Abstracts

Rationale: The diagnosis of electrical status epilepticus in sleep (ESES) is based on significant activation of spike wave activity from awake to asleep state. A spike wave index (SWI) of 50% or 85% is most commonly used at our institution to make the diagnosis with the threshold SWI being dependent on clinician reading the EEG (CE) as some clinicians are now using less rigid definitions. A computerized program to automate SWI would be easier and consistent across readers. We sought to evaluate the ability of an interictal spike detection software: Persyst Development Corporation's Persyst 11 to detect threshold SWI and compare diagnostic accuracy between Persyst software and CE in ESES . Methods: All patients with ESES from 2009 to 2014 were retrospectively analyzed. Only those EEGs with at least two 15 minute clips each in awake and asleep state OR one 30 minute clip each in awake and asleep state were included. Multiple clips were analyzed per EEG to improve accuracy. Using Event Density Spike Count per / clip, number of spikes per clip was recorded. An average awake and average asleep spike count was calculated by averaging multiple clips per each EEG. An average SWI in sleep of either 50% or 85% was coded as ESES depending on the definition used by the CE. All raw EEG was evaluated if there was a conflict by senior author. The sensitivity, specificity, positive predictive value and negative predictive value of the user-defined spike counting metrics configured in Persyst 11 was compared to the same by CE Results: Nineteen EEGs among thirteen patients satisfied inclusion criteria. Fifty -three clips were analyzed with an average of 2 clips per EEG. Persyst diagnosed ESES in 17/19 cases while expert did so in 15/19(p=0.6). The Persyst group had a sensitivity of 100%, specificity of 66%, positive predictive value of 94% and a negative predictive value of 100%. The CE had a sensitivity of 93.7 %, specificity of 100%, positive predictive value of 100% , negative predictive value of 75%. There was one false negative in the CE group and one false positive in the Persyst group. In one case, the CE assessed activation to be less than 85% by visual analysis but upon independent review this was inaccurate. What Persyst software identified as spikes in one case were actually showed sharply contoured sleep activity. Conclusions: Persyst software has a high sensitivity and high positive predictive value for identifying ESES by spike counting. Despite the possibility of false positives, reduced chances of missing a diagnosis and consistency in assessment makes this a useful tool to aid the clinician.