Risk prediction modelling of 30-day all-cause mortality following percutaneous coronary intervention in an Australian population: leveraging machine learning

Comprehensive Summary

In this study, researchers evaluated the performance of several machine learning (ML) models in the prediction of 30-day all-cause mortality following percutaneous coronary intervention (PCI). Over 93,000 PCI cases from hospitals within Victoria, Australia were gathered and evaluated for various risk factors, including age, sex, left ventricular ejection fraction (LVEF), and medical history. Seven ML models (Decision Tree, Decision Tree, Extreme Gradient Boosting, Gradient Boosting, Linear Discriminant Analysis, LR, Random Forest, and Stochastic Gradient Boosting) were trained on 70% of the data and validated on the remaining 30%. Out of the seven, the Extreme Gradient Boosting (XGB) model slightly outperformed the others, yielding the best metrics in accuracy (86.7%), RMSE (36.5%), and specificity (82.5%), among others. It also had satisfactory metrics in metrics such as sensitivity/recall (76.5%) and ROC-AUC (85.5%). The XGB model identified LVEF, ACS, eGFR, age and complex lesion as five leading factors contributing to 30-day all-cause mortality. Researchers employed a SHAP plot to explain the influence and importance of these variables.

Outcomes and Implications

PCI is an extremely common operation performed to treat coronary artery disease. While the procedure is largely safe, the prevalence of 30-day all-cause mortality is about 2%. Risk prediction models can help both clinicians and patients alike make better informed choices regarding pre- and post-operative care. However, there still remains a need for future research, especially for evaluating the performance of other machine learning models, including different risk factors into model prediction, and validating these models with diverse datasets.