A multivariate Bayesian model for assessing morbidity after coronary artery surgery.

Area of Science:

• Cardiovascular Surgery
• Medical Informatics
• Biostatistics

Background:

• Traditional risk stratification for coronary artery bypass grafting (CABG) primarily uses preoperative variables.
• Intraoperative and postoperative factors also significantly influence patient prognosis.
• Previous research evaluated predictors, but novel modeling approaches are needed.

Purpose of the Study:

• To develop and validate a Bayes linear model for discriminating morbidity risk post-CABG.
• To compare the performance of the Bayes model against existing scoring systems, including the Higgins' score and locally customized models.

Main Methods:

• Analysis of 1,090 consecutive adult patients undergoing CABG, with data split into training (740) and testing (350) sets.
• Identification of optimal predictor variables using a stepwise approach from 88 operative risk factors.
• Assessment of model discrimination via receiver operating characteristic (ROC) curves and calibration using the Hosmer-Lemeshow goodness-of-fit test.

Main Results:

• A set of 12 preoperative, intraoperative, and postoperative variables was identified for the Bayes linear model.
• The Bayes linear classifier demonstrated significantly higher discrimination capacity than the compared score models.
• Both calibration and discrimination were notably poorer with the Higgins' original scoring system.

Conclusions:

• While score models are simple for clinical practice and acceptable when locally customized, the Bayesian model presents a feasible alternative.
• The Bayesian approach offers superior discrimination and greater ease of tailoring to individual healthcare institutions.
• This suggests a potential shift towards more sophisticated, data-driven risk prediction in cardiovascular surgery.