Predicting Readmissions or Post-Discharge Mortality After Cardiac Surgery with Machine Learning Using an Australian...

Area of Science:

• Cardiovascular Surgery
• Health Informatics
• Machine Learning in Healthcare

Background:

• Post-discharge adverse events after cardiac surgery pose significant challenges.
• Predictive models are crucial for identifying high-risk patients and improving outcomes.
• The Australian & New Zealand Society of Cardiac & Thoracic Surgeons (ANZSCTS) Database offers a valuable resource for such analyses.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for predicting 30-day readmissions or post-discharge mortality following cardiac surgery.
• To assess the predictive performance of these models using established metrics.
• To identify potential improvements for future predictive modeling efforts.

Main Methods:

• Utilized data from 54 Australian hospitals (2017-2021) for coronary artery bypass grafting (CABG), valvular, and aortic surgery.
• Developed and tested various machine learning models, including deep neural networks.
• Performance was evaluated using sensitivity, specificity, predictive values, accuracy, AUROC, and AUPRC.

Main Results:

• A dataset of 61,721 cardiac surgery cases was analyzed, with a 10.3% incidence of 30-day readmission or mortality.
• The best-performing deep neural network model achieved an AUROC of 0.615.
• A calibrated model identified high-risk patients, showing a 2.4x increase in incidence rate for readmissions or mortality.

Conclusions:

• Machine learning models derived from the ANZSCTS Database showed limited predictive power for 30-day adverse events.
• The current ANZSCTS Database may lack sufficient variables for highly accurate predictions.
• Integrating electronic medical record data with the ANZSCTS Database could enhance model accuracy at the hospital or network level.