BACKGROUND: Filter and peak detection algorithms implemented in amplitude-integrated electroencephalogram (aEEG) systems are not standardized. New aEEG systems are continuously enriching the market and clinicians are faced with different aEEG devices whose tracings may vary. OBJECTIVES: The aim of this work was to determine the role of different aEEG systems on quantitative measurements of the aEEG. METHODS: In this observational study, a single-channel aEEG recording (Olympic CFM 6000) with corresponding EEG signal was obtained from 32 infants at a gestational age of 36-44 weeks. The signals were split into 334 episodes of 4 h. New aEEG tracings were generated using the NicoletOne Reader Software and aEEG emulations with varying filter profiles and peak detection settings. The aEEG amplitude margins and automated annotation of continuous normal voltage (CNV) were compared. RESULTS: The output of the Olympic and the NicoletOne systems are very similar but not identical; the Spearman rank correlations of the aEEG amplitude margins exceeded 0.9 and the differences in the lower and upper amplitude margins were 1.55 μV (SD 1.47) and -2.12 μV (SD 1.44) on average (n = 309), respectively. The aEEG emulation showed that the differences between the output of the Olympic and the NicoletOne system could be primarily ascribed to the peak detection algorithm. The differences in output can affect automated analyses with agreement rates in CNV detection of 76% (n = 32, positive) and 92% (n = 32, negative) when comparing the Olympic to the NicoletOne outputs. CONCLUSIONS: Commercial aEEG systems have similar but not identical outputs. Care is advised when interpreting automated aEEG classifications across different devices.
BACKGROUND: Filter and peak detection algorithms implemented in amplitude-integrated electroencephalogram (aEEG) systems are not standardized. New aEEG systems are continuously enriching the market and clinicians are faced with different aEEG devices whose tracings may vary. OBJECTIVES: The aim of this work was to determine the role of different aEEG systems on quantitative measurements of the aEEG. METHODS: In this observational study, a single-channel aEEG recording (Olympic CFM 6000) with corresponding EEG signal was obtained from 32 infants at a gestational age of 36-44 weeks. The signals were split into 334 episodes of 4 h. New aEEG tracings were generated using the NicoletOne Reader Software and aEEG emulations with varying filter profiles and peak detection settings. The aEEG amplitude margins and automated annotation of continuous normal voltage (CNV) were compared. RESULTS: The output of the Olympic and the NicoletOne systems are very similar but not identical; the Spearman rank correlations of the aEEG amplitude margins exceeded 0.9 and the differences in the lower and upper amplitude margins were 1.55 μV (SD 1.47) and -2.12 μV (SD 1.44) on average (n = 309), respectively. The aEEG emulation showed that the differences between the output of the Olympic and the NicoletOne system could be primarily ascribed to the peak detection algorithm. The differences in output can affect automated analyses with agreement rates in CNV detection of 76% (n = 32, positive) and 92% (n = 32, negative) when comparing the Olympic to the NicoletOne outputs. CONCLUSIONS: Commercial aEEG systems have similar but not identical outputs. Care is advised when interpreting automated aEEG classifications across different devices.