Machine Learning Algorithm Predicts Mortality Risk in Intensive Care Unit for Patients with Traumatic Brain Injury

Area of Science:

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

Background:

• Existing mortality prediction tools for moderate to severe traumatic brain injury (TBI) lack comprehensive machine learning approaches.
• A well-established algorithm using diverse features and machine learning methods for predicting outcomes in intensive care unit (ICU) TBI patients is needed.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting mortality in ICU patients with TBI.
• To compare the performance of machine learning models against traditional scoring systems (APACHE II, SOFA).

Main Methods:

• Retrospective analysis of 2260 TBI patients admitted to the ICU (January 2016 - December 2021).
• Utilized four machine learning models with 42 features, exploring various feature combinations.
• Assessed predictive performance using Area Under the Curve (AUC) from Receiver Operating Characteristic (ROC) curves and validated with the Delong test.

Main Results:

• Machine learning models achieved AUCs ranging from 0.877 to 0.921.
• Random forest with 22 features yielded the highest AUC (0.921).
• Machine learning models demonstrated superior predictive performance compared to APACHE II and SOFA scores.

Conclusions:

• LightGBM model shows superior predictive accuracy for TBI patient mortality compared to APACHE II and SOFA.
• Key predictive features are available on the first day of ICU admission.
• Integration into clinical platforms can provide immediate prognosis, aiding clinician-family communication.