Accuracy of clinical signs, SEP, and EEG in predicting outcome of hypoxic coma

Accurate prediction of neurologic outcome after hypoxic coma is important. Previous systematic reviews have not used summary statistics to summarize and formally compare the accuracy of different prognostic tests. We therefore used summary receiver operating characteristic curve (SROC) and cluster regression methods to compare motor and pupillary responses with sensory evoked potential (SEP) and EEG in predicting outcome after hypoxic coma.We searched PubMed, MEDLINE, and Embase (1966-2007) for reports in English, German, and French and identified 25 suitable studies. An SROC was constructed for each marker (SEP, EEG, M1 and M < or = 3), and the area under the curve (AUC), a measure of diagnostic accuracy, was determined. For comparison, we calculated the differences between the AUC for each test and M1 reference standard.The AUC for absent SEP was larger than those for M1, M < or = 3, absent pupillary response, and EEG when the examinations were performed within the first 24 hours. The difference between the AUC for SEP (AUC 0.891) and that for M1 (AUC 0.786) was small (0.105, 95% confidence interval 0.023-0.187), only reaching significance on day 1 after coma onset. The use of M < or = 3 improved the diagnostic accuracy of motor signs.This study demonstrated that sensory evoked potential (SEP) is marginally better than M1 at predicting outcome after hypoxic coma. However, the superiority of SEP diminishes after day 1 and when M < or = 3 is used. The findings therefore caution against the tendency to generalize that SEP is a better marker than clinical signs.