Autoregressive Modeling with Exogenous Input of Middle-Latency Auditory-Evoked Potentials to Measure Rapid Changes in Depth of Anesthesia

Abstract

Obtaining an adequate depth of anesthesia is a continuous challenge to the anesthetist. With the introduction of muscle-relaxing agents the traditional signs of awareness are often obscured, or difficult to interpret. These signs include blood pressure, heart rate, pupil size, etc. However, these factors do not describe the depth of anesthesia (DA) in a cerebral activity sense. Hence, a better measure of the DA is required. It has been suggested that Auditory-Evoked Potentials (AEP) can provide additional information about the DA. The general method of extracting AEP is by use of a Moving Time Average (MTA). However, the MTA is time consuming because a large number of repetitions is needed to produce an estimate of the AEP. Hence, changes occurring over a small number of sweeps will not be detected by the MTA average. We describe a system-identification method, an autoregressive model with exogeneous input (ARX) model, to produce a sweep-by-sweep estimate of the AEP. The method was clinically evaluated in 10 patients anesthetized with alfentanil and propofol. The time interval between propofol induction and the time when the Na-Pa amplitude was decreased to 25% of the initial amplitude was measured. These measurements showed that ARX-estimated compared to MTA-estimated AEP was significantly faster in tracing transition from consciousness to unconsciousness during propofol induction (p < 0.05).