The Role of Artificial Intelligence in Predicting Mortality in Patients with Variceal Gastrointestinal Bleeding: A Retrospective Study
Introduction: Variceal gastrointestinal bleeding is a severe complication of portal hypertension with high in hospital mortality, Objective: To develop and validate an artificial intelligence (AI) model to predict in-hospital mortality in patients admitted for variceal gastrointestinal bleeding associated with portal hypertension (PH). Methods: A single-center retrospective study was conducted between 2022 and 2025 on 100 patients with cirrhosis. A supervised learning algorithm of the Random Forest type was trained (90% training, 10% testing) using Python (Pandas, Scikit-learn). To compensate for the small sample size, k-fold cross-validation was applied. Results: Overall mortality was 22%. The model demonstrated an area under the ROC curve (AUC) of 0.756, a sensitivity of 70%, and a specificity of 66% (overall accuracy: 66%). On the test sample, accuracy reached 80% (specificity of 100%), outperforming the Glasgow-Blatchford and Rockall scores, which are particularly limited by their low variability. The critical variables identified are MELD-Na, creatinine, and the Child-Pugh score. Conclusion: Artificial intelligence enables superior prognostic stratification compared to nonspecific scores. This exploratory model must now undergo multicenter external validation to confirm its role as a clinical decision-support tool.