# Machine learning-selected inflammation biomarkers for stable coronary artery disease with intermediate coronary lesions: potential for long-term prognosis in a multicenter cohort study

## TL;DR

This study uses machine learning to identify inflammation biomarkers that predict long-term outcomes for patients with stable coronary artery disease and intermediate coronary lesions.

## Contribution

A novel predictive model using machine learning and EHR data to assess prognosis in SCAD patients with intermediate lesions.

## Key findings

- The model included platelet-to-lymphocyte ratio, diabetes, lipoprotein(a), and mean platelet width as key predictors.
- The model showed moderate discriminative ability with AUC values between 0.658 and 0.743 over 2-4 years.
- Calibration and decision curve analysis confirmed the model's clinical utility in identifying high-risk patients.

## Abstract

Stable coronary artery disease (SCAD) generally exhibits prolonged periods of stability. However, this condition can unpredictably progress into an unstable state, representing a complex pathological process involving multiple contributing factors. Thus, we aimed to utilize machine-learning techniques to identify predictive features from electronic health record (EHR) data for forecasting the long-term prognosis of patients with SCAD and intermediate coronary lesions.

Patients were divided into a training cohort (n = 403) and an external validation cohort (n = 247) according to their hospital of origin during the period from January 2018 to December 2020. Predictive features were determined using LASSO regression analysis and boruta algorithm, followed by multivariate Cox regression analysis for model construction.

The developed predictive model comprised four clinical variables: platelet-to-lymphocyte ratio, diabetes mellitus, lipoprotein(a), and mean platelet width. The area under the curve for predicting major adverse cardiovascular events (MACEs) within 2-, 3- and 4-year in the development cohort was 0.692 (95%CI:0.59-0.793), 0.709 (95%CI:0.625-0.792) and 0.743 (95%CI:0.672-0.813), respectively, while that in the external validation cohort was 0.658 (95%CI 0.542-0.773), 0.681 (95%CI:0.579-0.782) and 0.723 (95%CI: 0.635-0.811), respectively. Additionally, the developed predictive model was calibrated by analyzing the correlation between expected and observed MACEs in the development and external validation cohorts. Lastly, the clinical value of the developed predictive model was confirmed via decision curve analysis.

Our validated nomogram was based on inflammation biomarkers and EHR data, demonstrating moderate discriminative ability to detect individuals at high risk of poor outcome among patients with SCAD and angiographically intermediate coronary stenosis.