Wavelet analysis of middle latency auditory evoked responses - Calculationof an index for detection of awareness during propofol administration

Background: middle latency auditory evoked responses (MLAER) as a measure o f depth of sedation are critically dependent on data quality and the analys is technique used. Manual peak labeling is subject to observer bias. This s tudy investigated whether a user-independent index based on wave let transf orm can be derived to discriminate between awake and unresponsive states du ring propofol sedation. Methods. After obtaining ethics committee approval and written informed con sent, 13 volunteers and 40 patients were: studied. in all subjects, propofo l was titrated to loss of response to verbal command. The volunteers were a llowed to recover, then propofol was titrated again to the same end point, and subjects were finally allowed to recover. From three MLAER waveforms at each stage, latencies and amplitudes of peaks Pa and Nb were measured manu ally. in addition, wavelet transform for analysis of MLAER was applied. Wav elet transform gives both frequency and time information by calculation of coefficients related to different frequency contents of the signal. Three c oefficients of the so-called wavelet detail level 4 were transformed into a single index (Db3d4) using logistic regression analysis, which was also us ed for calculation of indices for Pa, Nb, and Pa/Nb latencies. Prediction p robabilities for discrimination between awake and unresponsive states were calculated for all MLAER indices. Results: During propofal infusion, subjects were unresponsive, and MLAER co mponents were significantly depressed when compared with the awake states ( P < 0.001). The wavelet index Db3d4 was positive for awake and negative for unresponsive subjects with a prediction probability of 0.92. Conclusion: These data show that automated wavelet analysis may be used to differentiate between awake and unresponsive states. The threshold value fo r the wavelet index allows easy recognition of awake versus unresponsive su bjects. In addition, it is independent of subjective peak identification an d offers the advantage of easy implementation into monitoring devices.