Machine learning-based long-term outcome prediction in patients undergoing percutaneous coronary intervention

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Traditional risk assessment for percutaneous coronary intervention (PCI) patients relies on limited clinical and imaging data.
• Machine learning (ML) offers potential for comprehensive cardiovascular risk characterization using complex variables.

Purpose of the Study:

• To evaluate the efficacy of ML models in predicting long-term all-cause mortality in patients with coronary artery disease (CAD) before PCI.
• To compare the performance of various ML algorithms for cardiovascular risk stratification.

Main Methods:

• Prospective enrollment of 9,680 CAD patients undergoing PCI.
• Training six ML models (including random forest) using 87 baseline measurements.
• Evaluating model performance using 10-fold cross-validation and predicting 5-year cardiovascular outcomes.

Main Results:

• The random forest (RF-PCI) model showed the best performance in predicting all-cause mortality (AUC: 0.71±0.04).
• Top 15 prognostic features included 11 laboratory measures.
• The RF-PCI model provided meaningful risk stratification for patients.

Conclusions:

• ML models enhance the prediction of long-term mortality in CAD patients undergoing PCI.
• The random forest model outperforms other ML approaches for improved patient risk stratification.