Abstracts

Rationale: ESES is considered a potentially reversible encephalopathy and defined by a spike wave index (SWI) in sleep of > 50%. It is cumbersome to calculate SWI. Human experts (HE) are considered the gold standard and semiautomatic methods of calculating SWI have not been systematically studied. The corresponding author has presented data on the sensitivity ,specificity and reliability of Persyst software against 2 HE in the past. However there is considerable disagreement among HE in marking spikes in an EEG. SWI readings across >2 readers have not been previously compared to establish the spread of “normality” for this gold standard- HE. Persyst 13 (P13) has never been tested for a non inferiority hypothesis against multiple human readers to establish reliability in SWI calculation. Methods: Comparison of calculated SWI across 4 epilepsy trained HE: Twenty 30-minute contiguous EEG clips in sleep from patients previously diagnosed with ESES were identified. Within each 30-minute EEG clip, each HE marked the same 300 second epoch for all spikes. Each HE reader calculated the SWI as number of spike-affected 1 second bins divided by 300, expressed as a percentage. P13 was utilized to mark spikes, using its high sensitivity setting, in the same 300 second epochs marked by HE, and SWI was calculated using the above criteria. Pairwise HE differences and pairwise P13-HE differences for SWI were calculated for each case. Bootstrap resampling (BCa, N = 3000) was performed to better estimate mean differences and their 95% confidence bounds between HE and P13-HE pairs. Potential non inferiority of P13 with respect to the HE was tested by comparing the 95% confidence bounds of the mean differences between pairs for SWI.Non inferiority hypothesis: If P13, at its high sensitivity setting, is found to have a non inferior mean pairwise difference of SWI output compared to one or more pairwise human comparisons, then that algorithm would be considered non inferior to this study’s set of human readers. We determined dSI, the acceptable decrease or increase in SWI for the pairwise difference between human and algorithm, by comparing HE and finding the 95% confidence bounds that allow all four HE to be considered non inferior to all four HE. Results: 20 EEG records analyzed. Average patient age was 7.5 years. Total of 100 minutes of EEG marked by each HE. HE 1, 2, 3, and 4 marked 10075, 8635, 9710, and 9898 spikes respectively. Total 38,318 spikes marked. Highest and lowest 95% confidence bound of the mean difference in SWI between HE pairs was :High: 10.3%;Low: -10.2% ( table). Highest and lowest 95% confidence bound of the mean difference in Spike Index between P13 and human reader pairings was: High: 9.5%; Low: -6.7% Conclusions: Spike markings by HE in ESES vary widely as previously published. There is up to 10% mean spread among HE in calculating SWI. P13 is non inferior to the HE in calculating SWI in ESES in this dataset. Funding: none