Effects of artefact rejection and Bayesian weighted averaging on the efficiency of recording the newborn ABR.

OBJECTIVES: This study seeks to identify whether the most efficient artefact rejection (AR) level to use for recording the newborn auditory brainstem response (ABR) changes with recording conditions and if so, to suggest a simple strategy for testers to adopt when faced with nonideal test conditions. DESIGN: Twenty-six babies referred from the English Newborn Hearing Screening Programme were tested with ABR as a routine component of their postscreening assessment but their raw EEG responses were recorded for off-line analysis. One hundred 61 second data samples (equivalent to 3000 stimuli) were reaveraged off-line at five AR levels and two AR levels with Bayesian averaging; a total of 700 waveforms. An objective measurement of residual noise was used to determine the most efficient AR level (i.e., associated with the highest signal to noise ratio) to use in low, moderate, and severe noise conditions. RESULTS: The best performing AR levels were as follows: (1) low noise conditions: conventional averaging with AR = ±5 μV, (2) moderate noise conditions: conventional averaging with AR = ±5 μV or ± 6.5μV; Bayesian averaging with AR ±10 μV, and (3) severe noise conditions: Bayesian averaging with AR = ±10 μV. In severe noise conditions a more lenient AR level was most efficient for conventional averaging but a greater number of sweeps would be needed to reduce the noise allowed to enter the average. An interactive AR strategy has been proposed, including the trade-off between AR level and the number of sweeps required to control residual noise. CONCLUSIONS: AR level does influence test efficiency and the optimum level depends on the prevailing noise levels, which can change during the test session. It is important that testers are aware of this and develop evidence-based skills to optimize test quality, particularly in challenging test conditions.