Machine Learning-Based Mortality Prediction for Acute Gastrointestinal Bleeding Patients Admitted to Intensive Care Unit.

Abstract

Objective: The study aimed to develop machine learning (ML) models to predict the mortality of patients with acute gastrointestinal bleeding (AGIB) in the intensive care unit (ICU) and compared their prognostic performance with that of Acute Physiology and Chronic Health Evaluation II (APACHE-II) score.

Methods: A total of 961 AGIB patients admitted to the ICU of Renmin Hospital of Wuhan University from January 2020 to December 2023 were enrolled. Patients were randomly divided into the training cohort (n = 768) and the validation cohort (n = 193). Clinical data were collected within the first 24 h of ICU admission. ML models were constructed using Python V.3.7 package, employing 3 different algorithms: XGBoost, Random Forest (RF) and Gradient Boosting Decision Tree (GBDT). The area under the receiver operating characteristic (ROC) curve (AUC) was used to evaluate the performance of different models.

Results: A total of 94 patients died with an overall mortality of 9.78% (11.32% in the training cohort and 8.96% in the validation cohort). Among the 3 ML models, the GBDT algorithm demonstrated the highest predictive performance, achieving an AUC of 0.95 (95% CI 0.90-0.99), while the AUCs of XGBoost and RF models were 0.89 (95% CI 0.82-0.96) and 0.90 (95% CI 0.84-0.96), respectively. In comparison, the APACHE-II model achieved an AUC of 0.74 (95% CI 0.69-0.87), with a specificity of 70.97% (95% CI 64.07-77.01). When APACHE-II score was incorporated into the GBDT algorithm, the ensemble model achieved an AUC of 0.98 (95% CI 0.96-0.99) with a sensitivity of 85.71% and a specificity up to 95.15%.

Conclusions: The GBDT model serves as a reliable tool for accurately predicting the in-hospital mortality for AGIB patients. When integrated with the APACHE-II score, the ensemble GBDT algorithm further enhances predictive accuracy and provides valuable insights for prognostic evaluation.