A simple Lactate-GCS model for predicting poor neurological outcome after pediatric traumatic brain injury: a single-center retrospective cohort study

Background Early prediction of poor neurological outcome after pediatric traumatic brain injury (TBI) remains challenging, particularly in the pediatric intensive care unit (PICU). This study aimed to develop and internally validate a simple prediction model. Methods This single-center retrospective cohort study included children with TBI admitted to the PICU between January 2022 and June 2026. Poor neurological outcome was defined as a Pediatric Cerebral Performance Category (PCPC) score greater than 2 at hospital discharge. Candidate predictors were assessed using clinical relevance, univariable analysis, and multicollinearity testing. A multivariable logistic regression model was developed and evaluated using discrimination, calibration, decision curve analysis, and bootstrap internal validation. Predictive performance was compared with models based on the core variables of the IMPACT Core and CRASH basic models, and subgroup and sensitivity analyses assessed robustness. Results Ninety-two children were included, and 32 (34.8%) had poor neurological outcome. Lower admission Glasgow Coma Scale (GCS) score (aOR = 0.28, 95% CI: 0.10–0.54; p < 0.001) and higher arterial lactate level (aOR = 2.09, 95% CI: 1.22–4.62; p = 0.004) were independently associated with poor neurological outcome. The final Lactate-GCS model formula was logit (P) = 5.19–1.24 × GCS + 0.70 × Lactate (P represents the predicted probability of poor neurological outcome), yielding an apparent AUC of 0.97 (95% CI: 0.93–1.00). In exploratory comparisons with locally refitted IMPACT Core and CRASH basic models within our pediatric cohort, the AUCs were 0.91 (95% CI: 0.86–0.97; p = 0.03) for the IMPACT Core model and 0.95 (95% CI: 0.91–0.99; p = 0.27) for the CRASH basic model. After bootstrap internal validation, the model demonstrated good calibration, with a calibration slope of 0.90, calibration-in-the-large of −0.03, a mean absolute error of 0.018 and a mean squared error of 0.00014. Decision curve analysis suggested a positive net clinical benefit across a range of threshold probabilities. Subgroup and sensitivity analyses supported the robustness of the predictive effects. Conclusion In pediatric TBI admitted to the PICU, higher admission arterial lactate levels and lower GCS scores were independently associated with poor neurological outcome at discharge. The simplified Lactate–GCS model showed preliminary predictive ability for early risk stratification in pediatric TBI.