Influence of Auditory Stimulation Rates on Evoked Potentials during General Anesthesia: Relation between the Transient Auditory Middle-latency Response and the 40-Hz Auditory Steady State Response

Approval for this study was obtained from the institutional review boards at the University of Miami and Jackson Memorial Hospital (Human Subject Research Office, Miami, Florida). With informed/written consent, we enrolled healthy subjects, with American Society of Anesthesiologists physical status I or II, scheduled to have either elective inguinal hernia repair or laparoscopic cholecystectomy. Subjects were excluded with a history positive for seizures or hearing impairment.

Patients were premedicated with midazolam (2 mg) and fentanyl (100 μg) and were induced with propofol (1–2 mg/kg) and either vecuronium (0.1 mg/kg) or succinylcholine (0.5–1 mg/kg) followed by trachea intubation with an endotracheal tube. A combination of nitrous oxide and/or sevoflurane and intermittent boluses of fentanyl as needed was used to maintain general anesthesia. For all surgeries, anesthesiologists, who were not directly involved in the study and who were blinded to specific study details and outcome measures, were asked to follow standard of care in maintaining adequate anesthetic depth. Titration of anesthetic concentrations was accomplished, but was not limited to, monitoring of patient vital signs, expired concentration of inhalational anesthetic(s), ventilator parameters, and patient movement. Recording experiments (fig. 1) were initiated at least 20 min after induction and at least 5 min after establishment of constant expired concentrations of inhalational anesthetics. Each experiment consisted of three or more recording sets, and each set consisted of three interleaved rounds of 512 sweeps that used traditional, medium jitter, and isochronic auditory stimulation sequences (described in the following section) to generate the 5-Hz transient, 40-Hz transient, and 40-Hz ASSR, respectively. Summarizing (fig. 1), (1) three interleaved recording windows (each 512 sweeps in length and corresponding to each type of evoked response) formed a set, (2) three or more sets were considered to constitute an experiment, and (3) each experiment corresponded to a single constant level of anesthetic. Because during a typical surgery several different anesthetic concentrations occurred because of changing anesthetic requirements, data from any single patient often consisted of several different experiments and expired anesthetic concentrations (table 1).

Auditory stimuli consisted of rarefaction clicks that were presented monaurally to the right ear using an insert earphone (ER-3A Ethmotic Research, Elk Grove Village, IL). A stimulus sound level of 82 peak dB sound pressure level was calibrated in a 2-ml acoustic coupler using a precision sound level meter (Quest Model 1800; Quest Technologies, Oconomowoc, WI). A full description of auditory stimulus generation and evoked response recording using CLAD can be found elsewhere.27Briefly, three stimulus sequences were used in this study and are shown schematically (fig. 1). Two sequences consisted of clicks presented with no jitter at constant (isochronic) rates of 4.9 and 39.1 Hz. These sequences were used to record the 5-Hz transient and 40-Hz ASSR, respectively. To generate the 40-Hz transient, a 39.1-Hz sequence of medium-jittered clicks was used.27Jittering is a requirement of CLAD so that the deconvolution matrix has a unique solution.24The SmartEP system (Intelligent Hearing Systems, Miami, FL) was used to generate all sequences.

The electroencephalographic recordings were obtained using the SmartEP system. Disposable electrodes (Vermed, Bellows Falls, VT) were applied to FPz (positive), the right mastoid (negative), and the forehead (ground). Signals were amplified (gain: 100,000), band-pass filtered (10–1,500 Hz, 6 dB/oct), and digitized at 200-μs sampling time, with each sweep consisting of 1,024 data points with a duration of 204.8 ms. Artifact rejection was set at 30 μV. The ASSRs and 5-Hz transients were calculated by using traditional nonoverlapping 204.8-ms sweep averaging. The 40-Hz transients were calculated by averaging the convolved recordings for each jittered sequence, and then deconvolving using CLAD. Any low-frequency noise remaining after analog filtering, averaging, and deconvolution was removed using a spectral digital high-pass filter cutting frequencies of 15 Hz and below.

Five- and 40-Hz transients were linearly superimposed (convolved) to generate predicted 40-Hz ASSRs—5-Hz predicted ASSR  and 40-Hz predicted ASSR , respectively. The method used is similar to those described elsewhere.18,21,28Initially, a transient (204.8 ms, 1,024 points) was demarcated into eight equal segments of 25.6 ms or 128 points. The demarcated transient was circularly time shifted 25.6 ms (one segment) to generate a second. The time shifting was repeated for the second recording, and so on until eight recordings were obtained that could be wrapped around to form a continuous loop. These eight recordings were then convolved (summed linearly) to generate a predicted ASSR. Because the convolution process can generate small artifactual peaks, the predicted ASSRs were digitally denoised (comb filtered) so that only the main frequency (f⁰= 39.06 Hz) and the first two harmonics (f¹= 78.12 Hz and f²= 117.18 Hz) remained (refer to the next section). For consistency, this denoising was also applied to the recorded ASSRs (fig. 2).

Phasor analysis29was used to decompose recorded and predicted ASSR waveforms. For this study, we used the first three harmonics (40, 80, and 120 Hz), which we have found to contain most of the signal energy (figs. 2A and B). Each harmonic, a pure cosine wave, is obtained as a single beam of the Fourier decomposition. The cosine wave is defined by the real and imaginary components of a complex number. These components can be represented as phasors (vectors) on a phasor diagram (fig. 3A). As such, each ASSR is describable as a six-dimensional array , consisting of a real and imaginary number for each of the three harmonics (40, 80, and 120 Hz). Because it is not easy to graphically represent six-dimensional arrays, for the purpose of data presentation, each array is depicted as three projections corresponding to the three harmonics (fig. 3A). Phasor arrays are amenable to algebraic manipulation and statistical comparison testing (fig. 3B).

For each experiment we calculated a difference-phasor array  by subtracting the predicted from the recorded ASSR phasor arrays (fig. 3A). Our null hypothesis was as follows: The difference-phasor array is statistically different than zero, i.e. , the origin (0, 0, 0, 0, 0, 0). Rejection of this null hypothesis would indicate that the predicted and recorded phasor arrays (and therefore the corresponding ASSR waveforms) are not statistically different from each other. We tested the null hypothesis by performing the Hotelling T  ²test,30a multivariate analog of the Student t  test. The T  ²test gives the likelihood that a vector signal is greater than background residual noise. In our study, the vector signal is the grand-averaged difference-phasor array calculated from the difference-phasor arrays from each experiment (fig. 3B). With the T  ²test, confidence limits are represented as ellipses centered at the grand-averaged difference-phasor array tips. If the origin is not contained in the ellipse, the difference phasor is considered to be significantly different than zero for a specified P  value. All phasor and statistical calculations were performed in MATLAB Release 13 (The Mathworks, Inc., Natick, MA). The T  ²test was performed using a program available at mathworks.com.§

A total of 13 subjects (1 female) aged between 26 and 66 yr participated (table 1). As was done previously,31in this article, the auditory brainstem and middle-latency responses are included in the presentation of transient response waveforms to allow voltage comparisons across recordings. Adequate brainstem responses were obtained for all subjects during recording confirming adequate delivery of auditory stimuli and effectively ruling out the presence of significant hearing impairment to the rarefaction clicks (fig. 4, first two columns). A total of 24 steady state experiments were conducted. The concentrations of anesthetic ranged from 0.7 to 2.0% ± nitrous oxide (42–70%), indicating that both intersubject and intrasubject differences in the anesthetic concentrations were required to maintain adequate anesthetic depth (table 1). Transient signals in the microvolt range were readily recorded in every experiment (fig. 4). For each experiment, peak V of the brainstem response was similar for 5- and 40-Hz transients. Qualitative differences in amplitudes and latencies of the 5- and 40-Hz transient components (middle-latency range) are apparent on visual inspection (fig. 4). The peak amplitudes of the 40-Hz transient components were generally more attenuated compared to those of the 5-Hz transient. In addition, in many of the 5-Hz transients, an early peak (approximately 20 ms) was also observed, and was rarely observed in the 40-Hz transients. These qualitative differences were more easily appreciated in the grand-averaged waveforms (fig. 4). An additional difference was the latency of the first peak after the brainstem response in the 40-Hz transient, which had a shorter latency than the comparable peak in the 5-Hz transient.

The 5-Hz predicted ASSRs (i.e. , ASSRs predicted by the 5-Hz transient), the 40-Hz predicted ASSRs (i.e. , ASSRs predicted by the 40-Hz transients), and the recorded ASSRs from each experiment (and grand averages) are shown alongside the 5- and 40-Hz transients (fig. 4). The 5-Hz predicted ASSR peak-to-peak amplitudes were generally larger than those of 40-Hz predicted ASSRs. In addition, the 40-Hz predicted ASSRs closely resembled the recorded ASSRs for each experiment. These qualitative findings were more easily appreciated in the grand-averaged predicted and recorded ASSRs (fig. 4, bottom waveforms). Of considerable note, regular oscillations seem to be present in the transient waveforms. Fourier analysis (data not shown) reveals these oscillations are centered around 20 Hz for both 5- and 40-Hz stimulation conditions; however, the power appears greater in the former.

The complex numbers (real and imaginary) that describe the recorded and synthetic phasor arrays from each experiment (and grand averages) are presented in table 2. Five- and 40-Hz difference-phasor arrays were calculated from these data, and their projections are shown in figures 5A and B, respectively. Confidence intervals (P < 0.05 and P < 0.01) are represented by ellipses that surround the grand-averaged difference-phasor array projections. Referring to the 5-Hz difference-phasor array projections (fig. 5A), the origin is not contained in any of the ellipses (except for the ellipse (P < 0.01) corresponding to the 120-Hz harmonic projection), indicating that the grand-averaged difference-phasor array is significantly different from zero. The null hypothesis as tested on the six-dimensional difference-phasor array is retained (T  ²= 38.919, F = 5.076, P = 0.003). However, referring to the 40-Hz difference-phasor array projections (fig. 5B), the origin is contained in all ellipses, indicating that this six-dimensional array is not significantly different from zero (T  ²= 2.626, F = 0.343, P = 0.905).

Based on the observed qualitative differences between the 5- and 40-Hz transient waveforms, we set out to estimate the contributions of the transient components to the final ASSR waveforms (figs. 6 and 7), similar to what was done previously.26The recorded grand-averaged 5- and 40-Hz transients are shown in figures 6 and 7, respectively, with tentative labeling of component waveforms (e.g. , P^x, N^y, P^y). The transient waveforms were altered to generate various curtailed waveforms (figs. 6A and 7A). Each curtailed waveform was convolved to generate a predicted ASSR (figs. 6B and 7B, solid traces). Finally, each predicted ASSR was converted to its 40-Hz harmonic phasor. The phasors were arranged tip to end in order of increasing curtailed waveform length (figs. 6C and 7C). Referring to figure 6B, none of the 5-Hz predicted ASSRs (solid traces) seems to visually resemble the recorded ASSR (dashed traces). However, in figure 7B, curtailed waveforms (b–d) yield predicted ASSRs that visually resemble the recorded ASSR. The phasor diagram (fig. 7C) shows that the curtailed phasor (c) resides closest to the recorded ASSR phasor—about as close as the grand-averaged (whole) ASSR phasor. These data suggest that it is the early components of the transient response that contribute the most to the generation of the ASSR waveform.

The current study is the first to address the prediction discrepancy by using the 40-Hz transient response to predict the 40-Hz ASSR waveform during general anesthesia. The major finding is that there was no statistical difference between the predicted ASSRs (obtained after convolving the 40-Hz transient) and the recorded ASSRs. This result strongly suggests that the transient response and ASSR are generated by the same underlying neuronal structures. There are conflicting data in the literature regarding the relation between ASSR and transient responses in humans. In support of a common neurogenesis, numerous studies have shown a close correspondence of superimposed transient (electrical) responses and ASSR.17–19,32,33However, evidence contrary to these results comes from magnetoencephalographic human studies,23,34in which the components of the transient field were shown to be tonotopically distinguishable from the ASSR (field), and from animal studies.35The reason for these conflicting data may depend on differences in the methodologies and analyses used in those studies to address the question. At the outset of the current study, our concern was that if we were to find a poor agreement between predicted and recorded ASSRs (i.e. , corroboration of the prediction discrepancy), it would be difficult or impossible to dismiss differences in the recording conditions used to obtain transient and steady state waveforms as confounding factors. Hence, we used interleaving recording schedules, only recorded during steady state anesthetic concentrations and averaged over multiple recording epochs. To the best of our knowledge, the only difference in recording conditions was between the jittered stimulation sequence used to obtain the 40-Hz transient and the isochronic stimulation sequence used to obtain the ASSR. We were relieved after our initial qualitative, time-domain results showed that there seemed to be a strong likeness of the predicted and recorded ASSR waveforms. We subsequently wanted to dismiss this finding as due to chance by performing quantitative, frequency domain statistical analysis. We adapted the Hotelling T  ²test to develop a statistical method to test whether the difference-phasor arrays derived from predicted and recorded ASSR waveforms were statistically greater than zero (e.g. , an indication that the predicted and recorded waveforms were different). This would seem to be the first use of such a test in this context. The statistical results allow us to state with a high degree of certainty that the differences between predicted and recorded ASSRs are not statistically significant and, by inference, that the 40-Hz ASSR is a composite (or epiphenomenon) of the transient waveform.

The underlying explanation for the prediction discrepancy is that the morphology of the transient response varies with the rate of auditory stimulation during general anesthesia, and previous studies that reported a prediction discrepancy used 5- or 10-Hz (not 40-Hz) transient waveforms to predict the 40-Hz ASSR. For the purposes of reporting the differences between 5- and 40-Hz transient waveforms, we elected to give tentative labels to the middle latency components of the 40-Hz transient (fig. 7A) and the first positive peak after peak V of the 5-Hz transient (fig. 6A). As such, the main differences between the 5- and 40-Hz transient responses obtained in anesthetized subjects involve the early middle-latency peaks from 15 to 60 ms: (1) There is an additional peak (at approximately 20 ms), P^x, that is only observed in 5-Hz transient recordings. (2) In 40-Hz transient recordings, a peak, P^yis present that has a shorter latency than P^ain 5-Hz transient recordings. (3) The middle-latency component amplitudes in the 40-Hz transient are generally smaller than those in the 5-Hz transient. When selected portions of the 40-Hz transient are time shifted and summed, it seems that mainly the early peaks V (of the brainstem response), N^y, and P^ycontribute to the generation of the steady state waveform (fig. 7). If N^yand P^ycorrespond to N^aand P^a, this finding is consistent with suggestions17and evidence15from others that it is these early peaks that lead to the ASSR waveform. It also suggests that the neurogenerators shared by transient response and ASSR are the lateral lemniscus (V), and/or mesencephalic structures including the inferior colliculus (N^a) and the primary auditory cortex (P^a). It would be premature to report on how the peaks in the 5- and 40-Hz transients relate definitively. Our study was not designed to elucidate the anesthetic dose dependency of component latencies and amplitudes; we only analyzed recordings obtained during constant expired concentrations of anesthetic. The question of whether the 40-Hz transient peaks N^y, P^y, N^z, and P^zare equivalent to N^a, P^a, N^b, and P^b, respectively, will be answered by investigating the dynamic changes of component amplitudes and latencies as a function of the concentration of anesthetic, similar to the study reported by others36for sevoflurane. We are currently working on such a study.

It is possible that P^x, observed here in the 5-Hz transient during anesthesia, corresponds to the early middle-latency peak P⁰observed by others.37The peak P⁰is not consistently seen in transient recordings. Others have observed P⁰recorded from anesthetized subjects. For example, Schwender et al.  36studied dose dependency of component latency and amplitude in 9.3-Hz transient recordings during sevoflurane anesthesia. Though not explicitly reported, P⁰appears in their grand-averaged waveforms with all concentrations of sevoflurane. This finding, along with our data, preliminarily suggests that P⁰is present during anesthesia at lower stimulation rates (e.g. , 5 Hz, 9.3 Hz) and that it disappears at higher stimulation rates (e.g. , 40 Hz). It should be noted that in the 40-Hz transient, P⁰, may in fact be present (fig. 7A, see asterisk label) within the ascending portion of P^y, which seems to have a shorter latency than P^ain the 5-Hz transient response, though this is purely speculative. Electrode placement may be important for observing evoked potential signals. For this study, we used FPz and mastoid (positive and negative electrodes, respectively) and forehead (ground). Others have used alternate locations for the negative electrode. Our laboratory has obtained preliminary results showing that P⁰is observable in unanesthetized, awake volunteers when the negative electrode is placed at the inion. It would seem that movement of the negative electrode from the mastoid to the inion serves to diminish N^a, which perhaps reveals the P⁰peak (verbal unpublished data, June 2008, Jorge Bohorquez, Ph.D., Department of Biomedical Engineering, Neurosensory Laboratory, University of Miami, Coral Gables, Florida). Future studies will address how electrode placement affects evoked potentials.

We observed what seems to be an oscillatory activity in the 5- and 40-Hz transient recordings obtained from subjects under general anesthesia, and we do not observe similar oscillations in unanesthetized, awake subjects.26To the best of our knowledge, this is the first time such activity has been observed in a transient waveform; typical use of recording window lengths of 100 ms or less in previous studies, and the decreased sensitivity of shorter window lengths for detection of oscillatory activity are probable explanations. With CLAD, window lengths can be as long as 800 ms, and for this study, we used window lengths of 200 ms. We believe that the absence and presence of oscillatory activity in the transient waveform during the awake and anesthetized states, respectively, are consistent with the presence of neurons that can engage in a resonance with periodic stimulation. We hypothesize that anesthetics allow certain neurons to become entrained so that resulting evoked electrical activity becomes phase locked with the auditory stimulus. The significance and full characterization of this oscillatory activity remains to be established, and we are currently working in this regard.

The results presented here may help to explain an unexpected finding from a previous study. The 40-Hz ASSR amplitude was reported to be attenuated in the presence of nitrous oxide with isoflurane11and enflurane.21However, in the latter report, attenuation was found not to be dose dependent, and to vary in a nonmonotonic way.21Our results show that the 40-Hz ASSR can be adequately described and reconstructed from 40-Hz transient and that both transient waveform amplitude and latency are important considerations. We hypothesize that the nonmonotonic relation observed by others can be explained by anesthetic effects on transient component latencies that are endogenously convolved during constant 40-Hz stimulation. If, for example, transient latencies increase with enflurane concentration, the degree of constructive (or destructive) summation of waveforms (and ultimately ASSR amplitude) will vary according to a polar—not linear—scale. This is illustrated in figure 8. The ASSR amplitude would vary nonmonotonically even if a linear anesthetic dose dependency for latency exists. In this model, the ASSR phase would also vary nonmonotonically. If our hypothesis is true, it would indicate that the 40-Hz ASSR, albeit an easily obtained steady state evoked potential that is unequivocally related to anesthetic depth, may offer a subset of the information that can be obtained from the transient waveform.

It would have been useful to obtain baseline auditory evoked potentials in the awake state for comparison. Unfortunately, we found it extremely difficult to perform recordings on our awake subjects awaiting surgery in the holding area. To study the prediction discrepancy—a phenomenon observed only during general anesthesia—it was not important to have baseline recordings. In addition, we assume that awake recordings would be similar to those found in the literature and in a study similar to this one performed with awake subjects, recently reported by our group.26We are unable to exclude the effect of muscle relaxant use on our recording because we were unable to determine relaxation status during the continuum of each of the surgeries. However, with our study design, which included common electrode placement, interleaving of recordings, and use of time-domain and phasor analyses, we believe that any myogenic response or artifact would affect all recordings equally and would not confound our data interpretation. A potential limitation of this study is the large range of anesthetic concentrations that were not targeted to a quantitative endpoint (e.g. , expired percentage concentration). We decided to target anesthetic concentration to patient requirement to maintain general anesthesia. This target endpoint reduces the risk of either overdosing or underdosing anesthetic. It is also likely to have more real-world significance for future studies that will attempt to correlate evoked potential measures with awareness.

In summary, the 40-Hz ASSR is a composite or epiphenomenon of the evoked, transient waveform. The inability of 5- or 10-Hz transients to predict the 40-Hz ASSR is explained by an apparent stimulation rate-dependent effect on the morphology of the transient waveform recorded during anesthesia. It is likely that the ASSR amplitude can vary in a nonmonotonic (nonlinear) way even though changes in the transient amplitude and latency likely vary monotonically with anesthetic concentration. The use of CLAD was instrumental in accomplishing these studies, and its use contributed to our serendipitous observation of oscillations in both 5- and 40-Hz transients during anesthesia. The significance of these oscillations (which are not evident in awake recordings), as it relates to anesthetic action and depth, remains to be investigated.

The authors thank the Stanley Glaser Foundation; Alexander Castro, M.S., Graduate Student (Department of Biomedical Engineering, University of Miami, Coral Gables, Florida); and David Lubarsky, M.D., M.B.A., Professor and Chair, David Birnbach, M.D., Professor, Albert Varon, M.D., Professor, Shawn Banks, M.D., Assistant Professor, Kim Blumberg, M.D., Assistant Professor, Nick Nedeff, M.D., Assistant Professor, Edgar Pierre, M.D., Assistant Professor, Sripad Rao, M.D., Assistant Professor, and Christian Diez, M.D., Assistant Professor (all at the Department of Anesthesiology, University of Miami School of Medicine, Miami, Florida).