Machine learning-based mortality prediction in critically ill patients with hypertension: comparative analysis,...

Area of Science:

• Critical care medicine
• Biomedical informatics
• Artificial intelligence in healthcare

Background:

• Hypertension is a major risk factor for cardiovascular, cerebrovascular, and renal diseases, increasing mortality in critically ill patients.
• Accurate mortality prediction is crucial for timely interventions in this high-risk population.
• Machine learning (ML) and deep learning (DL) offer advanced tools for analyzing electronic health record (EHR) data.

Purpose of the Study:

• To develop and evaluate ML and DL models for predicting in-hospital mortality in hypertensive patients.
• To assess the fairness and interpretability of these predictive models.
• To leverage the MIMIC-IV critical care dataset for robust model development.

Main Methods:

• Developed Gradient Boosting Machine (GBM), logistic regression, Support Vector Machine (SVM), random forest, Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM) models.
• Utilized comprehensive features from EHRs, including demographics, lab values, vital signs, and comorbidities.
• Evaluated models using 5-fold cross-validation, SHapley Additive exPlanations (SHAP) for feature importance, and demographic parity difference (DPD) and equalized odds difference (EOD) for fairness.

Main Results:

• The GBM model achieved the highest performance (AUC-ROC 96.3%, accuracy 89.4%).
• Key predictors of mortality included Glasgow Coma Scale (GCS) scores, Braden Scale scores, blood urea nitrogen, age, red cell distribution width (RDW), bicarbonate, and lactate.
• Models using top features showed reduced bias (lower DPD and EOD); debiasing techniques improved fairness for models with all features.

Conclusions:

• ML models demonstrate significant potential for predicting mortality in critically ill hypertensive patients.
• Feature selection enhances model interpretability, reduces complexity, and may improve fairness.
• Integrating interpretable and equitable AI tools can support clinical decision-making in critical care.