A bootstrap approach to testing for an association between bursts in the neonatal EEG and changes in cerebral blood flow

In the EEG of normal premature neonates, the Tracé Discontinu pattern is common, consisting of bursts of relatively high amplitude activity, followed by periods of low amplitude. Little is known about the neurophysiology underlying these bursts.We therefore investigated if cerebral blood flow (CBFV), as measured by transcranial Doppler ultrasound varies with the bursts. We analysed 12 simultaneous recordings of CBFV and EEG from 8 neonates (gestational ages 25-32 weeks), recorded within the first two days after birth. The bursts in the EEG were identified semi-automatically by an experienced neonatal electroencephalographer (GB). The mean CBFV in each heart-beat was identified, and interpolated to achieve a uniform sampling rate. The ensemble of segments of mean CBFV, synchronized with the start of each burst was then obtained, and coherently averaged. Signals from some of the recordings showed an apparently consistent change in CBFV, whose statistical significance we then investigated.Due to the wide variation in responses between subjects, we investigated simple measures of 'consistency' of the CBFV responses within each recording, rather than any specific pattern of response. We thus tested the statistical significance of the mean-square difference between each response and their coherent average, using a bootstrap approach. In this, the results from the ensemble of signal segments synchronized with burst-onset are compared to those from random selections of signal-segments within each recording. Simulation studies confirmed the statistical robustness of the approach, and its sensitivity and flexibility.