Explainable machine learning model for predicting traumatic brain injury-induced coagulopathy in elderly patients: A...

Traumatic brain injury-induced coagulopathy (TBI-IC) in the elderly is a severe complication of traumatic brain injury (TBI) that leads to unfavorable outcomes. Currently, there are no reliable machine learning (ML) models available for early identification. This study comprehensively assessed routine clinical characteristics at admission to develop a robust, generalizable, and explainable ML model for estimating TBI-IC risk in the elderly and to identify modifiable factors.

This study included two cohorts: a derivation cohort (n = 484) from the Medical Information Mart for Intensive Care IV (MIMIC-IV) and a validation cohort from the eICU Collaborative Research Database (eICU-CRD) (n = 788). Feature selection utilized the Boruta algorithm, followed by the implementation of a distinctive ML framework incorporating 12 ML algorithms to establish a consensus prediction model (TBI-IC index). The model and feature variable assessments were conducted using multiple analytical methods. Model interpretation and feature quantification relied on the Shapley additive explanation (SHAP) methodology for visualization purposes.

Through Boruta algorithm selection, 17 characteristics were identified across two cohorts and incorporated into 12 ML methodologies, generating 113 permutations and an optimal algorithm for identifying TBI-IC. The TBI-IC index demonstrated strong diagnostic capabilities, achieving a mean area under curve (AUC) of 0.801 across both cohorts, along with notable discriminatory power, model fit, and clinical utility. Multivariate logistic regression and subgroup evaluations confirmed the stability and broad applicability of the TBI-IC model. SHAP explains the importance of ranking features and visualizes global and individual TBI-IC risk predictions. Restricted cubic spline (RCS) regression and threshold effect analysis suggested a nonlinear link between the model features and TBI-IC, and generated inflection points for the features.

An optimized explainable model (TBI-IC index) incorporating several modifiable parameters was established and confirmed to deliver an economical, readily available, and accurate diagnostic tool, along with preventive guidance for TBI-IC among elderly patients, with potential applications in TBI clinical management.