Clinical Prediction Models for Valvular Heart Disease

Area of Science:

• Cardiology
• Medical Informatics
• Health Services Research

Background:

• Numerous clinical prediction models (CPMs) exist for valvular heart disease (VHD) treatment decisions.
• The comparative performance and external validation of these VHD CPMs are not well understood.

Purpose of the Study:

• To systematically review and describe available CPMs for VHD.
• To specifically assess the performance of these CPMs in external validation studies.

Main Methods:

• A systematic review identified 49 CPMs for VHD patients undergoing surgery, percutaneous intervention, or no intervention.
• Analyzed 204 external validations across these CPMs, noting populations and performance metrics.

Main Results:

• Only 71% of identified CPMs have undergone external validation.
• 65% of external validations used distantly related populations, showing a median -27.1% change in discrimination.
• EuroSCORE II and STS (2009) models showed variable performance but performed better in related populations; few validations exist for transcatheter aortic valve replacement CPMs.

Conclusions:

• Many VHD CPMs lack external validation, and isolated validations are insufficient to confirm trustworthiness.
• Existing surgical VHD models perform adequately in related populations.
• Transcatheter aortic valve replacement CPMs require additional external validation to ensure reliable predictions.