Comparative Analysis of Coronary Surgery Risk Stratification Models

Area of Science:

• Cardiovascular Surgery
• Health Services Research
• Biostatistics

Background:

• Preoperative risk assessment is crucial for coronary artery bypass grafting (CABG) surgery.
• Existing risk models require validation on independent datasets to ensure generalizability.
• Comparing the performance of various CABG risk stratification tools is essential for clinical utility.

Purpose of the Study:

• To compare the predictive accuracy of established and newly developed risk assessment models for hospital mortality in coronary artery bypass grafting (CABG) patients.
• To evaluate the validity of existing CABG models using a local patient database.
• To assess the variability in mortality predictions across different risk models.

Main Methods:

• Tested four existing models (Parsonnet, Cleveland, STS v1, STS v2) and two new models (Bayesian, Logistic Regression) on a dataset of 3,443 CABG patients.
• Calibrated Parsonnet and Cleveland models using historical data from 2,842 patients operated on before 1991.
• Compared models based on predicted vs. observed mortality, Brier score, C-Index (AUC), and predictive efficiency, including a subset of 3,237 isolated CABG patients.

Main Results:

• Observed mortality was 4.0% for all CABG patients and 3.7% for isolated CABG patients.
• C-Indexes for the models ranged from 0.77 to 0.83, indicating approximately 80% discrimination accuracy.
• Bayesian and Logistic Regression models showed strong performance, with C-Indexes of 0.83 and 0.80 respectively.
• Significant variation in predicted mortality was observed between different models.

Conclusions:

• Existing coronary artery bypass grafting (CABG) risk models demonstrate moderate predictive accuracy (around 80%).
• Models developed using national or non-local databases appear valid for local application.
• Current methods for comparing CABG risk models may be insufficient due to wide prediction variability.