Development and validation of a prediction model for postoperative cerebral edema progression in emergency traumatic brain injury patients.

To develop and internally validate a clinical prediction model for estimating the risk of postoperative cerebral edema progression in emergency traumatic brain injury (TBI) patients. This study was a single-center, retrospective, observational cohort study. A total of 201 TBI patients who were admitted to our hospital's emergency department and underwent craniotomy between January 2023 and December 2024 were consecutively enrolled. Based on imaging and clinical criteria within 72 hours postoperatively, patients were categorized into a cerebral edema progression group (n = 63) and a non-progression group (n = 138). Using the time-split method, 70% of the patients (n = 141) were assigned to the model development cohort, and 30% (n = 60) to the internal validation cohort. Demographic, clinical, imaging, and laboratory indicators were collected via systematic review of electronic medical records. In the development cohort, univariate logistic regression was first performed to screen potential predictors (P < .1), followed by multivariate logistic regression to identify independent predictors and build the prediction model. Model performance was assessed by the area under the receiver operating characteristic curve (AUC) for discrimination, the Hosmer-Lemeshow test and calibration plot for calibration, and decision curve analysis for clinical utility. Finally, subgroup analysis was conducted to verify the model's stability. Multivariate analysis ultimately identified 7 independent predictors of postoperative cerebral edema progression: increased age (adjusted odds ratio [aOR] = 1.02/yr), lower admission Glasgow Coma Scale score (aOR = 0.86/point), larger volume of cerebral contusion/hematoma (aOR = 1.06/10 mL), greater midline shift (aOR = 1.12/mm), higher maximum intracranial pressure within 24 hours postoperatively (aOR = 1.11/mm Hg), lower serum sodium level within 24 hours postoperatively (aOR = 0.91/mmol/L), and elevated C-reactive protein level (aOR = 1.03/10 mg/L). The prediction model constructed based on these factors demonstrated excellent discrimination in the development cohort (AUC = 0.901, 95% CI: 0.849-0.954) and maintained good predictive performance in the internal validation cohort (AUC = 0.875, 95% CI: 0.782-0.948). This study successfully developed and internally validated a prediction model incorporating clinical, imaging, and laboratory indicators. This model can accurately and stably predict the risk of postoperative cerebral edema progression in emergency TBI patients, facilitating the early identification of high-risk patients and guiding individualized intervention.