What is the clinical evidence from this study?

Study design: Cohort. Population: type 4b acute myocardial infarction (n=519). Primary outcome: Predictive performance of machine learning model for VLST-4b-AMI (null, 95% CI 0.640-0.809, p=null).

March 16, 2026Open Access

Risk factor analysis and machine learning-based prediction model for type 4b acute myocardial infarction induced by very late stent thrombosis

Q: What does this research mean for the field?

An XGBoost-based machine learning model demonstrates the best predictive performance for type 4b acute myocardial infarction induced by very late stent thrombosis, achieving an AUC of 0.724. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Key Result

The XGBoost model demonstrated a predictive performance with an AUC of 0.724 for type 4b acute myocardial infarction induced by very late stent thrombosis.

Key Points

The aim is to analyze risk factors related to type 4b acute myocardial infarction caused by very late stent thrombosis and develop a predictive model.
Analyzed patients with a history of coronary stent implantation and AMI after 1 year.
Classified patients into VLST-4b-AMI group and de novo AMI group based on angiography results.
Used logistic regression and Lasso regression to identify factors associated with VLST-4b-AMI.
Developed predictive models using machine learning techniques including XGBoost, SVM, RF, and KNN.
Identified 12 key risk factors associated with VLST-4b-AMI.
XGBoost model showed highest predictive performance (AUC = 0.724).
KNN followed with AUC = 0.698, and RF had AUC = 0.669.
SVM model had the lowest performance (AUC = 0.641).

Study Design

Type

Cohort (n=519)

Multicenter

Structured PICO

Can machine learning models accurately predict type 4b acute myocardial infarction induced by very late stent thrombosis in patients with prior stent implantation?

Population

519 patients with acute myocardial infarction (AMI) admitted to the Coronary Care Unit who had a history of coronary stent implantation >1 year prior to AMI occurrence and underwent coronary angiography. Mean age 65.65, 76.1% male. Single-center cohort in China.

Outcome

Predictive performance for type 4b acute myocardial infarction caused by very late stent thrombosis (VLST-4b-AMI), evaluated using the area under the curve (AUC)

An XGBoost-based machine learning model can moderately predict the occurrence of very late stent thrombosis-induced type 4b AMI in patients with prior stent implantation.

Main Result

Effect estimate: null (95% CI 0.640-0.809)

p-value: p=null

Limitations

Limited use of intravascular imaging examinations in the VLST-4b-AMI group (22/339).
Lack of external validation of the predictive model.

Abstract

To analyze the risk factors for type 4b acute myocardial infarction (AMI) caused by very late stent thrombosis (VLST) and develop a predictive model using machine learning techniques. Patients who had a history of coronary stent implantation and developed AMI more than 1 year later, and had undergone coronary angiography were included. Based on the presence of VLST on coronary angiography, patients were classified into the VLST-4b-AMI group and the de novo AMI group. Data including the first coronary stent implantation, drug treatment, baseline hospitalization data, and coronary angiographic findings were collected and compared between the two groups. Logistic regression and Lasso machine learning were used to identify risk factors for VLST-4b-AMI, and a predictive model was developed using machine learning techniques, followed by performance evaluation. Univariate logistic regression analysis identified risk factors and protective factors for VLST-4b-AMI. Lasso regression selected 12 variables closely associated with VLST-4b-AMI occurrence. Based on these risk factors, predictive models were developed using eXtreme Gradient Boosting (XGBoost), support vector machine (SVM), random forest (RF), and k-nearest neighbors (KNN). The XGBoost model demonstrated the best predictive performance (AUC = 0.724, 95% CI: 0.640–0.809), followed by KNN (AUC = 0.698, 95% CI: 0.614–0.783), and RF (AUC = 0.669, 95% CI: 0.574–0.763). SVM had the lowest performance (AUC = 0.641, 95% CI: 0.547–0.736). The XGBoost-based machine learning model showed the best predictive performance for VLST-4b-AMI, offering a promising tool for early prediction of this high-risk type of AMI.

Mark Helpful

Bookmark

Relay

View Full Paper