An Interpretable Machine Learning Model for Predicting Early Neurological Deterioration Following Posttraumatic Acute...

Area of Science:

• Neuroscience
• Medical Informatics
• Trauma Surgery

Background:

• Diffuse brain swelling (DBS) is a major cause of intracranial hypertension and early neurological deterioration (END) after traumatic brain injury.
• Accurate prediction of END is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting END in adult patients with traumatic DBS.
• To identify key predictors of END in this patient population.

Main Methods:

• Retrospective collection of clinical data from 208 adult patients with traumatic DBS.
• Identification of END predictors using regression analysis.
• Training and evaluation of six ML algorithms, including extreme gradient boosting, using AUROC, Brier score, and decision curve analysis.
• Internal cross-validation and external validation in a separate cohort.
• Interpretation of the optimal model using Shapley Additive Explanations.

Main Results:

• The incidence of END was 38.0% (79 patients).
• Key predictors of END included subdural hemorrhage, severe traumatic subarachnoid hemorrhage, hemoglobin, and fibrinogen levels.
• The extreme gradient boosting model demonstrated superior performance with an AUROC of 0.879, excellent calibration, and clinical utility.
• The model showed good generalizability across internal cross-validation and external validation cohorts.

Conclusions:

• The developed ML model shows significant clinical potential for accurately predicting END in patients with traumatic DBS.
• Further multicenter external validation is recommended prior to widespread clinical implementation.