A machine learning approach to high-risk cardiac surgery risk scoring

Area of Science:

• Cardiovascular Surgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• High-risk cardiac surgery presents outcome uncertainty, complicating clinical decisions.
• A machine learning (ML) approach was developed to personalize mortality risk assessment.
• This study aimed to identify weighted, patient-specific factors influencing mortality.

Purpose of the Study:

• To develop and validate ML models for predicting mortality in high-risk cardiac surgery.
• To identify and quantify individual patient factors contributing to surgical mortality.
• To enhance clinical decision-making through personalized risk prediction.

Main Methods:

• Utilized the American College of Surgeons National Surgical Quality Improvement Program database (2012-2019).
• Applied gradient boosting machine (GBM) modeling with significant predictors.
• Employed local interpretable model-agnostic explanations (LIME) for individual patient predictions.

Main Results:

• Included 1291 high-risk cardiac surgery patients with 194 deaths; GBM showed superior performance.
• Top mortality predictors identified by LIME: preoperative dialysis, emergent surgery, Hispanic ethnicity, steroid use, ventilator dependence.
• LIME provided individualized patient mortality probabilities and explanations.

Conclusions:

• ML techniques effectively predict individualized mortality risk in high-risk cardiac surgery.
• Identified key patient-specific factors influencing mortality, enabling personalized risk assessment.
• Suggests applying this ML model to other databases, like STS, for broader clinical utility.