SMOTE-random forest model enhances prediction of in-hospital mortality in hypertensive patients: a retrospective study based on the MIMIC-IV database

Background Hypertension is a major contributor to cardiovascular mortality, yet predicting in-hospital mortality remains challenging due to class imbalance in clinical data. Purpose To develop and validate a machine learning model for predicting in-hospital mortality in hypertensive patients while addressing data imbalance using R version 4.5.0. Methods We analyzed 2,493 hypertensive patients from the MIMIC-IV database (v3.0) with a mortality rate of 14.6%. Data were split into training (80%) and testing (20%) sets. Four models were compared: Base Elastic Net, SMOTE-Elastic Net, SMOTE-Random Forest (SMOTE-RF), and SMOTE-XGBoost. SMOTE was applied to the training set to mitigate class imbalance. Model performance was evaluated using AUC-ROC, sensitivity, and specificity. Results SMOTE-RF achieved the highest discriminative performance (AUC: 0.827; 95% CI: 0.784–0.870) (Figure 1a). At an optimized threshold of 0.255, the model demonstrated a sensitivity of 79.5% and specificity of 74.6%, substantially outperforming the imbalanced model (sensitivity: 15.1%) (Figure 2). Key predictors included white blood cell count, aspartate aminotransferase, international normalized ratio, and platelet count. (Figure 1b) Conclusion The SMOTE-RF model significantly improves the prediction of in-hospital mortality in hypertensive patients by effectively addressing class imbalance, enabling early identification of high-risk individuals for timely intervention.