Meta-analysis of EEG test performance shows wide variation among studies

Background: EEG results are used for counseling patients with seizures about prognosis and deciding on medications. Published sensitivities of interictal EEG vary widely. Objective: To account for variation in test characteristics between studies. Methods: Meta-analysis. Medline search, 1970 to 2000, of English language studies. Standard methods for meta-analysis of diagnostic test performance were used to determine the ability of EEG results to distinguish between patients who will and will not have seizures. Using linear regression, the authors assessed the influence of readers' thresholds for classifying the EEG as positive, sample probability of seizure, percent of subjects with prior neurologic impairment, percent treated, and years followed. Results: Twenty-five studies involving 4,912 EEG met inclusion criteria. Specificity (range 0.13 to 0.99) and sensitivity (range 0.20 to 0.91) of epileptiform EEG interpretations varied widely and were heterogeneous by chi(2) analysis (p < 0.001 for each). Diagnostic accuracy of EEG and the thresholds for classifying EEG as positive varied widely. In the multivariate model, differences in readers' thresholds accounted for 37% of the variance in EEG diagnostic accuracy, and no other reported factors were significant. Conclusion: This analysis suggests that there is wide interreader variation in sensitivity and specificity of EEG interpretations, and that this variation influences the ability of EEG to discriminate between those who will and will not have seizure recurrences. In clinical practice, interpreting the degree to which a positive EEG result predicts increased seizure risk in an individual patient is difficult. Interpreting EEG with higher specificity yields more accurate predictions.