What type of study is this?

October 2, 2025

Machine Learning Models for Predicting 28-Day Mortality in Gastrointestinal Bleeding with Acute Kidney Injury: A MIMIC-IV-Based Study

Key Points

XGBoost model achieved an area under the curve of 0.8027 for predicting 28-day mortality in patients with GIB-AKI.
Five machine learning algorithms were developed and compared, enhancing the interpretability of mortality predictors using SHAP.
The study utilized the MIMIC-IV database including 1,890 adult patients to build and validate these prediction models.
Results highlight the potential of machine learning to improve early mortality prediction and critical care management for GIB-AKI patients.

Abstract

Abstract Background: Gastrointestinal bleeding (GIB) is a common life-threatening condition in the digestive system that is frequently complicated by acute kidney injury (AKI), substantially increasing mortality and healthcare burden. To date, no precise tool exists for early prediction of short-term outcomes in patients with concurrent GIB and AKI (GIB-AKI).We conducted a retrospective cohort study aims to develop and validate machine learning (ML) models for predicting 28-day mortality . Methods: This retrospective cohort study was based on the MIMIC-IV database, including patients with first ICU admission who met criteria for GIB and AKI .From 64 clinical variables, we applied Least Absolute Shrinkage and Selection Operator (LASSO) regression and the Boruta algorithm to identify the key predictors of 28-day mortality. Five ML algorithms were employed to construct the prediction model. Model performance was evaluated based on identification, calibration, and clinical application.SHapley Additive exPlanations (SHAP) were used to interpret key variables influencing mortality in the best model. Results: A total of 1,890 adult GIB-AKI patients were included in this study. Five machine learning (ML) algorithms—Light Gradient Boosting Machine (LightGBM), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Logistic Regression, and Decision Tree—were developed and compared. Among all models, XGBoost demonstrated the best discriminative performance, achieving an area under the receiver operating characteristic curve (AUC) of 0.8027 in the validation set. Calibration and decision curve analyses confirmed its superior clinical utility. Furthermore, SHapley Additive exPlanations (SHAP)enhanced model interpretability by illustrating key variables influencing mortality. Conclusion: This study introduces a robust, interpretable ML model for early mortality prediction in GIB-AKI, offering a valuable tool for critical care management and precision medicine practice.

Perguntar à IA

Bookmark