By John Gever
Not all children with severe brain injuries need to be monitored for subclinical seizures, researchers said here, which means that resources can be focused on those at the highest risk.
Victims of abuse, those younger than 2, and those with bleeding within the brain rather than only in the epidural compartment are the pediatric ICU patients most likely to show significant seizure activity that should be detected and treated, said Rajsekar Rajaraman, MD, of the University of California Los Angeles (UCLA).
A separate study by many of the same investigators also found that, in a broader range of pediatric brain injury cases, risk of seizures could be predicted with "fair-to-good" accuracy on the basis of clinical characteristics that would be recorded routinely at admission.
Both studies were reported at the American Epilepsy Society's annual meeting here.
A senior author on both studies, Nicholas Abend, MD, of Children's Hospital of Philadelphia, said at an AES press briefing that identifying and treating seizures is important in the pediatric ICU. When seizures are extremely frequent or long-lasting -- and these can easily go without detection in hospitalized children who are unconscious or lethargic -- they significantly increase the likelihood of poor short- and long-term outcomes.
Such seizures can only be detected via continuous EEG monitoring, Abend explained, which also requires interpretation from trained electroneurologists.
Another investigator in the studies, UCLA's Jason Lerner, MD, noted that children may appear to be napping peacefully while actually undergoing continuous seizures.
Although it would be desirable to perform intense monitoring on all pediatric cases involving head trauma, that is not feasible at most centers, Abend said. He said the field could benefit from risk-stratification models that would allow the care team to track only those patients at the highest risk for damaging subclinical seizures.
Such models, he added, could be tailored to meet the needs of individual centers on the basis of their patient mix, staffing, and other factors.
In a platform session at AES, Rajaraman described one approach to developing such a model. He and colleagues collected data on 135 consecutive pediatric patients (ranging in age from infant to late adolescent) with traumatic brain injury who were treated in ICUs at UCLA and at Children's Hospital of Colorado in Denver. These children had continuous EEG monitoring for detecting subclinical seizures.
They found that all such seizures occurred in children younger than 2 and in those with intradural bleeding, and that the vast majority also involved abusive head trauma. Rajaraman and colleagues then sought to validate these associations in a separate cohort of 44 pediatric ICU patients with head injuries treated at Children's Hospital of Philadelphia. The same patterns were seen.
Across both cohorts, 81% of those with subclinical seizures were determined to have been victims of abusive head trauma, whereas the prevalence of such trauma in all the patients was 25%. Abend said it was uncertain why abusive trauma should be such a strong predictor of these seizures, but speculated that "shaken baby syndrome" -- the most common form of abuse of infants and toddlers -- may produce fundamentally different injuries in the brain compared with falls and car accidents.
Also, such abuse is often chronic, such that the episode that brings a child to the hospital is only the latest in a series of abusive incidents.
The other study, led by Abend, was aimed at producing a predictive model yielding a risk index score that pediatric centers could use to identify critically ill children who could benefit the most from continuous EEG monitoring. It was based on clinical information to which the attending neurologist would have ready access: age, seizure etiology, presence of clinical seizures prior to beginning continuous EEG, initial EEG background category, and interictal discharge category.
Data to design the model were drawn from a database of 336 patients from 11 centers, and then tested against a separate validation dataset of 222 patients treated at Children's Hospital of Philadelphia.
Normalized scores in the model could range from 0 to 1.0, and Abend and colleagues examined the sensitivity and specificity of various cutoffs. When set at 0.10 in the validation cohort, sensitivity was 86% but sensitivity was only 58% -- the high sensitivity meant that 43% of patients would be identified as candidates for continuous monitoring. At the other end, a cutoff of 0.45 reversed the sensitivity and specificity percentages to 19% and 97%, respectively, such that only 5% of patients would be assigned to monitoring.
Abend said the beauty of this approach is that an individual center could choose its own optimal cutoff depending on the resources it has available to monitor multiple patients at one time. A well-equipped and staffed ICU could thus opt for high sensitivity whereas one with more limited resources could be more restrictive.