Early clinical and laboratory indicators for differentiating infectious from autoimmune encephalitis in pediatric patients: A single-center retrospective cohort study.

BACKGROUND: Pediatric encephalitis is associated with significant morbidity. Early recognition of autoimmune encephalitis (AE) is critical for timely immunotherapy and improved outcomes. OBJECTIVES: To externally validate a previously published adult encephalitis prediction model in a pediatric cohort and identify predictors of pediatric AE. METHODS: This retrospective single-center study included children 90 days to 21 years with encephalitis between January 2016 and September 2025. Clinical and laboratory data were extracted using electronic health record. External validation used published coefficients. Discrimination and calibration were assessed using receiver operating characteristic analysis and slope estimation. An exploratory multivariable logistic regression model was constructed. RESULTS: One hundred and twenty-six patients were included (92 autoimmune, 34 infectious). Patients with AE were significantly older (median age 14 vs. 5.5 years, p < .001). Infectious encephalitis was associated with higher serum white blood cell (WBC) (10.8 vs. 8.0 k/µL; p = .033) and cerebrospinal fluid (CSF) WBC (11 vs. 1 cells/µL; p < .001). The model showed acceptable discrimination (area under the receiver operating characteristic curve 0.774; 95% confidence interval [CI] 0.635-0.913) and calibration (slope 1.14; 95% CI 0.45-1.83). In our exploratory model (n = 80), CSF protein ≥30 mg/dL, was associated with lower odds of AE (odds ratio [OR] 0.24; 95% CI 0.07-0.80; p = .021), while increasing age was associated with higher odds (OR 1.13/year; 95% CI 1.01-1.26; p = .031). CONCLUSIONS: Predictors derived from an adult encephalitis model may retain some relevance in pediatrics despite physiologic differences. Specific pediatric predictors of AE included CSF protein <30 and increasing age, highlighting the need for pediatric-specific prediction tools. Multicenter studies are needed to validate models and refine thresholds.