Variables Associated With Shunt Failure in Children With Cerebrospinal Fluid Diverting Shunts

Pediatric patients with shunted hydrocephalus are frequently evaluated in the emergency department (ED) with concern for shunt failure. Because shunt failure is potentially life-threatening, accurate diagnosis is essential. However, neuroimaging imperfectly detects children in failure, with reported sensitivities of 53% to 92% for computed tomography and 51% to 59% for magnetic resonance imaging. Objectives We sought to identify clinical characteristics that would negatively predict shunt failure, thus potentially obviating the need for further diagnostic workup or extended periods of observation. We hypothesized that viral symptoms and a patient history of epilepsy or chronic headaches would be negative predictors of shunt failure. Methods Data were retrospectively collected for children 19 years or younger with a cerebrospinal fluid diverting shunt in their medical history or problem list who underwent neuroimaging during an ED visit from March 2008 to September 2016. Patients were defined as having shunt failure if they required surgical exploration for shunt revision within 7 days of the ED visit. Descriptive statistics were used for patient demographics, current symptoms, and historical features. We conducted a logistic regression analysis to determine which characteristics were associated with the odds of shunt failure and used binary recursive partitioning to determine if there were features or a combination of features that were able to accurately classify patients without shunt failure. Results There were 606 visits by 277 patients during this interval, 34% of whom were experiencing shunt failure. Variables found to be significantly predictive of shunt failure were revision within the prior 6 months, vomiting, personality changes, family opinion of shunt failure, and cranial nerve palsies. Viral symptoms and a history of epilepsy or chronic headaches were not predictive of shunt failure. Binary recursive partitioning identified family opinion and personality changes as predictive of shunt failure, with a sensitivity of 72.2% (95% confidence interval, 65.5%-78.2%) and specificity of 46.6% (95% confidence interval, 41.7%-51.7%). Conclusions Although certain clinical and historical features have modest predictive value in children with shunted hydrocephalus, these factors are insufficiently sensitive to exclude shunt failure, arguing for liberal neuroimaging and extended observation.