Combining transient statistical markers from the EEG signal to predict brain sensitivity to general anesthesia

Abstract

Sensitivity to General Anesthesia (GA) remains difficult to anticipate, due to the lack of predictive tools to interpret the ElectroEncephaloGram (EEG). Using EEG, signal processing and classification methods, we develop a predictive framework to anticipate the sensitivity of a patient from the induction phase of GA. The procedure consists in estimating a reference time, where a stable α-band (8–12) Hz emerges from the spectrogram. To estimate the patient sensitivity during the induction phase, we introduce a decision tree that combines three statistical markers: 1-the proportion of time the alpha band disappears (α-suppressions, αS) in a sliding window, 2-the first occurrence time of αS and 3-the first occurrence time of an iso-electric suppression (IES). We extracted these three parameters within the first 15 min of anesthesia from 94 patients that had propofol target controlled infusion (TCI). To estimate the predictive values of the parameters, we use multivariate logistic regression at three time checkpoints. To confirm the patient sensitivity, we consider the entire GA duration and compare the results of our logistic approach, showing little deviation. We obtain the best predictor of the patient sensitivity by combining the slope of the proportion of αS and the first occurrence time of the IES computed during the induction phase, with an average AUC of 0.88. Finally, our analysis reveals that sensitive and non-sensitive patients are well classified into two distinct statistical groups. The present approach and algorithms can be used to detect the patient sensitivity during the induction phase of GA using propofol.