AUTOMATED MONITORING OF BRAINSTEM AUDITORY EVOKEDPOTENTIALS IN THE OPERATINGROOM j. R. Boston, PhD,* L. G. Deneault, MS,* L. Kronk,* and P.J. Jannetta, MD~-

Boston JR, Deneault LG, Kronk L, Jannetta PJ: Automated monitoring of brainstem auditory evoked potentials in the operating roonl.

J Clin Monit 1985;1:161-167 ABSTRACT. Wc monitored brainstem auditory evoked poten-

tials in 112 patients undergoing rctromastoid craniectomies for microvascular decompression. To provide information on latency changes as quickly as possible, we implemcnted a block averaging techniquc of data acquisition with automatic tracking of wave V latency, which is the most clinically useful information. A change in peak latency probably due to surgical manipulation was obscrvcd in 63% of the paticnts, and the change could bc at least partially corrected by modification of surgical technique. Twenty percent of the 89 patients who underwent preoperative and postoperative audiomctric testing showed a postoperative hearing decrement. Some patients had large intraoperative incrcascs in latency without suffering postoperative hcaring deficits, and somc incurred hearing deficits even though the intraopcrative latency increases wcrc relatively small. However, patients whosc brainstcm auditory cvoked potentials were lost during surgcry, cven temporarily, were likely to have postoperative hearing decrements. Patients who had deficits tended to have slightly greater increases in latency than patients without deficits, but the difference in the mean increases of thc two groups was not statistically significant. Most of the deficits were small, and all resolved over time. KEY WORDS. Monitoring; Brainstem auditory evoked poten-

tials, peak detection; Brain: evoked potentials; Surgery: retromastoid craniectomy.

Front the Departnaents of *Anesthesiology and "['Neurological Surgery, University of Pittsburgh School of Medicinc, Pittsburgh, PA 15261. Received Nov 9, 1984, and in revised form Feb 11, 1985. Accepted for publication Feb 20, 1985. Address correspondence to Dr Boston.

T h e brainstem auditory e v o k e d potential (BAEP) is an electrical signal that can be recorded with surface E E G electrodes in h u m a n s [1]. It is generated in ascending neural p a t h w a y s in the auditory periphery and brainstem in response to a transient acoustic stimulus, usually a click. Because the signal is small (less than one IxV) in amplitude, it is m a s k e d by the m u c h larger spontaneous E E G signal. Signal averaging techniques must be used to extract the B A E P f r o m the EEG; responses to 1,000 or m o r e stimuli are c o m m o n l y averaged. T h e B A E P recorded with scalp electrodes on the vertex and ipsilateral mastoid is a series o f vertex-positive peaks. T h e first five peaks, referred to as I through V, are the m o s t consistent, and the latency~: o f these peaks is the information that is m o s t widely used clinically. T h e origin o f the peaks is not completely clear, since evidence f r o m animal studies [2,3], clinical studies in h u m a n s [4], and direct intracranial recordings [5,6] is not c o m p l e t e l y consistent. Peak I p r o b a b l y originates in the peripheral auditory nerve. Peak II has been attributed to the intracranial portion o f the nerve and the :[:Latency is defined as the delay between the audible click and the given peak. Thus, the time between the stimulus and peak V is called the peak V latency.The term poststimulus latency is also used. 161

162 Journal of Clinical Monitoring I/ol I No 3 July 1985

cochlear nucleus. Peak III arises in the lower brainstem auditory pathways, and peaks IV, V, and the prominent downslope following peak V arc probably generated in the lateral lemniscus and the inferior colliculus. Increase in the interpeak interval is generally associated with neural dysfunction of structures near or between the proposed generators of the peaks. BAEPs can be recorded in anesthetized and in awakc humans, and the recording methods are noninvasive and relatively simple. Because the peak latencies appear to increase with stretching of the eighth ncrvc and brainstem retraction, they have becn used for intraoperative monitoring during retromastoid cranicctomies [7,8]. The purpose of BAEP monitoring is to indirectly detect changes in hearing during the procedure that are due to retraction and surgical manipulation. Change in BAEP latency can occur very rapidly (in seconds or minutes); we believe that surgeons should be notified of these changes as soon as possible to allow them to modify their approach. Stimulus rates higher than 10 to 15 per sccond reduce definition of the response, especially in the early pcaks [9]. (Higher rates yield a peak V, but, sincc change in the interpeak interval is of intcrest, thesc rates have not been commonly used for monitoring.) Conventional averaging requires 1,000 to 2,000 individual responses, and 1 to 3 minutes is required for one averaged BAEP. Several minutes can be required for display and analysis of the response, including measurement of peak parameters. Because considerable change can occur ovcr this period, we developed an automated monitoring tcchnique that provides new waveforms and latency values every 25 to 50 seconds. In an 18-month period, we monitored 112 patients undergoing retromastoid craniectomies for microvascular decompression. This article summarizes the technique and describes the intraoperative changes in latency we saw. METHODS

The study was approved by the Institutional Revicw Board for Biomedical Research of The University of Pittsburgh.

Acquisition of B A E P Acoustic stimuli were generated by delivering a 50 ~.s rectangular electrical pulse to a Madsen ear insert transducer (Madsen Electronics, Buffalo, NY). The transducer was taped into the ear ipsilateral to the surgical site before the patient was anesthetized. The patient's subjective threshold (within 10 dB) to clicks was ascertained in the operating room. (The clicks were masked

by the background noise in the room, and the thresholds wcre approximatcly 10 dB higher than those occurring in normal subjects in a quiet laboratory.) The acoustic intensity uscd for monitoring was 50 to 60 dB above the threshold, depending on the shape of the patient's audiogram and the responses obtaincd. The craniectomies were pcrformed undcr general endotrachcal ancsthcsia with the patient in a lateral position. After induction, sterile necdlc clcctrodes wcre placed at the vcrtex and on the cheek directly in front of the ear canal. A ground electrode was placed on the forehead. Elcctrode impcdanccs wcrc less than 4 kHz. The EEG was amplified with optically coupled amplifiers (Model HGA-200, Nicolct Biomedical, Madison, WI) with a gain of 100,000 Hz and a bandpass filter from 150 to 3,000 Hz. Although the usc of this high-pass cutoff frcqucncy distorts the BAEP waveform, it has little effect on peak latency values [10]. The contralatcral car was not tested in this series. Although such responses would indicatc systcmic cffccts (such as effects of temperaturc or anesthetic) on the BAEP, data from a previous study [8] showed no notable change in the response from the contralatcral car during the procedure. In addition, the need to acquire responses rapidly from the ipsilatcral ear during periods of rapid latency change precluded this additional monitoring. To elicit a BAEP, stimuli were presented at an avcrage rate of 12 per second; the mean interstimuhis intcrval was 83 ms. The actual intervals bctwecn stimuli were of random duration, with a variation of -+8 ms to avoid entrainment of the averaging system to 60 Hz interference. For each stimulus, the computer digitized a 15 ms interval of the EEG signal (the data window) at a sampiing rate of 11 kHz and included the resulting waveform in the current average. The stimulus occurred in the middle of the data window. The control interval, the part of the data window that prcccded the stimulus, showed the residual noise. This interval helped the operator evaluate thc effects of electrical interference and movement artifact from surgical manipulation and, hence, the rcliability of the rcsponse, ldcally, this interval would have zero amplitude. The sensory response occurred in the interval following the stimulus. The amplitude of response components could be compared directly with thc residual noise amplitude in the control interval. The monitoring system was implemented on an lntel Multibus using an SBC 80/30 CPU board (Intel, Santa Clara, CA) [11]. During surgical preparation of the patient before incision, we obtained baseline responses. Two responses (each an average of 1,024 individual stimulus presentations) were obtained and plotted. Peaks I, III, and V

Boston et ah Monitoring of BAEP

could usually be identified clearly in these responses; a second pair of responses was obtained if peak definition was unclear in the first pair. Continuous monitoring was initiated after the dura was exposed.

163

STIMULUS RESPONSE II1 V

CONTROL

Monitoring Algorithm We used a monitoring algorithm that is described in detail in this issuc [12]. Briefly, a block averaging technique stored five consecutive averaged responses in separate buffers in the computer. The average currently being acquired was displayed along with the average of the preceding four averages. With this technique, the current average could be compared with the average of the previous four while the current average was being acquired. When the current average was completed it was included in the large average, and the oldest average in the large average was deleted. The number of individual waveforms in each averaged response was the smallest that would yield a waveform that could be interpreted. For BAEPs, the response often could be identified after 200 stimuli and almost always after 500 stimuli. Although the waveform was noisy, it indicated whether a response was present or whether gross changes in waveform had occurred. This information is important in operating room monitoring, and the speed with which it can be obtained is critical. We used averages based on 256 or 512 individual responses. A new average was obtained every 25 to 50 seconds, excluding time lost during electrocautery interference. Although these averages were useful for determining gross changes in the response waveform, they were too noisy to provide reliable measures of peak latency. Changes of a few tenths of a millisecond over 5 to 10 minutes can be important during BAEP monitoring, but the response-to-response variability of peak latency in these noisy waveforms often exceeded that value. The average of four averages was a BAEP based on 1,024 or 2,048 individual responses; it provided a stable waveform when the response itself was not changing rapidly. Small changes in the response, occurring over periods of many minutes, could be reliably tracked with measurements obtained from this more stable average. Our computer system excluded large amplitude responses (usually caused by external interference) from the average. The major source of electrical interference during monitoring is caused by electrocautery. At times, the cautery artifact was sufficiently small that it was not rejected on an amplitude basis but still disrupted the BAEP. These instances were identified by large amplitude signals in the control interval of the response; at such times data acquisition was manually suspended.

BASELINE

MAXIMUM CHANGE

FINAL 0.51 0 ms

Fig I. Example qf brainstem auditory evoked response waveforms analyzed for changes in latency. The monitoring algorithm also tracked peak V latency as described elsewhere in this issue [12]. Before monitoring was initiated, the operator positioned a cursor on peak V of the baseline response. This peak was automatically detected in each large average, since these waveforms were sufficiently stable to provide accurate latency values. Peak detection required about 1 second, and the peak latency was displayed to the operator immediately. This function of the algorithm eliminated the time required to make latency measurements. The peak detection algorithm used several checks to differentiate the actual peak V from spurious peaks caused by noise. The algorithm was correct on about 95% of the responses. However, if the change in latency was rapid or was accompanied by a major change in the waveform, the algorithm could refuse to select a peak or select an incorrect one. The cursor marked the location of the peak detected by the algorithm on the displayed large average; the operator could therefore verify the accuracy of peak detection and return the cursor to the correct peak at any time.

Data Analysis To characterize the intraoperative changes in latency, the latencies of peaks I, III, and V were measured at three different times during the procedure, as shown in Figure 1; measurements were made on a baseline response obtained after the patient was anesthetized but before the dura was opened, on the waveform that showed the m a x i m u m change in peak V latency, and on

164 Journal of Clinical Monitoring Vol 1 No 3 July 1985

WAVE V I

TIME (rain) 0 BASELINE

68 72 83

120 I

I

I

0

5 TIME (ms)

10

END

STIMULUS Fig 2. Brainstem auditory evoked potentials recorded during a retromastoid craniectonl),Jor seventh nerve deconlpression. A maximum latenO, increase of I. 7 ms was observed. The time in mitlutes (after the start of the procedure) at which each waveform was obtained is at the right of that waveform.

a final waveform obtained after the dura was closed. Peaks I, II, and V could be identified for all three waveforms in 51 patients. For the remaining patients, either a peak could not be identified or the waveform was lost before measurements could be made. Analysis of variance was used to verify that significant differences existed among the changes in latency in these 51 patients. The data were analyzed by analysis of variance using Lotus 1-2-3 spreadsheet software on a Texas Instruments Professional Computer and the SCSS statistical package on the University of Pittsburgh DEC-10. Data were transferred between the computers as ASCII files sent over telephone lines. RESULTS A typical sequence o f changes in evoked potentials seen during monitoring in one o f the 51 patients is illustrated in Figure 2. The responses were obtained during a

seventh nerve decompression. For a baseline, two responses were superimposed, as described in Methods. At the end of the procedure, two responses were again obtained and superimposed. The other waveforms shown are large averages obtained with the monitoring algorithm during the procedure. For the first hour, during which the skull and dura were opened, little change in latency occurred. During manipulation of the nerve a rapid increase in latency was seen. The retractors were repositioned, and the response gradually returned toward the preoperative baseline. We attempted to monitor 117 procedures. These included 95 for trigeminal neuralgia, 11 for hemifacial spasm, and 11 miscellaneous procedures. Table 1 summarizes the outcome of this monitoring. In 2 cases we had technical problems with the transducers, and in 3 cases we were unable to obtain baseline responses, probably because of preexisting hearing deficits. In 41 of the 112 procedures that were actually monitored, there was no latency change or a total change of less than 0.3 ms. In the remaining 71 procedures changes in latency occurred that prompted warning to the surgeon; in 19 of these more than one warning was given. The changes included gradual increases in latency of more than 0.5 ms and rapid increases of more than 0.3 ms. Whenever feasible, retractors were repositioned or surgical approach was modified after an increase in latency. In 27 procedures latency decreased within 3 minutes of the first warning, and in 9 it decreased within 5 minutes. In 8 procedures no decrease in latency was seen for more than 30 minutes but some decrease was always seen eventually. In 27 of the 71 procedures there were no further increases in latency after the first warning was given and before a decrease occurred. In the remaining 44, however, the latency increased further before it began to decrease. That is, an initial increase in latency occurred that prompted a warning, and another increase occurred before the latency began to decrease. These additional increases were 0.3 ms or less in 19 of 44 cases and more than 0.3 ms in 25 cases. Generally, rapid increases in latency resolved rapidly and gradual increases resolved gradually. The BAEP disappeared during the procedure in 4 patients. In each case a warning was given and changes in surgical technique were made. The evoked potentials returned in 3 patients in 5, 7, and 10 minutes, respectively. In the fourth patient, all response peaks past peak II were lost and did not return during the procedure. Eighty-nine patients underwent preoperative and postoperative audiometric testing. In 1 patient the audiogram showed a postoperative hearing improvement; during the procedure this patient had shown a large

Boston et ah Monitoring of BAEP

165

Table I. hltraoperative Results and Postoperative Outconze of Monitoring Brainstem Auditory Evoked Potentials in 112 Patients

Increase in latency Return toward baseline latency Warning given Postoperative hearing loss

No. Patients (N = 112)

Audiometric Testing~ (N = 89)

Complete Data Obtained b (N = 51)

Postoperative Hearing Loss (N = 18)

78 72

63 59

42 40

16 15

71 17

56 17

36 10

13 ...

"Testing done both preoperatively and postoperatively. bldentification of peaks 1, 111,and V before and during opening and after closing of the dura.

Table 2. Changes in Mean Peak Latency and hlterpeak Interval ~

Peak I Maximum change from baseline latency (ms) Recovery from naaximum change (ms)

Peak I-Ill Interval

0.43 b

0.32 b

(0.04) 0.22b (0.04)

(0.06) 0.04 (0.05)

Peak III-V Interval 0.04 (0.03) 0.07 (0.04)

"Data arc based oll results from 51 patients in whom peaks I, 111,and V could be identified before, and after opening of the dura, and after closing of the dura. Estimates of standard error arc in parentheses. bSignificantly different from zero (p < 0.05).

increase in latency, which had recovered after a change in retractor position, Seventeen o f the patients who were monitored and one o f the patients who was not monitored were reported to have suffered some hearing decrement (a pure tone threshold change of at least 10 dB between 250 and 4,000 Hz). Five o f these patients had conductive hearing loss accompanied by impaired tympanic membrane mobility. Twelve had hearing loss that was between 5 and 20 dB and was consistent with cochlear or retrocochlcar dysfunction. T w o of the 3 patients whose evoked potentials had been temporarily lost during operation wcrc in this group. On routine postoperative follow-up visits these hearing losses appeared to have resolved in all 17 patients. The only patient who had a major change in hearing was the one whose evoked responses had been irretrievably lost during the procedure. This patient's diagnosis was atypical right trigeminal neuralgia with V2 distribution. A preoperative audiogram had shown a 20 dB bilateral sensorineural loss with 96% speech discrimination. A postoperative audiogram showed an additional loss o f 20 to 40 dB in the ipsilateral car at frequencies o f 250 and 2,000 Hz and 60 dB at 4,000 Hz; the speech discrimination score was 26%. There was no change in

the contralateral ear. A second postoperative audiogram two days later showed that the additional loss had decreased to 10 dB, except at 4,000 Hz, where it was 50 riB; the speech discrimination score was 52%. The hearing loss appeared to resolve over time; at a follow-up exam one year later the patient's hearing was noted to be intact. Patients with postoperative hearing deficits on audiometric testing tended to have a slightly larger intraoperative increase in latency than did those without postoperative deficits. The range o f change in maximal peak V latency for patients without postoperative deficit was 0 to 2.1 ms, and the range for patients with postoperative deficit was 0.4 to 2.1 ms. However, the difference between the mean values o f the change (1.0 ms for patients with a deficit versus 0.8 ms for those without) was found insignificant by analysis o f variance. The change from baseline in peak V latency at the end o f the procedure had the same range and mean value for both groups. Table 2 summarizes the results seen in the 51 patients in w h o m peaks I, III, and V could be identified before and after opening o f the dura, and after closing o f the dura. In 45 patients the maximum increase in peak I latency was greater than 0.1 ms; in 6 the increase was smaller. In contrast, an increase o f more than 0.1 ms in the peak I-peak III interval occurred in only 34 patients, and 17 showed a smaller increase or none at all; the mean increases in latency, however, were significantly different from zero. Only 15 patients had a change in the peak Ill-peak V interval. Although peak I latency in this group never decreased, both the peak I-III and the peak III-V intervals decreased in a few patients. In some patients changes occurred only in peak I, and in others they occurred in the peak I-III interval. Still others had changes in both. Peak I latency returned significantly toward baseline, but the peak I-III and peak III-V intervals did not.

166 Journal of Clinical Monitoritk~ I/ol 1 No 3 July I985

DISCUSSION

In most cases o f rapid increase in peak V latency, repositioning o f retractors or changes in surgical approach at least partially reduced latency. There is little question that these reductions resulted from surgical manipulation. We used measurement of peak V latency for monitoring for several reasons. It is the easiest peak to identify and measure. Changes in earlier peaks are reflected in peak V. Thus, we tend to look at it first and to have the most confidence in it. The fact that we were able to identify all three of the measured peaks at three different times in fewer than half the patients confirms the difficulty o f identifying these peaks. In general, with BAEPs, an increase in the latency of peak V usually indicates pressure on or stretching of the eighth cranial nerve. A decrease in latency after an increase probably indicates that the pressure or stretching has been relieved. We saw systematic changes in both peak I latency and the peak I-III interval. These changes reflected the sum of both gradual and rapid increases that were observed during a procedure. An increase in peak I latency (seen in 45 of 51 patients) was the most common change; latency usually returned toward baseline by the end of the procedure. This increase may result partly from cooling. BAEPs are known to be sensitive to temperature [9], and we observed increases of several tenths of a millisecond in latency when room temperature saline was used for irrigation. We have been unable to show a significant dependence o f peak V latency on body core temperature, but it is likely that exposure of the eighth nerve to room air causes significant local cooling that is not directly related to core temperature. A gradual increase in latency was common at the beginning of a procedure, and this change would be consistent with local cooling. Changes in the peak I-III interval (seen in 34 of 51 patients) were less common and slightly smaller. H o w ever, the peak I-III interval was less likely to return to baseline after closing o f the dura. Using a direct recording technique, Moller et al [5] showed that peak II of the BAEP is probably generated at least partly by the peripheral auditory nerve. Hence, change in the peak I-III interval could also be due to manipulation o f the eighth nerve or to cooling. However, since the change did not occur as consistently as the increase in peak I latency and showed little recovery, it seems less likely to be caused by cooling. We saw a few instances in which the BAEP changed

so rapidly that the large average did n o t show a response, even though a response could be seen in each individual average. In these instances, conventional averaging also would not have shown a response. This rapid change did not continue for more than 2 or 3 minutes; the individual averages were used for monitoring during these intervals. Patients whose BAEPs were lost even temporarily during a procedure were likely to have a postoperative audiometric deficit. Patients with a postoperative deficit tended to have had a large increase ill latency. The most severe hearing loss in our series was in the patient whose BAEP was lost and not recovered during the procedure. However, because of the large variability among patients and the overlap ill changes ill latency between patients who had a postoperative hearing deficit and those who did not, we were unable to show ally statistically significant difference between the two groups ill changes in latency. The block averaging algorithm rapidly detected changes ill the BAEP waveform during intraoperativc monitoring. The changes were usually due to surgical manipulation and could be at least partially corrected by modification of surgical technique. The control interval provided a mechanism for evaluating the reliability of the response. Large, abrupt changes could be detected in the individual averages. When the BAEP itself was not changing rapidly, the average of four averages provided a sufficiently stable response for automatic tracking of peak V latency and amplitude. We observed systematic intraoperative change ill both peak I latency and ill the peak I-Ill interval. Peak l latency increased most commonly and usually returned toward baseline by the end of the procedure. Changes in the peak I-Ill interval were less common and smaller, but the peak I-IIl interval was also less likely to recover. Twenty percent of patients showed a hearing decrement on postoperative audiometric testing. Patients whose BAEPs had been lost, even temporarily, during the procedure were likely to have a postoperative hearing decrement. Patients with a deficit tended to have a greater increase ill latency than patients without a deficit, but the difference in the mean increases of the two groups was not statistically significant. In all patients the hearing deficits resolved over time. Patients can have a large intraoperative increase in latency without suffering a postoperative hearing decrement, or they can incur a mild, temporary postoperative hearing deficit even with a relatively small intraoperatire increase in latency. By detecting early changes in hearing, however, BAEP monitoring during retromastoid craniectomy apparently helps avoid catastrophic

Boston et al: Monitoring of BAEP

hearing loss. In addition, since all paticnts in our series who had postoperative hearing decrements had had increases o f 0.4 ms or more in latency, it may bc possible to reduce the incidence o f temporary dcficits if increases can bc kcpt below this value. Complctc monitoring was achicvcd in 112 patients. Complete data for analysis after the procedure was ovcr were obtained in 51 patients. These data were studied in an effort to dctcrminc where the observed changes in peak V latency occurrcd. U n k n o w n changcs in local temperature prevented us from attributing change in peak V latency solcly to surgical intervention. In addition, since we intervened as soon as we saw changes in pcak V latency it is difficult to relate intraopcrativc changcs with postoperative hearing deficits. It is probable that postopcrativc hearing deficits would havc bccn greater had wc not intervened, but our study cannot establish this hypothesis. The often rapid changes in peak V latency sccn upon surgical retraction and the frequently rapid reversal o f these changes upon repositioning o f the retractors or modification o f surgical approach have convinced us that monitoring these changes is worth the considerable effort. REFERENCES 1. Jewett DL, Williston JS: Auditory-evoked far fields averaged from the scalp of humans. Brain 1971;94:681-696 2. Buchwald JS, Huang ChM: Far-field acoustic responses: Origins in the cat. Science 1975;189:382-384 3. Achor LJ, Starr A: Auditory brain stem responses in the cat. I. Intracranial and extracranial recordings. Electroencephalogr Clin Neurophysiol 1980;48:154-173 4. Starr A, Hamilton A: Correlation between confirmed sites of neurological lesions and abnormalities of farfield brain stem responses. Electroencephalogr Clin Neurophysiol 1976;41:595-608 5. Moller AR, Jannetta pJ, Bennett M, Moller MB: lntracranially recorded responses from the human auditory nerve: New insight into the origins of brain stem evoked potentials (BSEPs). Electroencephalogr Clin Neurophysiol 1981;52:18-27 6. Moiler AR, Jannetta pJ; Evoked potentials from the inferior colliculus in man. Electroencephalogr Clin Neurophysiol 1982;53:612-620 7. Raudzens PA, Shelter AG: Intra-operative monitoring of brain stem auditory evoked potentials. J Neurosurg 1982;57:341-348 8. Grundy BL, Jannetta PJ, Procopio PT, et al: lntraoperative monitoring of brain-stem auditory evoked potentials. J Neurosurg 1982;57:674-681 9. StockardJJ, StockardJE, Sharbrough FW: Nonpathologic factors influencing brainstem auditory evoked potentials. Am J Electroencephalogr Technol 1978;18:177-204 10. Boston JR: Effects of digital filtering on the waveform and peak parameters of the auditory brainstem response. J Clin Eng 1983;8:79-84

167

11, Boston JR, Deneault LG: Sensory evoked potentials: A system for clinical testing and patient monitoring. Int J Clin Monit Comput 1984;1:13-19 12. Boston JR: An algorithm for continuous monitoring of sensory evoked potentials. J Clin Monit 1985;1:201-206