What question did this study set out to answer?

To develop machine learning models for predicting in-hospital death among patients using intra-aortic balloon pumps during cardiac surgery.

April 23, 2026Journal of Cardiothoracic Surgery0 citationsOpen Access

Application of supervised machine learning algorithms to predict in-hospital death risk in patients receiving intra-aortic balloon pump therapy during the perioperative period of cardiac surgery

CYChangqing YangNantong University HXHan XieObstetrics and Gynecology Hospital of Fudan University HZHong ZhouWuhan University

Key Points

To develop machine learning models for predicting in-hospital death among patients using intra-aortic balloon pumps during cardiac surgery.
Data from 297 patients was analyzed, categorized into survival and death groups.

Structured PICO

Can machine learning algorithms accurately predict the risk of in-hospital death in patients receiving intra-aortic balloon pump therapy during the perioperative period of cardiac surgery?

Population

297 adult patients receiving intra-aortic balloon pump (IABP) support therapy during the perioperative period of cardiac surgery (isolated CABG, isolated cardiac valve surgery, or combined surgeries), mean age 65.06, 79.46% male, based in China. Indications for IABP included cardiogenic shock, unstable hemodynamics, low cardiac output, and progressive drop in blood pressure.

Outcome

In-hospital mortality (defined according to discharge status)hard clinical

An XGBoost machine learning model utilizing perioperative clinical variables can accurately predict in-hospital mortality in high-risk patients requiring IABP support during cardiac surgery.

Abstract

This study aimed to construct machine learning (ML) models to predict in-hospital death in patients with intra-aortic balloon pump (IABP) support during cardiac surgery. 297 patients were enrolled and divided into the survival and death groups. Eight ML models were developed and validated, with performance evaluated using the area under the receiver operating characteristic curve (AUC). Feature importance was analyzed using Shapley Additive exPlanations (SHAP). 70 (23.57%) patients were in the death group. The XGBoost model achieved the best predictive ability, with an AUC of 0.91 in the training set and 0.83 in the testing set. SHAP analysis identified low-density lipoprotein cholesterol (LDLC), intensive care unit stay on postoperative day 5 (ICU 5d), blood urea nitrogen (BUN), cardiac troponin T (cTnT), continuous renal replacement therapy (CRRT), body mass index (BMI), mechanical ventilation support on postoperative day 5 (MV 5d), and N-terminal pro-B-type natriuretic peptide (NT-pro BNP) as the top predictors. SHAP force plots demonstrated individualized prediction. The XGBoost model predicts the risk of in-hospital death in patients, possibly contributing to improving perioperative management and reducing mortality.

问 AI

Bookmark

View Full Paper