Screening of biomarkers and machine learning prediction in traumatic brain injury

Traumatic brain injury (TBI) is a major cause of disability and mortality worldwide. Accessible blood and hormonal biomarkers may be related to injury severity and cognitive status in patients undergoing neurorehabilitation. This study used explainable machine learning to explore the associations of routine blood parameters and hormones with radiological injury severity and cognitive status in TBI.

In this prospective cross-sectional multicenter study, 154 patients with TBI were enrolled. Blood samples were collected within 1 week after admission to assess routine hematological indices, liver and renal function, blood lipids, and hormone levels. Injury severity was evaluated using the Helsinki CT Score, and cognitive status was assessed using the mini-mental state examination (MMSE). After preprocessing, data were split into training and validation sets at a ratio of 7:3. LASSO regression was used for feature selection, and six machine learning models were developed. Model performance was evaluated using R ², mean squared error, and mean absolute error. SHapley Additive exPlanations were used for interpretation.

LASSO identified eight features for the Helsinki CT Score and four for MMSE. Random forest performed best for the Helsinki CT Score (validation R ² = 0.06), whereas CatBoost performed best for MMSE (validation R ² = 0.103). SHAP analysis indicated that IGF-1 was an important feature in both models. IGF-1 showed a possible nonlinear association with both outcomes.

Routine blood and hormonal biomarkers, particularly IGF-1, may be associated with radiological injury severity and cognitive status in TBI. These findings are exploratory and require validation in larger longitudinal studies.

Chinese Clinical Trial Registry: ChiCTR2300072902. Medical Research Registration Number: MR-11-23-023826.