Machine Learning in Cardiac Surgery: Predicting Mortality and Readmission.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Cardiovascular Surgery

Background:

• Predicting patient outcomes in open-heart surgery is complex, with significant economic impacts from readmissions, prolonged hospital stays, and mortality.
• Machine learning (ML) offers potential for improved outcome prediction in this domain.

Purpose of the Study:

• To evaluate the performance of machine learning (ML) in data visualization and predicting patient outcomes following open-heart surgery.
• To compare ML model performance at both cohort and patient levels against established risk models.

Main Methods:

• Analysis of 8,947 patients undergoing cardiac surgery from April 2006 to January 2018.
• Utilized clustering, correlation matrix, and seven predictive models for outcome classification (Discharged, Died, Readmitted).
• Employed cross-validation, hyperparameter optimization, and data imputation for robust model training and testing.

Main Results:

• Machine learning demonstrated strong predictive performance for mortality (AUC 0.83 ± 0.03) and readmission (AUC 0.75 ± 0.035).
• Cohort-level ML analysis showed performance comparable to the Society of Thoracic Surgeons (STS) risk model, even with fewer samples.
• Patient-level analysis with all cases yielded ML performance consistent with reported STS model benchmarks.

Conclusions:

• Machine learning provides a systematic and effective approach to analyzing and predicting outcomes in open-heart surgery.
• The study highlights the significant predictive utility and clinical implications of ML in cardiac surgery.
• Further validation is needed to assess the generalizability of the developed ML models beyond the study institution.