Abstracts

Rationale: The electrode placement of P3-P4 is most commonly used in neonatal ICUs for single channel amplitude integrated EEG (aEEG) registrations, and all the published literature concerns this channel. Restricted 10-20 system of electrode placement is used commonly for standard EEG registrations in neonates and does not include P3 and P4 electrodes. Most new digital EEG systems have the ability to display trends like aEEG. This prompted us to study the accuracy of single channel aEEG (channels C3-C4 vs P3-P4) in seizure detection and in classifying background EEG activity (6 vs 30 cm/hour time bases) compared with raw EEG. Methods: Polygraphic EEG recordings (full 10-20 system) of 8 hours, showing seizures, from 14 term neonates with birth asphyxia were studied. Seizures in the raw EEG were scored for onset, location, amplitude and frequency by a clinical neurophysiologist. Background EEG activity was scored for voltage and discontinuity. Using single channel aEEG with corresponding raw EEG (C3-C4 as well as P3-P4), the seizures (at 6 cm/hour) and background activity (at both 6 and 30 cm/hour time bases) were scored independently by a clinical neurophysiologist, neonatologist and neurotechnologist, experienced in reading neonatal EEG and aEEG.Results: A total of 305 seizures (median 15, range 2-67) were scored on raw EEG. For seizure detection aEEG C3-C4 showed a median sensitivity of 83 (mean 79.4, range 39-100) and a median positive predicitve value (PPV) of 77 (mean 74.7, range 24-100), while P3-P4 showed a median sensitivity of 85 (mean 78.2, range 38-100)and a median PPV 88.5 (mean 85.3, range 58-100). Seizures missed by aEEG were relatively low in amplitude (<50 µV), short (<20 sec), focal (especially frontal, occipital), or occurring along with too many EEG artifacts. False positive seizure detection by aEEG occurred due to artifacts (movements, muscle or ECG) or too many bursts in the EEG. Inter-rater agreement for aEEG seizure detection was ~80% between the three raters and for the aEEG background activity, almost 100%. For classifying background EEG activity, there was 100% agreement between the raw EEG as well as aEEG C3-C4 and P3-P4 at 6 and 30 cm/hour time bases in 8/14 patients. However, in 6 patients with moderately severe tracé discontinu, the upper margin of the aEEG trace was higher for C3-C4 channels, as well as for 6 cm/hour time base. Conclusions: aEEG has only moderate sensitivity for seizure detection, even in expert hands. Sensitivity was similar for C3-C4 and P3-P4 channels, but P3-P4 had less false-positive detections. aEEG is accurate in classifying background EEG activity. Time bases of 6 cm/hour and 30 cm/hour time bases tend to perform similarly, although in moderately severe tracé discontinu, the 30 cm/hour time base may be more accurate.