Methods for updating a risk prediction model for cardiac surgery: a statistical primer

Area of Science:

• Medical Statistics
• Cardiovascular Surgery
• Health Informatics

Background:

• Risk prediction models in cardiac surgery often experience a decline in predictive accuracy over time.
• Maintaining model performance is essential for reliable clinical decision-making and patient outcomes.
• Developing new models de novo is resource-intensive; updating existing models offers a potentially efficient alternative.

Purpose of the Study:

• To provide a statistical overview of methods for updating risk prediction models in cardiac surgery.
• To discuss the strengths and weaknesses of various model updating techniques.
• To emphasize the importance of model updating for sustained predictive performance.

Main Methods:

• Review of statistical methods for model updating, including intercept recalibration, logistic recalibration, model revision, closed test procedure, and Bayesian modeling.
• Application of logistic recalibration to the EuroSCORE II model using a Dutch national cohort.
• Recommendations for transparent reporting using the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement.

Main Results:

• Logistic recalibration demonstrated a significant improvement in model performance.
• Logistic recalibration successfully avoided the risk of overfitting.
• The study illustrated that increased data availability permits more extensive model updating strategies.

Conclusions:

• Model updating is a viable and efficient strategy to maintain the performance of cardiac surgery risk prediction models.
• Logistic recalibration is an effective method for improving model performance without overfitting.
• Transparent reporting and performance assessment (calibration/discrimination plots) are vital for updated models.