The use of machine learning approaches for mortality prediction and classification in traumatic brain injury: a literature review.

OBJECTIVE

This systematic review analyzed machine learning (ML) techniques for predicting and classifying mortality in traumatic brain injury (TBI) victims.

METHODS

The PICOS (People, Intervention, Comparison, Outcome, Study) strategy was used for the research question. The included studies were indexed in the Virtual Health Library and the National Library of Medicine, written in English, Portuguese, or Spanish, and focused on TBI-related mortality. Exclusions comprised animal studies, molecular research, reviews, and unavailable full texts.

RESULTS

Of 1,752 screened studies, 36 met inclusion criteria. Most publications emerged in 2023 (n = 10, 27.8%), with the highest number of publications belonging to Asian countries (n = 12, 33.3%). The general mean age of TBI was 50.5 years for male adults and the main causes were motor vehicle accidents and falls. The main predictors of mortality were low Glasgow Coma Scale score (n=17, 12.9%) and age (n=14, 10.6%). The best machine learning techniques to predict mortality were Support Vector Machine, Logistic Regression, Random Forest and eXtreme Gradient Boosting. The areas under the curve (AUC) ranged from 0.757 to 0.968.

CONCLUSION

Machine learning techniques seem to be useful and promising tools for predicting mortality in TBI victims, and thus help in clinical decisions.