Abstracts

Rationale: EEG helps narrow the differential diagnosis in emergency department (ED) patients with altered mental status (AMS). We sought to determine whether a 7-minute EEG provided the same diagnostic information as a 30-minute recording in this patient population. Methods: As part of a study evaluating a new EEG technology, 30-minute EEGs were obtained with a commercially available device and a full set of 10-20 electrodes on 261 sequential ED patients with AMS. The 227 EEGs available for review were de-identified and each one was interpreted by 2 of 6 board-certified electroencephalographers. The interpreters knew that the patients were ≥ 13 years old and had AMS in the ED, but did not know clinical history or medications. Interpreters assigned each study to one of 6 categories: Normal (Nl), Status Epilepticus (SE), Seizure (Sz), Interictal epileptiform with or without independent slowing (Ep), Slowing only (Sl), and Uninterpretable (U). Several months later, the same EEGers blindly reinterpreted the first 7 minutes of each EEG they had read previously. A 6 x 6 table of all 454 (2 x 227) interpretations was constructed (table). Accuracy measures of the 7m interpretations as predictive of the "gold-standard" 30m interpretations (predicting abnormal [including U] vs. normal) were obtained using a generalized mixed linear model with dependent variable 7m interpretation, fixed effect 30m interpretation, and random effects patient and rater. Similar models were used to predict SE vs. other and Sz vs. other. Bhapkar's test of homogeneity of row and column marginals was also performed. Results: With respect to the 30m recording, accuracy measures (95% CI) of the 7m recording were: sensitivity 0.93 (0.85, 0.97), specificity 0.57 (0.36, 0.76), PPV 0.93 (0.87, 0.96), NPV 0.55 (0.36, 0.73). Bhapkar's test was highly significant (p < 0.001) due primarily to the finding that Ep was much more frequent in the 30m recording than in the 7m recording. SE was interpreted 19 times in the 30m recording compared to 13 in the 7m segments, with 10 interpretations in common (NPV = 0.98 [0.96, 0.99]). Sz was diagnosed 8 times in both the 30m and 7m studies, but only 2 interpretations were common to both (NPV = 0.99 [0.98, 1.00]). Two Sz interpretations in the 7m segments became SE in the full study. Of 75 Nl interpretations in the 7m studies, 42 were Nl in the full study, and 26 of the remaining interpretations were Sl. Conclusions: As expected, if the first 7m of an EEG are interpreted as abnormal it is highly likely that the entire study will be interpreted as abnormal (PPV = 0.93). Conversely, a normal first 7m has low predictive value (NPV = 0.55) for the entire study in this patient population. However, if the first 7m are normal and the entire EEG abnormal, the abnormality is highly likely to be mild diffuse slowing, a relatively benign finding in ED patients with AMS. Also of clinical significance, if the first 7m are NOT interpreted as Sz or SE, it is extremely unlikely that the entire study will be interpreted as Sz or SE.