Prediction of Mortality in Intubated Patients Following Admission to the Intensive Care Unit After an Emergency Room Visit: A Retrospective Cohort Study of Machine Learning Techniques Using Electronic Medical Records

Abstract

Objectives: The purpose of this study is to develop a mortality prediction model for Intensive Care Unit (ICU) patients following endotracheal intubation using machine learning and identify key predictors of outcomes. Methods: A retrospective cohort study analysis was conducted using electronic medical records of 1,229 adult patients who were admitted to the ICU through the Emergency Department (ED) from January 2018 to December 2022. The collected data included general characteristics, blood test results at the time of ED and ICU admission, vital signs, the Braden scale, the Acute Physiology and Chronic Health Evaluation (APACHE) II score, and the duration of stay in both the ED and ICU. A comparison of five machine learning models (Logistic Regression, Decision Tree, Support Vector Machine, Random Forest, and Xtreme Gradient Boosting (XGBoost)) was performed. Model performance was evaluated using stratified k-fold cross-validation, and key metrics were reported as mean ± standard deviation. The bestperforming model was further analyzed using SHapley Additive exPlanations (SHAP) to ensure interpretability. Results: The Logistic regression analysis revealed that intubation time, time to transfer to the ICU after intubation, duration of stay in the ICU, total length of hospital stay, lactic acid levels in both the ED and ICU, APACHE II scores, and oxygen saturation significantly influenced mortality. Among the five machine learning models compared, the XGBoost model showed the highest predictive performance based on stratified k-fold cross-validation. SHAP analysis of the XGBoost model identified Total Length of Stay (T_LOS), ICU Length of Stay (I_LOS), and the APACHE II score as the most influential variables for predicting outcomes. Conclusions: The XGBoost model demonstrated high accuracy in predicting mortality. The combination of this high-performing model with SHAP analysis provides a powerful tool for clinical decision-making, offering both predictive accuracy and transparent, patient-specific interpretations. Implications for Clinical Practice: In managing patients in the ED and ICU, total length of stay ICU length of stay and APACHE II score can be considered to predict patient prognosis and develop tailored treatment plans.