RESUMO
OBJECTIVE: The study aimed to validate admission clinical and radiographic features of pediatric patients with traumatic epidural hematoma (EDH) that lead to safe observation. METHODS: A Level I trauma center radiology and electronic medical record databases were retrospectively queried for pediatric patients with EDH on CT scan between 1/1/2016 and 10/1/2016. Patient imaging, treatment and outcome variables were abstracted. Characteristics of the cohort were compared to an external cohort used to develop prediction rules for surgical intervention. External validity of the prediction rules was assessed. RESULTS: 195 eligible subjects were included in the study, 37 of which failed observation and required surgery while 158 underwent successful observation. The surgical cohort had significantly thicker (p < .001) and higher volume (p < .001) EDH, increased midline shift (p < .001) and higher likelihood of mass effect (p < .001). There was significantly higher residual neurologic deficit rate (54% vs 23%, p < .001) and hospital mortality (5% vs 0%, p = .035) amongst the surgical group. There were significant differences in patient demographic, clinical and imaging characteristics between the internal and external cohorts. The predictive rules externally developed yielded positive predictive value of 97.7% (95% CI = 93.3-99.5%), negative predictive value of 24.5% (95% CI = 16.2-34.4%), specificity of 88.5% (95% CI = 69.9-97.6%), and sensitivity of 63.8% (95% CI = 56.6-70.5%) for successful observation. CONCLUSION: The current study validates previously developed prediction rules for safe observation of pediatric EDH in a cohort with distinct characteristics from the external cohort. Specifically, patients with no mass effect, EDH volume <15 ml and no neurological deficits are less likely to fail observation. ADVANCES IN KNOWLEDGE: The current study validates prediction rules for safe observation of pediatric EDH in a distinct pediatric cohort that provides further support to conservative management in these circumstances.