# Development and validation of a machine learning predictive model for one-month post-revascularization angina in patients who had undergone PCI or CABG

## TL;DR

This study developed a machine learning model to predict angina recurrence after heart procedures, showing strong accuracy and potential for personalized care.

## Contribution

A novel machine learning model for predicting post-revascularization angina using diverse clinical and psychological factors.

## Key findings

- The random forest model achieved an AUC of 0.90 in internal validation and 0.87 in external validation.
- Key predictors included NYHA classification, cardiac troponin T, and depression severity.
- The model integrates cardiac, hemostatic, psychological, and metabolic factors for risk stratification.

## Abstract

Recurrent angina pectoris following coronary revascularization via percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG) poses significant clinical challenges, associated with reduced quality of life and increased healthcare burden. Traditional risk tools have limitations in predicting short-term recurrence. This study aimed to develop and validate a machine learning (ML) predictive model for post-revascularization angina (PRA).

This study used patient data from 38 clinical research centers in 23 provinces of China from 2016 to 2018. Data from 626 patients in a derivation cohort recruited from 28 centers across 16 Chinese provinces and 127 in an external validation cohort from another 10 centers across 10 provinces were analyzed. The Boruta algorithm selected key features, and eight ML models were trained on 70% of the derivation cohort, internally validated on 30%, and externally validated. Performance metrics included area under the curve (AUC), decision curve analysis (DCA), accuracy, sensitivity, specificity, and F1 score. The Shapley Additive explanation (SHAP) values provided model interpretability.

The Boruta algorithm selected six features: New York Heart Association (NYHA) classification, cardiac troponin T (cTnT), prothrombin time (PT), depression severity, abdominal circumference, and diastolic blood pressure (DBP). The Random forest (RF) model outperformed others, achieving an AUC of 0.90 (accuracy 0.88, sensitivity 0.77, specificity 0.92, F1 0.78) in internal validation and 0.87 in external validation. The SHAP algorithm confirmed the features’ predictive importance, with higher NYHA class, elevated cTnT, and depression severity positively influencing PRA risk.

This RF model offers a robust, interpretable tool for early PRA risk stratification, integrating cardiac, hemostatic, psychological, and metabolic factors. It supports personalized post-revascularization care, though prospective, multi-ethnic validation is needed to enhance generalizability.