Abstracts

Rationale: Many studies describe a high incidence of seizures in pediatric ICU populations and show that seizures are more likely to first occur early in recordings. A descriptive statistic (percent of studies with seizures) may not reflect the true probability of a seizure occurring at any point during EEG monitoring since all patients are not monitored for the same duration. With a limited duration of monitoring, some patients experience a seizure while others have monitoring discontinued and can be classified as censored . For censored patients, we are unable to ascertain if a seizure would occur with longer monitoring. Utilizing methodology for censored data may provide a better estimate of the true probability of a seizure occurring during a finite duration of monitoring. Methods: The duration of ICU-EEG monitoring studies and time to first seizure were documented for 463 consecutive pediatric studies completed between 6/12/2010 and 6/6/2013 at Children s Healthcare of Atlanta at Egleston. Descriptive statistics were used to compare these cases to previously reported series and calculate the incidence over a fixed duration of monitoring time. Kaplan-Meier survival analysis was used to estimate the probability of recording seizures during a 256 hour period by adjusting for censored cases.Results: The trend of seizures occurring in arbitrary time intervals was similar to previous reports. Cumulative incidence curve shows 28.3% of EEGs demonstrated seizures, and 98% of these events occur by 52 hours. These results are similar to prior reports although with a slightly lower cumulative incidence. Kaplan-Meier estimator suggests that 40.9% of studies would have recorded a seizure if monitored for the full 256 hours. This analysis accounts for case dropout (censored cases) because EEG ended without seizures occurring. The probability that a seizure would be recorded with further monitoring if none occurs by a given time remained substantially higher using the Kaplan Meier conditional probability estimates for all recording durations less than 6.3 days. For example, after 52 hours, only 3 patients 5.6% (3/ 54) experienced a seizure while the KM analysis estimated that approximately 10.6% of patients would have had a seizure had the monitoring continued for the full duration. Conclusions: Kaplan-Meier survival analysis should better reflect the probability that a seizure will or will not occur because it accounts for censoring of studies that end without recording a seizure. In our series the K-M estimate of seizure probability was 40.9/28.3=44% greater than suggested by the percent of EEGs with seizures. Furthermore, the probability that a seizure would be recorded with additional monitoring if none had occurred by a given time can be used to offer guidance in termination of EEG monitoring. These methods are potentially useful in comparing outcomes by etiology and/or interventions.