Machine learning-based mortality prediction models for emergency department patients: a comparative analysis.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Prediction Models

Background:

• Accurate prediction of in-hospital mortality in emergency departments (ED) is critical for patient care and resource management.
• This study addresses the need for improved predictive tools in the ED setting.

Purpose of the Study:

• To develop and compare various machine learning models for predicting in-hospital mortality among emergency department patients.
• To identify key clinical and laboratory parameters that contribute to mortality prediction.

Main Methods:

• Retrospective analysis of 1,389 emergency department patients.
• Development and comparison of nine machine learning models, including LightGBM and an ensemble voting classifier.
• Evaluation using metrics such as area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity.

Main Results:

• The LightGBM model achieved the highest performance with an AUROC of 0.9605, sensitivity of 78.12%, and specificity of 93.90%.
• Key predictors included serum lactate, Glasgow Coma Scale (GCS), albumin, base excess (BE), and systolic blood pressure (SBP).
• Calibration and decision curve analyses confirmed the clinical utility and accuracy of the models.

Conclusions:

• Machine learning models, especially LightGBM, offer highly accurate mortality prediction for emergency department patients.
• The integration of accessible clinical and laboratory data facilitates early risk stratification and targeted interventions.
• These findings can potentially improve patient outcomes through timely and informed clinical decision-making.