores and to counsel patients on the expected impact of lifestyle and pharmacologic interventions tailored to their genetic and clinical profiles.

Limitations

• Potential calibration and transportability issues across diverse healthcare systems and ancestries
• Black-box aspects of meta-features may limit interpretability without transparent model documentation

Future Directions

Prospective impact studies, clinical integration trials, and assessments across ancestries and health systems are needed; open-source tools and calibration frameworks could accelerate adoption.