Objective: Early prognostication after cardiac arrest is limited since the introduction of targeted temperature management treatment, but continuous EEG (cEEG) monitoring seems a promising tool. We aimed to improve an existing automatic outcome prediction algorithm, the cerebral recovery index (CRI) by adding a newly developed automatic detection algorithm for burst suppression with and without identical bursts and evaluating the relevance of the features in the current CRI.

Methods: For development of the automatic detection algorithm for burst suppression with and without identical bursts, three methods for burst onset detection (based on non linear energy operator and based on recurrence rate with an without embedding) were compared to manual burst onset detection in a training set of 46 patients with burst suppression after cardiac arrest. The optimal performing method was evaluated in a test set of 19 additional EEGs and implemented in the CRI. Weight factors of the other features in the CRI were optimized in a different training set of 88 patients and the optimal performing CRI was evaluated in a test set of another 88 patients.

Results: Recurrence rate without embedding resulted in 100% specificity and 67% sensitivity for detection of burst suppression with identical bursts in the test set. The CRI was optimised by adding the burst identicity score, removing Shannon entropy and alpha/delta ratio and changing weight factors for regularity and coherence in the delta band. This resulted in higher sensitivities for good (30% at 12-24h post resuscitation) but especially for poor outcome (33% at 12-24h, 36% at 24-36h and 63% at 36-48h) compared to the old CRI.

Conclusion: The revised CRI predicts especially poor outcome with a high sensitivity and specificity and can be implemented in clinical practice to aid prognostication in patients after cardiac arrest.

Wednesday 7 January 2015, 16:30 - 17:30 h

Building Carré - room CR 3.718