Utility of Operational Criteria for Identifying Interictal Epileptiform Discharges on EEG: A Decision Hygiene Approach
Doyle Yuan1, Roohi Katyal2, Irfan Sheikh3, Adam Greenblatt4, Niravkumar Barot5, Ushtar Amin6, Daniel Weber7, Ioannis Karakis8, Kollencehri Puthenveetti Vinayan9, Selim Benbadis10, Sandor Beniczky11, M. Westover12, Fabio Nascimento13
1Dallas VA Medical Center, 2Louisiana State University Health Sciences Center, 3UT Southwestern, 4Washington University in St. Louis, 5University of PIttsburgh, 6University of South Florida - JAHVAH, 7St. Louis University, 8Emory University, 9Amrita Institute of Medical Sciences, 10University of South Florida, 11Aarhus University & Danish Epilepsy Center, 12MGH, 13Washington University Medical School
Objective:
To determine if implementing the operational definition of interictal epileptiform discharges (IEDs) proposed by the International Federation of Clinical Neurophysiology (IFCN) improves expert diagnostic performance and interrater reliability (IRR).
Background:
Accurate and reliable EEG interpretation is critical for the diagnosis and management of epilepsy. However, variability exists even among experts, which may largely be due to differences in IED identification thresholds. Standardized criteria such as those proposed by the IFCN may minimize this unwanted variability and improve diagnostic performance.
Design/Methods:
In a 2-part prospective analysis, 9 EEG experts classified 200 candidate IEDs (100 expert-validated [Series A], 100 epilepsy monitoring unit-validated [Series B]) as epileptiform or non-epileptiform, in random order. In Part I, raters performed binary classification only. In Part II, raters selected the applicable IFCN criteria for each candidate IED in addition to classifying it as epileptiform vs not. Parts I and II were separated by at least 30 days to reduce the effect of visual recall. Finally, we compared aggregate performance and IRR between both parts, and we separately assessed IRR for each of the six IFCN criteria.
Results:
Overall, there were no major differences in performance (AUC; 0.81 vs 0.82) or IRR (Gwet’s AC1; 0.48 vs. 0.47) between Parts I and II. Similarly, there were no major differences in performance or IRR for Series A (0.81 vs. 0.79; 0.34 vs. 0.31) and Series B (0.95 vs. 0.95; 0.62 vs. 0.63). IRR was fair for criteria 2, 3, 5 and 6, and moderate for criteria 1 and 4.
Conclusions:
The findings of this study suggest that the IFCN criteria to define IEDs may not significantly improve performance or IRR among experts. Increasing expert IRR for each criterion separately may enhance the utility of the IFCN criteria in clinical practice.