Interictal epileptiform discharges on scalp electroencephalography: Reproducibility of visual interpretation-A...

Interictal epileptiform discharges (IEDs) are a key diagnostic finding in first seizure, early epilepsy, and acute brain dysfunction. Variability in visual electroencephalographic (EEG) analysis contributes to diagnostic uncertainty. We performed a systematic review of the reproducibility of IED detection in clinical EEG.

We updated a systematic review. We searched four databases (Pubmed, Embase, PsycInfo, Cochrane) for studies reporting interrater or intrarater reproducibility of clinical scalp EEG interpretation, covering records until April 2025. Four raters independently screened 4841 citations and 119 full texts in duplicate. Study quality was assessed using Guidelines for Reporting Reliability and Agreement Studies. Studies reporting data on IED detection were analyzed. This study is registered with PROSPERO (Prospective Register of Systematic Reviews; number CRD42018099318).

Thirty-two studies reported reproducibility of IED detection (25 interrater, seven intrarater). No other IED-related EEG feature was examined in more than two studies. Ten studies met criteria for high clinical relevance, defined as relatively unselected patient populations without enrichment for epilepsy or IED occurrence. Reproducibility was highly variable across studies. For whole-EEG assessment, percent agreement was approximately 80% in first seizure; kappa-type statistics were higher in known epilepsy (interrater κ ≈ .8) than in acute brain dysfunction (κ ≈ .5) and first seizure (interrater κ ≈ .4-.6). Short EEG clips showed consistently lower reproducibility than whole-EEG assessments. Among studies reporting both intra- and interrater data with the same metric, intrarater reproducibility was equal to or higher than interrater, indicating intrarater disagreement accounts for a substantial share of interrater disagreement. "Paroxysmal" was used inconsistently.

IED detection has moderate reproducibility in unselected clinical populations (κ ≈ .4-.6), higher in known epilepsy settings (κ ≈ .8), and moderate to fair in acute brain dysfunction (κ ≈ .3-.5). Intrarater reproducibility is understudied and likely contributes substantially to interrater variability. Some traditional EEG terminologies ("EEG dialects") are mutually incompatible. Future studies should use consecutive unselected populations and standard nomenclature, standardize rater experience, and be designed for cross-study comparability.