Development and internal validation of a clinical prediction model for postoperative venous thromboembolism after neurosurgical intervention in patients with traumatic brain injury.

Postoperative VTE occurred in 14.1% of TBI patients who underwent neurosurgical intervention in this retrospective cohort.

• The cohort enrolled 1,450 TBI patients who underwent surgery between 2016 and 2024.
• 205 patients (14.1%) developed postoperative VTE.
• The study was a retrospective cohort design at a single reported dataset with internal validation only.

Seven predictors were selected by Recursive Feature Elimination for the prediction model.

• The predictors were age, Barthel index, ASA classification, multiple surgeries, general anesthesia, limb AIS score, and serum magnesium.
• All seven are described as routinely available clinical variables.
• Feature selection used Recursive Feature Elimination.

Logistic regression was selected as the final model after comparing six machine learning algorithms, and it showed good discrimination and calibration.

• The LR model achieved an AUC-ROC of 0.868 (95% CI: 0.815-0.920).
• Calibration was good, with a Brier score of 0.085.
• Performance was evaluated using AUC-ROC, calibration, and decision curve analysis.

The model discriminated better than the Caprini score for postoperative VTE in this population.

• The Caprini score had an AUC of 0.686.
• The difference in AUC was 0.182 in favor of the new model.
• The net reclassification improvement (NRI) was 0.559.

The authors provide a nomogram and a web-based calculator to support clinical use of the model.

• The model is described as interpretable.
• Conventional risk assessment tools were not developed for neurosurgical trauma populations, which limits their applicability.
• Validation was internal; the paper describes the performance as promising.