Ensemble machine learning prediction and variable importance analysis of 5-year mortality after cardiac valve and...

Area of Science:

• Cardiovascular Surgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Cardiac valve operations have higher mortality than coronary artery bypass grafting (CABG).
• Predictors for long-term mortality after cardiac surgery are not well-established.
• Accurate prediction models are crucial for patient management and risk stratification.

Purpose of the Study:

• To develop and validate an ensemble machine learning model for predicting 5-year mortality after cardiac operations.
• To identify novel peri-operative predictors of long-term mortality.
• To compare the predictive performance across different cardiac surgical procedures.

Main Methods:

• Utilized an ensemble machine learning algorithm (Super Learner).
• Trained the model on prospectively collected peri-operative data from 8241 patients.
• Included 88 routinely collected peri-operative variables for analysis.

Main Results:

• Ensemble machine learning demonstrated high predictive accuracy for 5-year mortality (e.g., mitral valve: 0.846, aortic valve: 0.838).
• Post-operative urea was identified as a novel and significant predictor of mortality.
• The model effectively integrated known risk factors like age and postoperative creatinine with new predictors.

Conclusions:

• Ensemble machine learning accurately predicts long-term mortality using routine peri-operative data in cardiac surgery.
• Post-operative urea is a significant, previously unrecognized predictor of mortality after cardiac operations.
• This approach enhances risk stratification and can inform clinical decision-making for cardiac surgery patients.