2006 Annual Scientific Meeting - CLICK FOR PROGRAMME

Entropy

Stefan Schraag, MD PhD

What is entropy?

"No one really knows what

entropy is, so in a debate you

will always have the advantage!“

J. Von Neumann to Claude Shannon:

Scientific American 1971; 224:178

The concept of entropy comprises the hypothesis that the EEG should have a higher degree of order and less randomness during increased levels of anaesthesia. In other words, entropy measures the irregularity of the signal and is Iindependent of the absolute amplitude/frequency scale. Different algorithms to determine entropy for EEG measurements have been described.

Approximate entropy and Shannon entropy quantify the predictability of amplitudes in the EEG signal, whereas Spectral entropy describes the randomness of frequencies in the EEG signal. Approximate or Shannon entropy is a new statistical parameter derived from Kolmogorov-Sinai entropy formula and describes amount of regularity in the data set. It quantifies the predictability of subsequent amplitude values, and the degree of skew in the distribution of values.Thus increasing hypnotic drug concentrations modify EEG from a skew distribution towards equal distribution.

Entropy is generated in three steps:

1. Normalization of power spectrum P(fi):

2. Mapping and summation:

3. Normalization:

The commercially available Entropy monitor distributed by Datex-Ohmeda/GE is composed of time-frequency balanced spectral entropy and provides both state entropy (0.8-30Hz) and response entropy (0.8-47Hz) indices. The latter is meant to detect increased EMG activity and thus identify better levels of light anaesthesia or arousal. The optimization of response time is designed by a set of time windows with lengths from 2 sec up to 30/60 sec. At all levels of anesthesia, RE and SE are computed using the same algorithm. During burst suppression, periods of suppression contribute zero entropy with all entropy confined in the bursts.

In clinical studies, Entropy correlates well with clinical signs of anaesthetic depth and the Pk values for APE were slightly better than those for BIS, SEF 95 and the combination of drug concentrations, whereas a much lower Pk was observed for tolerance of airway manipulations and noxious stimulation than for hypnotic endpoints. In direct comparison though, correlation between EEG and Propofol Ce was better for BIS than for Entropy. In  general, entropy has a good correlation with graded levels of sedation and hypnosis incl. EMG. It is an open source calculated algorithm with reasonable artefact robustness and baseline stability. However, entropy is clinically evaluated only in few studies and has less good Pk during stimulation. Only a few comparative data on performance during combinations of anaesthetic drugs are available yet.

Conclusion

Entropy will always quantify the degree of cortical activity/suppression, regardless of the underlying level of consciousness. Prediction probability is higher for hypnotic endpoints and lower for identifying noxious stimuli. Drug interactions as used in anaesthesia practice are less good described by Entropy. There are no conclusive studies about preventing awareness with Entropy monitoring.

References for further reading:

1. Pincus SM et al. J Clin Monit 1991; 7: 335-45
2. Rezek IA et al. IEEE Trans Biomed Eng 1998; 45: 1186-91
3. Bruhn J et al. Anesthesiology 2000; 92: 715-26
4. Bruhn J et al. Anesthesiology 2001; 95: 30-5
5. Bruhn J et al. Anesthesiology. 2002; 96:54-9
6. Vanluchene A at al. Anesthesiology 2004,101:34-42
7. Vanluchene A et al. Br J Anaesth 2004;93:645-54
8. Anderson RE et al. BJA 2004; 92:167-70