Investigation of the inter-observer variability effect on the prognostic value of somatosensory evoked potentials of the median nerve (SSEP) in cardiac arrest survivors using an SSEP classification

Background and purpose: We investigated the inter-observer variability in interpretation of median nerve SSEPs with regard to neurological prognosis in survivors of cardiac arrest (CA).

Methods: Four experienced neurologists analyzed 163 median nerve SSEPs on the basis of a pre-defined classification of SSEPs into five patterns (A-E), with consideration of cortical potentials up to a latency of 150ms. Of these, 133 recordings were from CA survivors and 30 were from healthy volunteers. The experts were blinded to whether a SSEP finding was from a CA survivor or a healthy volunteer. They were also unaware of the neurological outcome for the resuscitated patients. Three categories were defined for decision making. These were "good neurological outcome" represented by patterns A-C, "poor neurological outcome" (patterns D and E), and "not evaluable". Experts' agreement was calculated using the kappa-coefficient.

Results: The mean correct prediction by the experts was 81.8% (range 76.3-86.6%) in resuscitated patients with good neurological outcome. In those with poor neurological outcome, however, correct prediction was achieved in only 63% (60.5-66%). All SSEPs from healthy volunteers were classified as "good neurological outcome". The kappa-coefficient (κ) for all decision-making classifications was 0.75; for patients with poor outcome it was 0.76 and for those with good outcome 0.88. The predictive value for poor neurological outcome of the SSEP pattern D achieved a specificity of 93.5% and that of E a specificity of 98.4%.

Conclusion: Our study demonstrates good inter-observer agreement in the interpretation of median nerve SSEPs in CA survivors on the basis of a pre-defined SSEP evaluation set. The strongest correlation with poor outcome was found for pattern E, bilateral absence of the N20 peak.