A Clinical and Biomarker-Based Model for Predicting in-Hospital Mortality in Patients Undergoing Extracorporeal...

Area of Science:

• Critical Care Medicine
• Biostatistics
• Machine Learning in Healthcare

Background:

• Extracorporeal membrane oxygenation (ECMO) is a life-support technology with significant in-hospital mortality.
• Accurate risk stratification is crucial for optimizing patient management and resource allocation during ECMO therapy.

Purpose of the Study:

• To develop and validate a predictive nomogram for in-hospital mortality in patients undergoing ECMO.
• To identify key clinical indicators associated with ECMO patient mortality.

Main Methods:

• Retrospective cohort study of 122 ECMO patients for model development and 22 patients for external validation.
• Utilized a two-step Boruta-LASSO feature selection strategy to identify predictors.
• Constructed and evaluated five machine learning models, selecting the best for nomogram development.
• Validated the nomogram using ROC analysis, calibration curves, decision curve analysis, and SHAP interpretability.

Main Results:

• Identified four key predictors of in-hospital mortality: malignant tumor, Lactic Acid, monocyte-to-white blood cell ratio (MWR), and activated partial thromboplastin time-to-albumin ratio (AAR).
• The logistic regression-based nomogram achieved a cross-validated AUC of 0.765 in the primary cohort and 0.838 in the external validation cohort.
• The model demonstrated satisfactory discriminatory ability and acceptable calibration.

Conclusions:

• Malignant tumor, elevated Lactic Acid, decreased MWR, and elevated AAR are significant predictors of in-hospital mortality in ECMO patients.
• The developed nomogram shows potential as an exploratory tool for bedside risk stratification.
• Further validation in larger, prospective, multicenter studies is recommended prior to clinical implementation.