Machine Learning Methods for Predicting Long-Term Mortality in Patients After Cardiac Surgery.

Area of Science:

• Cardiovascular Surgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Predicting long-term mortality post-cardiac surgery is crucial for patient management.
• Existing risk stratification models may not fully capture complex patient data.
• Machine learning offers potential for more accurate prognostic assessments.

Purpose of the Study:

• To develop and validate machine learning algorithms for predicting long-term mortality in cardiac surgery patients.
• To identify significant risk factors associated with mortality after cardiac surgery.
• To compare the performance of various machine learning models for prognostic accuracy.

Main Methods:

• Retrospective analysis of the MIMIC-III database.
• Inclusion of demographics, comorbidities, vital signs, lab results, and treatment data.
• Application of eight machine learning models: logistic regression, neural network, naïve bayes, gradient boosting machine, adapting boosting, random forest, bagged trees, and eXtreme Gradient Boosting.
• Evaluation using Area Under the Receiver Operating Characteristic Curves (AUC), calibration curves, and Decision Curve Analysis (DCA).

Main Results:

• The Adapting Boosting (Ada) model achieved the highest discriminatory ability (AUC=0.801) and best goodness of fit.
• Random Forest (RF), Ada, and Bagged Trees (BT) models showed superior net benefit in Decision Curve Analysis.
• Top predictors identified by the Ada model included Red Blood Cell Distribution Width (RDW), Blood Urea Nitrogen (BUN), SAPS II, anion gap, age, urine output, chloride, creatinine, congestive heart failure, and SOFA.

Conclusions:

• The Adapting Boosting (Ada) model is a highly effective tool for predicting 4-year mortality after cardiac surgery.
• These findings support the development of early warning systems to improve post-operative care.
• Machine learning models can significantly enhance risk prediction and clinical decision-making in cardiac surgery patients.