Prognosis prediction in traumatic brain injury patients using machine learning algorithms.

Area of Science:

• Neurology
• Medical Informatics
• Biostatistics

Background:

• Predicting outcomes for traumatic brain injury (TBI) patients presents significant global challenges.
• Accurate prognostic models are crucial for effective TBI patient management and resource allocation.

Purpose of the Study:

• To develop and evaluate machine learning (ML) algorithms for predicting TBI treatment outcomes.
• To identify the most influential demographic, laboratory, imaging, and clinical features for outcome prediction.

Main Methods:

• Utilized data from 1653 TBI patients admitted to a tertiary trauma center.
• Employed machine learning algorithms, including Random Forest (RF) and Decision Tree (DT), with ten-fold cross-validation.
• Evaluated predictive performance for both short-term (in-hospital mortality) and long-term survival.

Main Results:

• The motor component of the Glasgow Coma Scale, pupil condition, and cistern status were key predictors of in-hospital mortality.
• Patient age emerged as a critical factor for predicting long-term survival.
• The RF algorithm excelled in predicting short-term mortality, while the Generalized Linear Model (GLM) achieved 82.03% accuracy for long-term survival prediction.

Conclusions:

• Machine learning models demonstrate significant potential for predicting TBI patient survival.
• Specific clinical and demographic features serve as reliable markers for short- and long-term outcome prediction.
• Further development of ML approaches can enhance prognostic accuracy in TBI care.