Abstracts

Rationale: There is variable opinion regarding rhythmic and periodic EEG patterns in critically ill patients. Prior to research of treatment of these patterns, a universal terminology needs to be established, and consistent observer agreement should be demonstrated using said terminology. The ACNS recently proposed standardized terminology (Hirsch et al 2005) for these patterns. This study evaluated interobserver agreement using this terminology.Methods: A pilot study using 10-second EEG samples has been presented previously. Seven epileptologists from our center were selected as readers. A total of 30 EEG samples (all 30-minute recordings except one 20-minute) from a previously established research data set, comprised of comatose patients with subarachnoid hemorrhage, were used. The evaluators were presented the EEGs in random order, and used the terminology to “score” each EEG. The terminology addresses the following elements: presence or absence of rhythmic or periodic patterns, localization, morphology, persistence, duration, rhythm, and frequency. Descriptive terms for sporadic (i.e. non-rhythmic) discharges are also included. After “scoring” each EEG, the results from each reader were tabulated and interobserver agreement was calculated using pairwise kappa statistics, which were then averaged for each variable. Results: Mean kappa values for each item in the terminology are presented in Table 1. Moderate agreement beyond chance was seen for the presence/absence of rhythmic or periodic patterns (mean k=0.443337) and for localization of rhythmic or periodic patterns (mean k=0.420072). Fair agreement was seen for morphology (mean k=0.377358), persistence (mean k=0. 388502), and duration (mean k=0.341988). Slight agreement was seen for frequency (mean k=0.147592), response to stimulus (mean k=0.175718), change over time (mean k=0.069521), and the terms “quasi” (mean k=0.068124), “sudden onset” (mean k=0.027413), “gradual onset” (mean k=0.047109), “frontally predominant” (mean k=0.074278), and “plus” (mean k=0.081564). The frequency of sporadic discharges also showed slight agreement (mean k=0.046824). Certain findings appeared to have a low overall prevalence in the sample, which can affect kappa values. In this situation, the kappa value and the positive agreement may be low, despite a high degree of negative agreement (Table 2). Conclusions: Even when using standardized terminology, significant variability exists between readers in the description of rhythmic and periodic EEG patterns. Compared to the pilot study, which used 10-second EEG samples, interobserver agreement deteriorated with longer, 30-minute samples. In particular, detailed descriptors (such as “quasi”) show a higher degree of variability compared to basic descriptors (such as “generalized” vs. “lateralized”). The utility of standardized terminology is dependent upon consistent use between observers. Further refinement or enhanced definition of the terminology may be required to improve interobserver agreement.