Predicting in-hospital mortality in ICU patients with lymphoma using machine learning models

Area of Science:

• Computational biology
• Medical informatics
• Oncology

Background:

• Lymphoma is a critical condition with high mortality, often necessitating intensive care unit (ICU) admission.
• Existing risk stratification tools (SOFA, APACHE) have limitations in capturing complex clinical interactions for lymphoma patients.
• Machine learning (ML) offers advanced predictive capabilities for patient outcomes by analyzing extensive clinical datasets.

Purpose of the Study:

• To develop and validate ML models for predicting in-hospital mortality in ICU lymphoma patients.
• To enhance risk stratification and clinical decision-making for this high-risk population.
• To utilize data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database for model development.

Main Methods:

• Retrospective cohort study using MIMIC-IV database data.
• Lasso regression for significant risk factor identification (e.g., BUN, platelets, PT).
• Development and comparison of 15 ML models (including CatBoost Classifier, logistic regression, random forest, gradient boosting, neural networks) using ROC/AUC analysis and SHAP values.

Main Results:

• 1591 lymphoma patients included; 342 (21.5%) in-hospital deaths.
• CatBoost Classifier achieved the highest predictive performance (AUC = 0.7766).
• Key mortality predictors identified: BUN, platelets, and PT; SHAP analysis provided individualized risk insights.

Conclusions:

• ML models, especially CatBoost Classifier, effectively predict in-hospital mortality in ICU lymphoma patients.
• These models surpass traditional methods in accuracy and provide crucial risk stratification insights.
• External validation and clinical implementation are recommended for improving outcomes in this patient group.