Abstract
Objective: To investigate the early predictive value of neutrophil-to-lymphocyte ratio (NLR) for surgical requirement in infants with traumatic subdural effusion (TSE).Methods: A single-center retrospective cohort study was conducted, enrolling 629 patients aged ≤6 months with isolated traumatic brain injury. NLR was calculated based on a complete blood count within 48 hours post-injury. Logistic and LASSO regression were used to screen for independent risk factors. Machine learning was employed to impute missing values for continuous variables. The association between NLR and subdural burr hole drainage was analyzed using the area under the receiver operating characteristic curve (AUC). Combined prediction models were constructed.Results: 52 cases (8.27%) required surgical intervention. Injury mechanism, seizure (P<0.001), shortened APTT (original P=0.02, after imputation P=0.03), and elevated NLR (original P=0.01, after imputation P=0.004) were independent risk factors for surgical requirement. The optimal cutoff value for NLR was 0.80 (original apparent AUC=0.73, after imputation cross-validated AUC=0.69). The surgery rate in the high-NLR group (≥0.80) (13.70%) was significantly higher than that in the low-NLR group (4.18%) and was more frequently associated with traffic accident injury (20.00% vs. 9.75%) and seizure (16.30% vs. 6.41%). The model combining NLR with injury mechanism, seizure, and APTT significantly improved predictive performance (original apparent AUC=0.84, after imputation cross-validated AUC=0.78).Conclusion: NLR, as a low-cost and available biomarker, can effectively predict surgical requirements in infants with TSE. When combined with clinical indicators, it significantly enhances risk stratification capability. Machine learning handling missing data can improve the reliability of predictive models.