Prediction of Mortality in Surgical Intensive Care Unit Patients Using Machine Learning Algorithms.

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Accurate prognosis prediction for in-hospital patients is crucial but challenging.
• Identifying critically ill patients at high risk of in-hospital death is a significant clinical need.

Purpose of the Study:

• To evaluate the accuracy of machine learning algorithms in predicting in-hospital death for critically ill patients.
• To identify key factors contributing to the predictive power of these algorithms.

Main Methods:

• Utilized medical data from 1,384 Surgical Intensive Care Unit (SICU) patients.
• Compared three machine learning algorithms (including decision tree and neural network) to predict in-hospital death.
• Defined outcome as unexpected postoperative mortality within 30 days or same hospital stay.

Main Results:

• Machine learning algorithms achieved high accuracy in classifying mortality.
• The decision tree model demonstrated superior performance (AUC = 0.96) compared to the neural network (AUC = 0.80).
• Key predictors identified include serum albumin, total prenatal nutritional intake, and peak dopamine dose.

Conclusions:

• Machine learning, particularly decision tree methods, can accurately predict hospital mortality in SICU patients.
• These algorithms offer structured and explainable decision-making pathways for prognosis.
• The findings highlight the potential of AI in improving critical care patient management.