Auditory Evoked Potentials
Effect of Hearing Loss of Cochlear Origin on the Auditory Brain Stem Response Neil T. Shepard, PhD; John C. Webster, MD;* Midge Baurnen, BA; Pamula Schuck, BS Vestibular Testing Center, Department of Otolaryngology and Head and Neck Surgery, University of Michigan, Ann Arbor, Michigan (N. 1,s); Providence Hospital, Southfield, Michigan (J.C.W.); and Division of the Otology Research Laboratory, Department of Otolaryngology, Henry Ford Hospital, Detroit, Michigan (M.B., PS.)
ABSTRACT Auditory brain stem response (ABR) testing is widely used to detect lesions of the auditory neural pathways. The ABR waves depend not only on the integrity of the neural pathways, but also on the condition of the cochlea. To properly interpret the ABR response, it is necessary to understand the effects of cochlear hearing loss on the ABR wave latencies. We studied two populations of subjects with cochlear hearing loss: one with varying degrees of high-frequency hearing loss and the other with varying degrees of flat configuration hearing loss. The degree of cochlear hearing loss was quantified in several different ways and subjected to one linear and three nonlinear regression analyses to test for accuracy in predicting ABR wave latencies and interpeak intervals (waves I, 111, V, I-V, 1-111, and Ill-V) for three click intensities. Hearing loss levels from 2 to 6 kHz, in particular 4 kHz, were superior to other audiometric test frequencies as predictors of ABR wave latencies for the group with the high-frequency losses. No particular characterization was found to be superior for the flat hearing loss configurations. From these results, modeled predictions of wave latencies as a function of degree and configuration of hearing loss were made. The modeled predictions are then used to suggest guidelines for interpretationsof ABR results where hearing impaired patients are involved. (Ear Hear 13:173-180)
A PRIMARY USE OF ABR testing is the identification of patients with cerebellar pontine mass lesions, more specifically, space occupying lesions of the VIIIth nerve. A large percentage of these patients have some degree of unilateral or bilateral sensorineural hearing loss. A knowledge of the effects of sensorineural hearing loss on the ABR is needed if test performance is to be maximized (i.e., reduction in false-positive identificaEar and Hearing, Vol. 13, No. 3,1992
tion as low as possible without affecting hit rate). This issue has received some attention in the literature over the past decade. In general, the results across studies show the following: (1) some effect on the wave I and V latencies with high-frequency hearing loss of cochlear origin (HFHL) at or greater than 60 dB HL in the 2 to 6 kHz range (Coats & Martin, 1977; Gorga, Worthington, Reiland, Beauchaine, & Goldgar, 1985b; Selters & Brackman, 1977); and (2) when the latency intensity function was investigated, the change in slope of wave V versus click intensity was found to be greater for HFHL subjects than for normal-hearing subjects (Coats & Martin, 1977; Gorga et al, 1985b). Attempts have been made to characterize the status of the cochlea between degree of hearing loss and wave latency changes. The audiometric test frequency of highest correlation typically centers around 4 kHz (Coats & Martin, 1977; Gorga, Reiland, & Beauchaine, 1985a; Gorga et al, 1985b; Montandom, Megill, Kahn, Peake, & Gang, 1975a). Most studies have investigated the effect of hearing loss on wave latencies with HFHL subjects [re: normalhearing groups] (Gorga et al, 1985b) and reviewed the effects of sensorineural loss of both HFHL and flat hearing loss of cochlear origin (FLATHL) configuration on wave V latency. They reported that (a measure of) the slope of the latency-intensity function for wave V was greater for HFHL than for normals, but was less than normals for the FLATHL group. They suggested that the results are a reflection of the cochlear region that predominates in the ABR potentials recorded for different click intensities. Effects of HFHL of cochlear origin on the latencies of waves I and V have been reported (Coats & Martin, 1977), suggesting a rather predictable effect of increasing hearing loss on the wave V latency using a linear regression model and loss at 8 kHz. For wave I, there is a discontinuity in latency at 50 to 60 dB hearing loss which causes a poor fit with linear regression models. This result would be predicted from findings in which filtered click stimuli were used to assess the contributions of the various regions of the cochlea to both ear canal and vertex recordings on normal-hearing subjects (Coats, Martin, & Kidder, 1979). Other investigations, delineating the various frequency region contributions to the ABR by the use of high-pass masking, may also * The authors regret that J.C.W. passed away before publication of this article.
0196/0202/92/1303-0173$03.00/0 * EARAND HEARING Copyright Q 1992 by Williams & Wilkins * Printed in the U.S.A.
Cochlear Hearing Loss and ABR
173
be interpreted to predict the effects of HFHL of cochlear origin on waves I, 111, and V. However, these do not indicate any significant differential effects on wave latencies (Eggermont & Don, 1982). A complete comparison of the effects of HFHL and FLATHL of cochlear origin on wave latencies and interpeak intervals using linear and nonlinear models has not been reported. The purpose of this study was (1) to characterize the effects of HFHL and FLATHL of cochlear origin on the latencies of waves I, 111, and V and the interpeak intervals; 2) model the behavior of the wave latencies; 3 ) utilize the data for use in ABR interpretation of identification of cochlear versus retrocochlear site of lesion. METHODS
Two groups of subjects underwent ABR testing using an electrical pulse stimulus (100 psec in duration) with alternating polarity at a presentation rate of I1.3/sec presented through a matched pair of earphones (TDH-49, MX/AR41 cushions). All testing was conducted in a sound-attenuated room. Active electrodes (silver disk style filled with conductive paste) were placed on the vertex and ipsilateral mastoid, with a ground electrode on the contralateral mastoid, and held in place with tape. Impedance measurements between all combinations of electrodes were equal to or less than 3 kohm. Raw EEG was band-pass filtered from 150 to 3000 Hz (at -3 dB points). Dedicated averager units (Nicolet 1070, CA- 1000) were utilized to produce the averaged responses from the FLATHL and HFHL subjects, respectively. The averaged responses consisted of approximately 2048 sweeps with about 256 data points at 95, 85, and 75 dB attenuator settings (a dial setting of 10 dB produced a click intensity that was the average click threshold for normal-hearing adults; i.e., 0 dB nHL, therefore, 95 dB dial setting was 85 dB nHL; 0 dB dial setting = 20 dB pe SPL [re: 20 pPa]). A minimum of three points on the latency intensity function was obtained for each ear, testing the ear with the better thresholds first in descending order of click intensity. The two experimental groups were (1) a HFHL group consisting of 10 females and 2 1 males, with the predominant etiology noise exposure. All ears in the HFHL group had normal thresholds at 250 Hz and thresholds of 25 dB HL or less at 500 Hz, except for one with a loss of 40 dB HL at 500 Hz. Thresholds above 500 Hz ranged from mild sloping sensorineural loss of sensitivity from 3 through 8 kHz to a moderate to profound sloping configuration beginning with 1 kHz. All HFHL subjects had measurable thresholds at all frequencies. (2) A FLATHL configuration consisting of 34 females and 18 males, all with thresholds that showed no more than a 25 dB variation in threshold from 250 to 8000 Hz. The thresholds in the FLATHL group ranged from a mild to a severe loss of sensitivity. Like the HFHL group, all of the FLATHL subjects had measurable thresholds at all frequencies and no evidence of middle ear involvement. All 52 ears from the FLATHL group were diagnosed with MeniCrSs disease or cochlear endolymphatic hydrops. The subjects ranged in age from 20 to 65 yr and were all selected based on the following criteria: (1) no retrocochlear involvement as indicated by otoneurological investigation other than ABR; (2) no evidence of middle ear involvement; and (3) no evidence of progressive or fluctuant hearing within 6 mo of their participation. A group of 24 young volunteers ranging in age 174
Shepard et al
from 18 to 35 yr with audiometric thresholds equal to or less than 15 dB HL at all audiometric test frequencies were used to establish the behavioral click thresholds and the normative range of absolute latencies for ABR waves I, 111, and V and the interpeak intervals, The same data collection protocol was used on the normal-hearing subjects as used on the hearingimpaired subjects. Analysis of ABR data consisted of regressing the latencies (measured by hand from plots of the averaged ABR waveforms) of waves I, 111, and V and the interpeak intervals, 1-111, 111-V, and I-V individually with all of the FLATHL subjects’ hearing loss thresholds at 1, 2,4, and 8 kHz and the average loss at 1 to 2, 1 to 4, 1 to 8, 2 to 4, 2 to 8, and 4 to 8 kHzfor each click level used. This same analysis was used for the HFHL group with the addition of hearing loss thresholds at 3 and 6 kHz measured and used in the regression process. Four regressions of the above type were performed, one each with the following models: (1) Linear fit to the data, (2) a second-order polynomial fit, (3) a log,,, of the latency with a linear fit, and (4) a third-order polynomial fit (see Appendix for details of the models used in the regression analysis). The r2 values in percentage equals the square of the Pearson correlation coefficient between the observed ABR latencies and predicted ABR latencies from the model prediction equations times 100. This is interpreted as that percentage of the variation in the predicted latencies that can be explained by means of the prediction equation. That is, the percentage of variation in the particular ABR latency or interpeak interval that can be explained given the hearing loss information. The level (pvalues) of statistical significance of the correlation coefficients and r2 values were used to compare relative effectiveness of the four models and hearing loss threshold as a prediction parameter. A two-tailed t-test of significance was used to develop the p-values for the correlation coeficents. RESULTS
Absolute Latencies Determination of the hearing test frequency or frequencies that best predict changes in the absolute ABR latencies was developed by the use of the p-values for the correlation coefficient and r2 information. A sample of these data are shown in Tables 1 to 3, giving the p values and the r2 values for the HFHL and FLATHL groups at all of the hearing loss predictor frequencies for the third-order polynomial model at the 95 dB attenuator setting. Waves V, I11 and I are given in Tables 1, 2, and 3, respectively. A contrast in predictability for a click level of 95 dB attenuation setting (Table 1) is noted for the HFHL and FLATHL groups. For the HFHL group, the third-order polynomial model was superior for waves I11 and V with both the secondand third-order polynomial models superior for wave I. For waves I11 and V, the best predictor frequency was at 3 to 4 and 4 kHz, respectively, for the HFHL group (Tables I and 2). Note that although the FLATHL group shows statistically significant correlation coeficients, the r2 values for the HFHL were larger than those for FLATHL (Tables 1-3). Wave I (Table 3) for the HFHL shows a broader dependence on threshold frequency for prediction, showing the best results for the 2 to 8 kHz combination. Ear and Hearing, Vol. 13, No. 3,1992
Table 1.p-Values for correlation coefficients and r 2 values 95 dB wave V third-order polynomial model.
Hearing Loss Predictor Frequencies in kHz
1 2 3 4 6 8
HFHL
p-values
FLATHL
[r2%1
p-values
[rzo/0]
NS [22.8] NS [10.1] 0.02< p 5 0.05 [47.3]0.005< p I0.01 [23.2] 0.01< p 5 0.02 [50.8] p < 0.005 [82.6]0.005< p 5 0.01 [26.6] NS [24.2] NS [26.1]0.005< p 5 0.01 [21.9]
2-8
4-8
NS NS NS 0.02< p 5 0.05 0.02< p 5 0.05 NS
[11.8] 0.01< p 5 0.02 [17.2] [31.7]0.005< p 5 0.01 [24.1] [32.8] p < 0.005 [26.4] < 0.005 [27.7] [44.9] [40.0] p < 0.005 [28.9] [26.4] p < 0.005 [27.0]
NS,not statistically significant, p > 0.05.
1 2 3 4 6 8
HFHL
p-Values
FLATHL
[r2%]
p-Values
0.01 [71.2] 0.01 < p [39.4]0.005 < p p < 0.005 [85.7] [86.9]0.005 < p p < 0.005 NS [41.4] 0.005< p I 0.01 [74.7]0.005 < p 0.005< p
5
NS
5
[r20/o]
0.02 [14.9] [14.9]
I0.01 I0.01
[26.5]
I0.01
[21.7]
NS [3.0] 0.02< p I0.05 0.02< p I0.05 [45.7]0.005< p 5 0.01 0.01 < p d 0.02 [56.8]0.005< p I0.01 0.005 < p I0.01 [75.0]0.005< p I0.01 0.005< p 5 0.01 [71.7]0.005< p 5 0.01 0.01 < P I 0.02 155.71 0.005 < P 5 0.01
[12.6] [18.7] [20.8] [23.0] [23.6] [24.41
Average of
1-2 1-4 1-8 2-4 2-8 4-8
1
2 3 4 6 8
HFHL
p-Values
0.01 < p 0.02< p NS NS NS NS
I0.02 I0.05
FLATHL
IrzYol
p-Values
rr2%1
[66.3] 0.02< p I0.05 [19.7] [56.6] 0.02< p 5 0.05 [16.8] [38.3] [5.9]0.005< p I0.05 [31.l] [30.2] [39.8]0.005c p I0.01 [25.3]
NS, not statistically significant, p > 0.05.
The FLATHL group correlation coefficient p-values and r2 results showed no clearly superior model for fit to the experimental data and no indication of a threshold frequency or combination of frequencies that adequately predicts latency behavior (Tables 1-3). These findings are evident for waves I, 111, and V for all three levels of click stimuli investigated. Correlation coefficient pvalues and r2 values for wave V for the HFHL group, as intensity of the click stimulus was changed from 95 to 75 dB, indicated that the predictor superiority of 4 kHz at 95 dB was no longer seen. It was replaced by a more broadly based region (2-8 kHz) of predictor frequency. This change also holds, in general, for waves I11 and I as click level was decreased. Based on the correlation Coefficient p-values and r2 data, a model was developed for wave latency behavior Ear and Hearing, Vol. 13,No. 3,1992
1-2 1-4 1-8 2-4 2-8 4-8
NS [48.1] 0.02< p I0.05 [17.1] NS [28.8]0.005< p I0.01 [24.0] NS [29.6]0.005< p 5 0.01 [25.5] NS [38.5] 0.01< p I0.02 [22.4] 0.005 < p 5 0.01 [96.4]0.005< p I0.01 [25.2] NS [31.l]0.005< p 5 0.01 [26.8]
NS, not statistically significant, p > 0.05.
Table 2. p-Values for correlation coefficients and r 2 values 95 dB wave 111 third-order polynomial model.
Hearing Loss Predictor Frequencies in kHz
Hearing Loss Predictor Frequencies in kHz
Average of
Average of
1-2 1-4 1-8 2-4
Table 3. p-Values for correlation coefficients and r 2 values 95 dB wave I third-order polynomial model.
versus degree of hearing loss. For HFHL, the basic assumptions were: ( I ) the best predictor frequency was 4 kHz for waves I11 and V,with 2 to 8 kHz average for wave I; (2) the experimental data were best fit by the third-order polynomial model. For FLATHL, these were: (1) given no clearly superior predictor frequency for the FLATHL group, it was decided to remain consistent with those chosen for the HFHL group; (2) no clear regression model was best, so the simplest, linear, was chosen for use. Application of the described models to the HFHL and FLATHL group’s audiometric data at the specified individual or combination frequencies was used to predict the mean ABR latencies and to predict +2 SD above the predicted mean latencies for waves I, 111, and V and interpeak intervals 1-111, 111-V, and I-V. This was done for click stimulus intensities of 95, 85, and 75 dB attenuator settings. Figure 1 provides a summary of the findings produced by application of the models described. Shown in Figure 1 is the wave latency delay in milliseconds above the normal-hearing upper limit cutoff (+2 SD above the mean for normal-hearing subjects) plotted versus hearing lossdegree. Only latencies greater than the cutoff were plotted, otherwise data are zero. The figure illustrates the following: (1) The rate of growth of the wave latencies outside the normal range is consistently greater for HFHL group than for the FLATHL group. (2) As click level is decreased (95 dB compared to 85 and 75 dB together), the effect on wave V latency increases for a given degree of loss for either hearing loss configuration. (3) For a given click level (95 dB), wave V appears less effected by a degree of loss than waves I11 and I for either hearing loss group. Interpeak Intervals For the HFHL group at 95 dB, the 2 to 8 kHz frequency combination and the third-order polynomial model gave the lowest p-values for the correlation coefCochlear Hearing Loss and ABR
175
w
1.0-
nl c
g
1.1-
2
'
1: .- 1 ;o.5:3 2
.
E
& ,
9
*--
0.0
B "w
P
'
ln X
ln - 0 . 5 .
0.0
'
-0.5.
.
-1.0-
-1.0-
U
U
0.5. E
B 6
.
* ln
--_----
0.0.
-0.5.
-1.0-
__--__---
L
r"
"
0.1.
0 w
4 ln
X
-0.5.
-1.0-
Figure 1. Wave V at click levels of 95, 85, and 75 dB in the top three panels showing the number of milliseconds above the normal-hearing subject upper limit of wave V latency as a function of degree of hearing loss. In the bottom left panel, the same information is provided for wave 111 and for wave I in the bottom right panel. Both bottom panels are for click levels of 95 dB. In all panels, the HFHL group is shown by a solid line, with the FLATHL group indicated by the dashed line.
ficient in the I-V interval ( p < 0.005). The 4 kHz frequency and the third-order model were found superior for the 111-V interval at 95 dB ( p < 0.05). At 75 dB for the HFHL population, no predictor frequency or model was found to be superior. As with the individual latencies, no superior predictor frequency was found for the FLATHL. The third-order model did show slightly lower p-values for the correlation coefficient for the I-V and 111-V intervals at 95 dB, with no difference in the models at 75 dB. Figure 2 (a-d) shows the individual interpeak intervals and the predicted mean intervals from the thirdorder model regression (HFHL) and linear model regression (FLATHL). The 2 to 8 kHz predictor frequencies are used for the I-V intervals and 4 kHz predictor frequency is used for the 111-V intervals. HFHL group results for a 95 dB click level are shown in Figure 2 (a and b), with the FLATHL group in Figure 2 (c and d). For the HFHL I-V interval, a definite trend for a reduction in the interval was noted as hearing loss increased (also true for a linear model, not shown). DISCUSSION
Comparison to Other Studies of Hearing-Impaired Subjects The results demonstrate that a loss of hearing of cochlear origin can cause prolongations in latencies of 176
Shepard et al
waves I, 111, and V beyond that for normal hearing. It is evident that the specific effect on latency is related to the configuration of the loss. This concept was presented by Coats and Martin (1977) and more recently in an article by Gorga et a1 (1985a, b). Gorga et a1 (1985b) found a significantly lower slope to the wave V latency intensity function for FLATHL compared to that for the HFHL. It would be anticipated that configuration of the loss (HFHL versus FLATHL) would correlate with the locus of lesion site within the cochlea, especially in those ears with a more severe loss. Therefore, the configuration could perhaps be interpreted to indicate those regions of the cochlea most likely dominating or participating in the production of the ABR. A HFHL would suggest a progressive loss of basal influence with increasing degree of loss and, therefore, an increasing dependence on the more apical areas. With decreasing click intensity, fewer basal fibers would respond, with more of the response coming from apical fibers. This would cause a rapid growth in latency with decreasing click intensity. In contrast, a FLATHL would provide for participation by all cochlear regions, giving a rate of latency growth parallel to that of the normal cochlea and a rate of growth less than that for HFHL. Our results are consistent with both this mechanism and the previous findings. Our results indicate that a cochlear hearing loss must reach 60 to 65 dB HL for our 85 or 75 dB click to cause Ear and Hearing, Vol. 13, No. 3,1992
a
C HFHL 95 I - V I N T . MODEL ( 3 r d o r d e r poly.]
v, E
5.0
=
-ft I
.A
FLATHL 95 I - V I N T . MODEL (linear]
I
c .r(
I
r(
m
> L w
I
4.0
cr
w C
>
* 3.0
3.0-1, 0
20
40
aa
60
Hearing Loss a t 2k-EkHz
[in
100
0
dB HLI
.
, 20
,
. 40
,
,
. . . ,I eo 100 l i n dB HLI
60
Hearing Loss a t 2k-BkHz
b
d HFHL 95 111-V I N T . MODEL ( 3 r d o r d e r poly.1
-c In
L
3.0
In
E
E
FLATHL 95 111-V I N T . MODEL (linear] 3.0
-
c
-
-n
.r(
m >
m
5 L W
$
-
L
W
g
2.0
2.0
w
W
>
>
H
U I
I
I+
w
r(
U
1.0
I
1.0
0
20
40
60
eo
I
loo
Hearing Loss a t 4kHz [ i n dB HLI
r)
Figure 2. Interpeak interval actual data points versus degree of loss for a frequency combination of 2 to 8 kHz and an individual frequency of 4 kHz, and either the third-order model predictionof interval for HFHL or linear model predictionof interval for FLATHL are shown. Panels a and b give the data for HFHL at 95 dB I-V and Ill-V intervals, respectively. Panels c and d give the same data for the FLATHL group.
any appreciable prolongation in the latency of the three waves studied. This is generally consistent with other investigations, which show the same degree of loss required to produce an effect on wave V latency (Coats & Martin, 1977; Gorga et al, 1985b; Selters & Brackman, 1977; Sohmer, Kinarti, & Gafni, 1981). Agreement is found across other studies with the result that the 4 kHz or 2 to 8 kHz regions provided the maximum predictability of wave V latency variation for HFHL when threshold is used as the available characterization of the status of the cochlea. For the FLATHL configuration, our results demonstrated no superior hearing test frequency for prediction of wave V latency behavior. It is suggested that this may reflect a broad locus of damage in the cochlea leaving intact the relative contributions to the responses based on synchronous activity of the VIIIth nerve. This would be in contrast to the HFHL configuration, where the typical dominating influence of the basal cochlear region is lessened as either click intensity is decreased or the magnitude of the loss is increased, thus causing a disruption in the Ear and Hearing, Vol. 13, No. 3,1992
synchronousfiring of the high-frequency region relative to the undisturbed activity of the apical regions.
Comparison to Normal System Studies The present results are consistent with predictions from studies of normal ears where wideband noise or derived narrowband responses are used to restrict the regions of the cochlea available to participate in the generation of auditory evoked potentials. As wideband noise is introduced with a click stimulus, the slope of the latency intensity function for the VIIIth nerve compound action potential (intracanicular component corresponding to wave I of the ABR complex) is found to be unchanged from that without the noise, even though the absolute latencies are prolonged (Montandon et al, 1985a; Montandon, Shepard, Marr, Peake, & Gang, 1985b). As contribution of the basalward cochlear regions are selectively removed by use of filtered clicks or derived responses, wave latencies progressively increase (Coats et al, 1979; Don & Eggermont, 1978; Eggermont & Don, 1980, 1982). For filtered clicks, the rate of Cochlear Hearing Loss and ABR
177
change in the latency intensity function for waves I, 111, and V appears to increase as the cochlear region of dominant contribution decreases below 2 kHz. This effect is not shown as clearly for the derived response paradigm using high-pass noise, but is suggested by the data (Eggermont & Don, 1980). Using these results, it could be hypothesized that a selective loss of basal contribution (HFHL) to the ABR would produce an increase in latency beyond that seen for a normal system. Also, the slope of the latency intensity function would be greater than that for normal hearing or a flat loss, as was shown with the present results. From the filtered click studies, the slope of the latency intensity function for wave I is greater than that for waves 111 and V because the contribution from the basal end of the cochlea is reduced (as intensity is lowered), resulting in reduced neural synchrony (Coats et al, 1979). This would predict the present result that indicates a decrease in the I-V and 1-111 intervals as the degree of the HFHL is increased. It may be hypothesized that this effect is, in part, related to the decreasing dependence on synchronous firing of the respective generating sites for waves I, 111, and V. Therefore, in the case of the FLATHL group with some preservation of the relative synchronous contributions, we find that the 1-111 and I-V intervals are independent of degree of loss. Predicting Effects of Hearing Loss on Wave V Latency Wave V normative range for HFHL and FLATHL groups can be developed by defining 2 SD above the predicted mean latency for a given degree of loss as the upper boundary for normal. Latency beyond this range would be defined as retrocochlear. Using degree of loss at 4 kHz of 60, 80, and 100 dB HL, a plot of 2 SD above the predicted mean for wave V latency versus click intensity was developed for each degree of loss for both HFHL and FLATHL groups. These plots, along with the upper limit for the normal-hearing male and female norms, are shown in Figure 3a for the HFHL and Figure 3b for the FLATHL. Ideally, normative data of this type are designed to reduce the false-positive rate without a reduction in hit rate. Our clinical experience over a 2Y2 yr period using hearing-impaired norms shows this to be correct (Shepard, Turner, & Burtka, 1986). Obviously, only clinical environments with a referral base with a substantial population of severe hearing loss patients stand to benefit from the use of norms of this type. It is exactly clinical situations of this type that have the need for this process. We suggest that the rate of false positives is proportional to the percentage of patients with significant loss of sensitivity. This may provide an explanation for the wide range of false-positive results (0-38%) reported by Turner, Shepard, and Frazer (1984) after a review of 15 yr of literature. Other measures of cochlear function (e.g., compound action potential tuning curves and psychophysical tun178
Shepard et al
High F r e q u e n c y H e a r i n g L o s s N o r m s
a
WAVE
v
UTENCY
10.0
3.0
8.0 v)
E
7.0
6 0 I
.
1 .
25
45
"
I
T
'
I I
I
7
~
; 1
- -
i
I
1
1
65
I
.
I
1 :
I
II
85
105
d 8 Attenuator Setting
Flat Hearing Loss N o r m s
WAVE V L A W
k Y
10.0
9.0
8.0 In E
7.0 6.0
5.0 25
45
65
85
105
dE Attenuator Setting Figure 3. Normal-hearing subject upper limit of the normal range for females (A-A) and males @€I) for wave V latency versus click intensity in dB attenuator setting. In panel a, the upper limit of the acceptable range for a cochlear loss of sensitivity for a 4 kHz loss of 60,80, or 100 dB HL is given by the dashed lines for HFHL. Panel b shows the same information for a cochlear hearing loss of a flat configuration.
ing curves) have been shown to accurately reflect the status of the cochlea (Shepard & Abbas, 1983). These measures have a higher probability of increased correlation and, therefore, increased predictability of wave latency performance. These measures provide for both loss of sensitivity information as well as frequency resolution capabilities of the cochlea. This may be of Ear and Hearing, Vol. 13, NO. 3,1992
use especially in cases where the cochlear lesion does not manifest itself through significant change in behavioral threshold. The utility of the compound action potential tuning curve or the psychophysical tuning curve for a clinically usable model is limited by the time required to obtain this information. Therefore, the pragmatic compromise is to trade reduction in accuracy for availability of the hearing threshold information. The decision to use hearing thresholds may well be a major reason for the inability to completely account for the increased false-positive test results seen with a severely hearing-impaired population. As stated under “Methods,” the FLATHL group was comprised of only those patients diagnosed as having endolymphatic cochlear hydrops or a specific form of endolymphatic hydrops, Meniirss disease. The possibility that the specific results obtained for the FLATHL group may be pathognomonic to this disorder classification cannot be ruled out, but is unlikely. This group of patients constitutes the largest percentage with cochlear lesions that are suspect for retrocochlear involvement. Their choice for the present study was, therefore, appropriate. CONCLUSIONS
1. We have demonstrated a differential effect of cochlear hearing loss configuration on ABR waves I, 111, and V latencies. 2. For FLATHL configuration, interpeak intervals show no effect with increasing loss of sensitivity. HFHL configuration shows a reduction in the intervals of I-V and 1-111 with increasing degree of loss. The 111-V interval showed no effect. 3. It is hypothesized that an explanation for the differential effects on latency across configuration and within the HFHL configuration for wave I versus waves I11 and V reflect the locus of the lesion within the cochlea and relative disruption of synchronous neural activity of wave sources, respectively. 4. Although use of hearing threshold information is not optimal for development of a model of this type, it is the only practical measure to use as a predictor in a clinical setting. 5 . Normative data were developed for ABR interpretation that would be predicted (and shown by clinical trial under separate reporting) to reduce the false alarm rate without a significant effect on the hit rate in a clinical population when losses of auditory sensitivity are prevalent. REFERENCES Coats CA and Martin JL. Human auditory nerve action potentials and brain stem evoked responses effects of audiogram shape and lesion location. Arch Otolaryngol 1977;103:605-622. Coats CA, Martin JL, and Kidder HR. Normal short-latency electrophysiological filtered click responses recorded from vertex and external auditory meatus. J Acoust SOCAm 1979;65(3):747-758. Don M and Eggermont JJ. Analysis of the click-evoked brainstem potentials in man using high-pass noise masking. J Acoust SOCAm
Ear and Hearing, Vol. 13, No. 3,1992
1978;63(4): 10984-1092. Eggermont JJ and Don M. Analysis of click-evoked brainstem potentials in human using high-pass noise masking. 11. Effect of click intensity. J Acoust SOCAm 1980;68(6):1671-1675. Eggermont JJ and Don M. Analysis of click-evoked brainstem auditory electric potentials using high-pass noise masking and its clinical application. Ann NY Acad Sci 1982;82:471-486. Gorga MP, Reiland JK and Beauchaine KA. Auditory brainstem response in a case of high-frequency conductive hearing loss. J Speech Hear Disord 1985a;50:346-350. Gorga MP, Worthington DW, Reiland JK, Beauchaine KA, and Goldgar DE. Some comparisons between auditory brainstem response thresholds, latencies and the pure-tone audiogram. Ear Hear 1985bl6:105- 1 12. Montandon PB, Megill ND, Kahn AR, Peake WT, and Kiang NYS. Recording auditory-nerve potentials as an ofice procedure. Ann Otol Rhinol Laryngol 1975a;84:2-11. Montandon PB, Shepard NT, Marr EM, Peake WT, and Kiang NYS. Auditory-nerve potentials from ear canals of patients with otologic problems. Ann Otol Rhinol Laryngol 1975b;84:164-173. Selters WA and Brackmann DE. Acoustic tumor detection with brainstem electric response audiometry. Arch Otolaryngol 1977;103:181-187. Shepard NT and Abbas PJ. Compound action-potential tuning curves in normal and acoustically traumatized cats. Ann Otol Rhinol Laryngol 1983;92:496-503. Shepard NT, Turner RG, and Burtka MJ. BAER experience using cochlear hearing impaired normative data. J Acoust SOCAm 1986;81:S2. Sohmer K, Kinarti R and Gafni M. The latency of auditory nervebrainstem responses in sensorineural hearing loss. Arch Otolaryngol 1981;230:189-199. Turner RG, Shepard NT, and Frazer GJ. Clinical performance of audiological and related diagnostic tests. Ear Hear 1984;5:187-194. ~~
Acknowledgment: We would like to express our gratitude to Dr. Robert Turner for his critical review of the manuscript and his helpful suggestions. Address reprint requests to Neil T. Shepard, Vestibular Testing Center, Dept. of Otolaryngology, Box 0816, Rm. C166A, Med-Inn Bldg., University of Michigan, 1500 E. Medical Center Dr., Ann Arbor, MI 48109. Received June 23,1989; accepted February 10,1992.
APPENDIX
Linear Model
Y=a+bx where Y = predicted wave latency or interpeak interval; x = measured hearing loss thresholds in dB HL; and a and b = calculated from the regression analysis for best fit to empirical data by least squares methods.
b = [C(xi - X U i - y)I/[Z(xi- x)’I a=y-bx where x = mean of the hearing loss thresholds in db HL; y = mean of the observed wave latencies for a given wave or interpeak interval across thresholds; (xi, yi) = datum point of hearing loss and wave latency for a given wave I, 111, or V, or hearing loss and interpeak interval; and n = number of data points for each wave latency or interpeak interval. Second-Order Polynomial Model Y = bo + bix
+ b2X2
Cochlear Hearing Loss and ABR
179
where Y = predicted wave latency or interpeak interval; x = measured hearing loss threshold; and bo, bl, bz = calculated coefficients from the regression analysis. Transform Model Y‘ = a’
+ b‘x
where Y’ = predicted latency or interpeak interval after transforming of observed latency or interpeak interval values; x = measured hearing loss thresholds in dB HL; and a’, b’ = calculated by regression analysis after first transforming the observed latency and interpeak interval values as follows:
where x = mean of the hearing loss thresholds in dB HL for a given wave or interpeak interval; y‘ = mean of the transformed observed latency or interpeak interval values; and (x,,y,’) = datum point of hearing loss threshold and transformed latency or interpeak interval value for a given wave or interpeak interval.
Third-Order Polynomial Model Y = ho -C hlx + bzx2 + b s 3
Y’ = logloY b’
=
[C(xi - 2 ) ( ~-’ ,jj’)l/[Cx,- n)’] a’ = j j ‘ - b’2
where y = predicted wave latency or interpeak interval; x = measured hearing loss threshold; and bo, bl, b2, and b3 = calculated coefficients from the regression analysis.
~~
180
Shepard et al
Ear and Hearing, VOI. 13, NO. 3,1992
Effect of Hearing Loss of Cochlear Origin on the Auditory Brain Stem Response Neil T. Shepard, PhD; John C. Webster, MD;* Midge Baurnen, BA; Pamula Schuck, BS Vestibular Testing Center, Department of Otolaryngology and Head and Neck Surgery, University of Michigan, Ann Arbor, Michigan (N. 1,s); Providence Hospital, Southfield, Michigan (J.C.W.); and Division of the Otology Research Laboratory, Department of Otolaryngology, Henry Ford Hospital, Detroit, Michigan (M.B., PS.)
ABSTRACT Auditory brain stem response (ABR) testing is widely used to detect lesions of the auditory neural pathways. The ABR waves depend not only on the integrity of the neural pathways, but also on the condition of the cochlea. To properly interpret the ABR response, it is necessary to understand the effects of cochlear hearing loss on the ABR wave latencies. We studied two populations of subjects with cochlear hearing loss: one with varying degrees of high-frequency hearing loss and the other with varying degrees of flat configuration hearing loss. The degree of cochlear hearing loss was quantified in several different ways and subjected to one linear and three nonlinear regression analyses to test for accuracy in predicting ABR wave latencies and interpeak intervals (waves I, 111, V, I-V, 1-111, and Ill-V) for three click intensities. Hearing loss levels from 2 to 6 kHz, in particular 4 kHz, were superior to other audiometric test frequencies as predictors of ABR wave latencies for the group with the high-frequency losses. No particular characterization was found to be superior for the flat hearing loss configurations. From these results, modeled predictions of wave latencies as a function of degree and configuration of hearing loss were made. The modeled predictions are then used to suggest guidelines for interpretationsof ABR results where hearing impaired patients are involved. (Ear Hear 13:173-180)
A PRIMARY USE OF ABR testing is the identification of patients with cerebellar pontine mass lesions, more specifically, space occupying lesions of the VIIIth nerve. A large percentage of these patients have some degree of unilateral or bilateral sensorineural hearing loss. A knowledge of the effects of sensorineural hearing loss on the ABR is needed if test performance is to be maximized (i.e., reduction in false-positive identificaEar and Hearing, Vol. 13, No. 3,1992
tion as low as possible without affecting hit rate). This issue has received some attention in the literature over the past decade. In general, the results across studies show the following: (1) some effect on the wave I and V latencies with high-frequency hearing loss of cochlear origin (HFHL) at or greater than 60 dB HL in the 2 to 6 kHz range (Coats & Martin, 1977; Gorga, Worthington, Reiland, Beauchaine, & Goldgar, 1985b; Selters & Brackman, 1977); and (2) when the latency intensity function was investigated, the change in slope of wave V versus click intensity was found to be greater for HFHL subjects than for normal-hearing subjects (Coats & Martin, 1977; Gorga et al, 1985b). Attempts have been made to characterize the status of the cochlea between degree of hearing loss and wave latency changes. The audiometric test frequency of highest correlation typically centers around 4 kHz (Coats & Martin, 1977; Gorga, Reiland, & Beauchaine, 1985a; Gorga et al, 1985b; Montandom, Megill, Kahn, Peake, & Gang, 1975a). Most studies have investigated the effect of hearing loss on wave latencies with HFHL subjects [re: normalhearing groups] (Gorga et al, 1985b) and reviewed the effects of sensorineural loss of both HFHL and flat hearing loss of cochlear origin (FLATHL) configuration on wave V latency. They reported that (a measure of) the slope of the latency-intensity function for wave V was greater for HFHL than for normals, but was less than normals for the FLATHL group. They suggested that the results are a reflection of the cochlear region that predominates in the ABR potentials recorded for different click intensities. Effects of HFHL of cochlear origin on the latencies of waves I and V have been reported (Coats & Martin, 1977), suggesting a rather predictable effect of increasing hearing loss on the wave V latency using a linear regression model and loss at 8 kHz. For wave I, there is a discontinuity in latency at 50 to 60 dB hearing loss which causes a poor fit with linear regression models. This result would be predicted from findings in which filtered click stimuli were used to assess the contributions of the various regions of the cochlea to both ear canal and vertex recordings on normal-hearing subjects (Coats, Martin, & Kidder, 1979). Other investigations, delineating the various frequency region contributions to the ABR by the use of high-pass masking, may also * The authors regret that J.C.W. passed away before publication of this article.
0196/0202/92/1303-0173$03.00/0 * EARAND HEARING Copyright Q 1992 by Williams & Wilkins * Printed in the U.S.A.
Cochlear Hearing Loss and ABR
173
be interpreted to predict the effects of HFHL of cochlear origin on waves I, 111, and V. However, these do not indicate any significant differential effects on wave latencies (Eggermont & Don, 1982). A complete comparison of the effects of HFHL and FLATHL of cochlear origin on wave latencies and interpeak intervals using linear and nonlinear models has not been reported. The purpose of this study was (1) to characterize the effects of HFHL and FLATHL of cochlear origin on the latencies of waves I, 111, and V and the interpeak intervals; 2) model the behavior of the wave latencies; 3 ) utilize the data for use in ABR interpretation of identification of cochlear versus retrocochlear site of lesion. METHODS
Two groups of subjects underwent ABR testing using an electrical pulse stimulus (100 psec in duration) with alternating polarity at a presentation rate of I1.3/sec presented through a matched pair of earphones (TDH-49, MX/AR41 cushions). All testing was conducted in a sound-attenuated room. Active electrodes (silver disk style filled with conductive paste) were placed on the vertex and ipsilateral mastoid, with a ground electrode on the contralateral mastoid, and held in place with tape. Impedance measurements between all combinations of electrodes were equal to or less than 3 kohm. Raw EEG was band-pass filtered from 150 to 3000 Hz (at -3 dB points). Dedicated averager units (Nicolet 1070, CA- 1000) were utilized to produce the averaged responses from the FLATHL and HFHL subjects, respectively. The averaged responses consisted of approximately 2048 sweeps with about 256 data points at 95, 85, and 75 dB attenuator settings (a dial setting of 10 dB produced a click intensity that was the average click threshold for normal-hearing adults; i.e., 0 dB nHL, therefore, 95 dB dial setting was 85 dB nHL; 0 dB dial setting = 20 dB pe SPL [re: 20 pPa]). A minimum of three points on the latency intensity function was obtained for each ear, testing the ear with the better thresholds first in descending order of click intensity. The two experimental groups were (1) a HFHL group consisting of 10 females and 2 1 males, with the predominant etiology noise exposure. All ears in the HFHL group had normal thresholds at 250 Hz and thresholds of 25 dB HL or less at 500 Hz, except for one with a loss of 40 dB HL at 500 Hz. Thresholds above 500 Hz ranged from mild sloping sensorineural loss of sensitivity from 3 through 8 kHz to a moderate to profound sloping configuration beginning with 1 kHz. All HFHL subjects had measurable thresholds at all frequencies. (2) A FLATHL configuration consisting of 34 females and 18 males, all with thresholds that showed no more than a 25 dB variation in threshold from 250 to 8000 Hz. The thresholds in the FLATHL group ranged from a mild to a severe loss of sensitivity. Like the HFHL group, all of the FLATHL subjects had measurable thresholds at all frequencies and no evidence of middle ear involvement. All 52 ears from the FLATHL group were diagnosed with MeniCrSs disease or cochlear endolymphatic hydrops. The subjects ranged in age from 20 to 65 yr and were all selected based on the following criteria: (1) no retrocochlear involvement as indicated by otoneurological investigation other than ABR; (2) no evidence of middle ear involvement; and (3) no evidence of progressive or fluctuant hearing within 6 mo of their participation. A group of 24 young volunteers ranging in age 174
Shepard et al
from 18 to 35 yr with audiometric thresholds equal to or less than 15 dB HL at all audiometric test frequencies were used to establish the behavioral click thresholds and the normative range of absolute latencies for ABR waves I, 111, and V and the interpeak intervals, The same data collection protocol was used on the normal-hearing subjects as used on the hearingimpaired subjects. Analysis of ABR data consisted of regressing the latencies (measured by hand from plots of the averaged ABR waveforms) of waves I, 111, and V and the interpeak intervals, 1-111, 111-V, and I-V individually with all of the FLATHL subjects’ hearing loss thresholds at 1, 2,4, and 8 kHz and the average loss at 1 to 2, 1 to 4, 1 to 8, 2 to 4, 2 to 8, and 4 to 8 kHzfor each click level used. This same analysis was used for the HFHL group with the addition of hearing loss thresholds at 3 and 6 kHz measured and used in the regression process. Four regressions of the above type were performed, one each with the following models: (1) Linear fit to the data, (2) a second-order polynomial fit, (3) a log,,, of the latency with a linear fit, and (4) a third-order polynomial fit (see Appendix for details of the models used in the regression analysis). The r2 values in percentage equals the square of the Pearson correlation coefficient between the observed ABR latencies and predicted ABR latencies from the model prediction equations times 100. This is interpreted as that percentage of the variation in the predicted latencies that can be explained by means of the prediction equation. That is, the percentage of variation in the particular ABR latency or interpeak interval that can be explained given the hearing loss information. The level (pvalues) of statistical significance of the correlation coefficients and r2 values were used to compare relative effectiveness of the four models and hearing loss threshold as a prediction parameter. A two-tailed t-test of significance was used to develop the p-values for the correlation coeficents. RESULTS
Absolute Latencies Determination of the hearing test frequency or frequencies that best predict changes in the absolute ABR latencies was developed by the use of the p-values for the correlation coefficient and r2 information. A sample of these data are shown in Tables 1 to 3, giving the p values and the r2 values for the HFHL and FLATHL groups at all of the hearing loss predictor frequencies for the third-order polynomial model at the 95 dB attenuator setting. Waves V, I11 and I are given in Tables 1, 2, and 3, respectively. A contrast in predictability for a click level of 95 dB attenuation setting (Table 1) is noted for the HFHL and FLATHL groups. For the HFHL group, the third-order polynomial model was superior for waves I11 and V with both the secondand third-order polynomial models superior for wave I. For waves I11 and V, the best predictor frequency was at 3 to 4 and 4 kHz, respectively, for the HFHL group (Tables I and 2). Note that although the FLATHL group shows statistically significant correlation coeficients, the r2 values for the HFHL were larger than those for FLATHL (Tables 1-3). Wave I (Table 3) for the HFHL shows a broader dependence on threshold frequency for prediction, showing the best results for the 2 to 8 kHz combination. Ear and Hearing, Vol. 13, No. 3,1992
Table 1.p-Values for correlation coefficients and r 2 values 95 dB wave V third-order polynomial model.
Hearing Loss Predictor Frequencies in kHz
1 2 3 4 6 8
HFHL
p-values
FLATHL
[r2%1
p-values
[rzo/0]
NS [22.8] NS [10.1] 0.02< p 5 0.05 [47.3]0.005< p I0.01 [23.2] 0.01< p 5 0.02 [50.8] p < 0.005 [82.6]0.005< p 5 0.01 [26.6] NS [24.2] NS [26.1]0.005< p 5 0.01 [21.9]
2-8
4-8
NS NS NS 0.02< p 5 0.05 0.02< p 5 0.05 NS
[11.8] 0.01< p 5 0.02 [17.2] [31.7]0.005< p 5 0.01 [24.1] [32.8] p < 0.005 [26.4] < 0.005 [27.7] [44.9] [40.0] p < 0.005 [28.9] [26.4] p < 0.005 [27.0]
NS,not statistically significant, p > 0.05.
1 2 3 4 6 8
HFHL
p-Values
FLATHL
[r2%]
p-Values
0.01 [71.2] 0.01 < p [39.4]0.005 < p p < 0.005 [85.7] [86.9]0.005 < p p < 0.005 NS [41.4] 0.005< p I 0.01 [74.7]0.005 < p 0.005< p
5
NS
5
[r20/o]
0.02 [14.9] [14.9]
I0.01 I0.01
[26.5]
I0.01
[21.7]
NS [3.0] 0.02< p I0.05 0.02< p I0.05 [45.7]0.005< p 5 0.01 0.01 < p d 0.02 [56.8]0.005< p I0.01 0.005 < p I0.01 [75.0]0.005< p I0.01 0.005< p 5 0.01 [71.7]0.005< p 5 0.01 0.01 < P I 0.02 155.71 0.005 < P 5 0.01
[12.6] [18.7] [20.8] [23.0] [23.6] [24.41
Average of
1-2 1-4 1-8 2-4 2-8 4-8
1
2 3 4 6 8
HFHL
p-Values
0.01 < p 0.02< p NS NS NS NS
I0.02 I0.05
FLATHL
IrzYol
p-Values
rr2%1
[66.3] 0.02< p I0.05 [19.7] [56.6] 0.02< p 5 0.05 [16.8] [38.3] [5.9]0.005< p I0.05 [31.l] [30.2] [39.8]0.005c p I0.01 [25.3]
NS, not statistically significant, p > 0.05.
The FLATHL group correlation coefficient p-values and r2 results showed no clearly superior model for fit to the experimental data and no indication of a threshold frequency or combination of frequencies that adequately predicts latency behavior (Tables 1-3). These findings are evident for waves I, 111, and V for all three levels of click stimuli investigated. Correlation coefficient pvalues and r2 values for wave V for the HFHL group, as intensity of the click stimulus was changed from 95 to 75 dB, indicated that the predictor superiority of 4 kHz at 95 dB was no longer seen. It was replaced by a more broadly based region (2-8 kHz) of predictor frequency. This change also holds, in general, for waves I11 and I as click level was decreased. Based on the correlation Coefficient p-values and r2 data, a model was developed for wave latency behavior Ear and Hearing, Vol. 13,No. 3,1992
1-2 1-4 1-8 2-4 2-8 4-8
NS [48.1] 0.02< p I0.05 [17.1] NS [28.8]0.005< p I0.01 [24.0] NS [29.6]0.005< p 5 0.01 [25.5] NS [38.5] 0.01< p I0.02 [22.4] 0.005 < p 5 0.01 [96.4]0.005< p I0.01 [25.2] NS [31.l]0.005< p 5 0.01 [26.8]
NS, not statistically significant, p > 0.05.
Table 2. p-Values for correlation coefficients and r 2 values 95 dB wave 111 third-order polynomial model.
Hearing Loss Predictor Frequencies in kHz
Hearing Loss Predictor Frequencies in kHz
Average of
Average of
1-2 1-4 1-8 2-4
Table 3. p-Values for correlation coefficients and r 2 values 95 dB wave I third-order polynomial model.
versus degree of hearing loss. For HFHL, the basic assumptions were: ( I ) the best predictor frequency was 4 kHz for waves I11 and V,with 2 to 8 kHz average for wave I; (2) the experimental data were best fit by the third-order polynomial model. For FLATHL, these were: (1) given no clearly superior predictor frequency for the FLATHL group, it was decided to remain consistent with those chosen for the HFHL group; (2) no clear regression model was best, so the simplest, linear, was chosen for use. Application of the described models to the HFHL and FLATHL group’s audiometric data at the specified individual or combination frequencies was used to predict the mean ABR latencies and to predict +2 SD above the predicted mean latencies for waves I, 111, and V and interpeak intervals 1-111, 111-V, and I-V. This was done for click stimulus intensities of 95, 85, and 75 dB attenuator settings. Figure 1 provides a summary of the findings produced by application of the models described. Shown in Figure 1 is the wave latency delay in milliseconds above the normal-hearing upper limit cutoff (+2 SD above the mean for normal-hearing subjects) plotted versus hearing lossdegree. Only latencies greater than the cutoff were plotted, otherwise data are zero. The figure illustrates the following: (1) The rate of growth of the wave latencies outside the normal range is consistently greater for HFHL group than for the FLATHL group. (2) As click level is decreased (95 dB compared to 85 and 75 dB together), the effect on wave V latency increases for a given degree of loss for either hearing loss configuration. (3) For a given click level (95 dB), wave V appears less effected by a degree of loss than waves I11 and I for either hearing loss group. Interpeak Intervals For the HFHL group at 95 dB, the 2 to 8 kHz frequency combination and the third-order polynomial model gave the lowest p-values for the correlation coefCochlear Hearing Loss and ABR
175
w
1.0-
nl c
g
1.1-
2
'
1: .- 1 ;o.5:3 2
.
E
& ,
9
*--
0.0
B "w
P
'
ln X
ln - 0 . 5 .
0.0
'
-0.5.
.
-1.0-
-1.0-
U
U
0.5. E
B 6
.
* ln
--_----
0.0.
-0.5.
-1.0-
__--__---
L
r"
"
0.1.
0 w
4 ln
X
-0.5.
-1.0-
Figure 1. Wave V at click levels of 95, 85, and 75 dB in the top three panels showing the number of milliseconds above the normal-hearing subject upper limit of wave V latency as a function of degree of hearing loss. In the bottom left panel, the same information is provided for wave 111 and for wave I in the bottom right panel. Both bottom panels are for click levels of 95 dB. In all panels, the HFHL group is shown by a solid line, with the FLATHL group indicated by the dashed line.
ficient in the I-V interval ( p < 0.005). The 4 kHz frequency and the third-order model were found superior for the 111-V interval at 95 dB ( p < 0.05). At 75 dB for the HFHL population, no predictor frequency or model was found to be superior. As with the individual latencies, no superior predictor frequency was found for the FLATHL. The third-order model did show slightly lower p-values for the correlation coefficient for the I-V and 111-V intervals at 95 dB, with no difference in the models at 75 dB. Figure 2 (a-d) shows the individual interpeak intervals and the predicted mean intervals from the thirdorder model regression (HFHL) and linear model regression (FLATHL). The 2 to 8 kHz predictor frequencies are used for the I-V intervals and 4 kHz predictor frequency is used for the 111-V intervals. HFHL group results for a 95 dB click level are shown in Figure 2 (a and b), with the FLATHL group in Figure 2 (c and d). For the HFHL I-V interval, a definite trend for a reduction in the interval was noted as hearing loss increased (also true for a linear model, not shown). DISCUSSION
Comparison to Other Studies of Hearing-Impaired Subjects The results demonstrate that a loss of hearing of cochlear origin can cause prolongations in latencies of 176
Shepard et al
waves I, 111, and V beyond that for normal hearing. It is evident that the specific effect on latency is related to the configuration of the loss. This concept was presented by Coats and Martin (1977) and more recently in an article by Gorga et a1 (1985a, b). Gorga et a1 (1985b) found a significantly lower slope to the wave V latency intensity function for FLATHL compared to that for the HFHL. It would be anticipated that configuration of the loss (HFHL versus FLATHL) would correlate with the locus of lesion site within the cochlea, especially in those ears with a more severe loss. Therefore, the configuration could perhaps be interpreted to indicate those regions of the cochlea most likely dominating or participating in the production of the ABR. A HFHL would suggest a progressive loss of basal influence with increasing degree of loss and, therefore, an increasing dependence on the more apical areas. With decreasing click intensity, fewer basal fibers would respond, with more of the response coming from apical fibers. This would cause a rapid growth in latency with decreasing click intensity. In contrast, a FLATHL would provide for participation by all cochlear regions, giving a rate of latency growth parallel to that of the normal cochlea and a rate of growth less than that for HFHL. Our results are consistent with both this mechanism and the previous findings. Our results indicate that a cochlear hearing loss must reach 60 to 65 dB HL for our 85 or 75 dB click to cause Ear and Hearing, Vol. 13, No. 3,1992
a
C HFHL 95 I - V I N T . MODEL ( 3 r d o r d e r poly.]
v, E
5.0
=
-ft I
.A
FLATHL 95 I - V I N T . MODEL (linear]
I
c .r(
I
r(
m
> L w
I
4.0
cr
w C
>
* 3.0
3.0-1, 0
20
40
aa
60
Hearing Loss a t 2k-EkHz
[in
100
0
dB HLI
.
, 20
,
. 40
,
,
. . . ,I eo 100 l i n dB HLI
60
Hearing Loss a t 2k-BkHz
b
d HFHL 95 111-V I N T . MODEL ( 3 r d o r d e r poly.1
-c In
L
3.0
In
E
E
FLATHL 95 111-V I N T . MODEL (linear] 3.0
-
c
-
-n
.r(
m >
m
5 L W
$
-
L
W
g
2.0
2.0
w
W
>
>
H
U I
I
I+
w
r(
U
1.0
I
1.0
0
20
40
60
eo
I
loo
Hearing Loss a t 4kHz [ i n dB HLI
r)
Figure 2. Interpeak interval actual data points versus degree of loss for a frequency combination of 2 to 8 kHz and an individual frequency of 4 kHz, and either the third-order model predictionof interval for HFHL or linear model predictionof interval for FLATHL are shown. Panels a and b give the data for HFHL at 95 dB I-V and Ill-V intervals, respectively. Panels c and d give the same data for the FLATHL group.
any appreciable prolongation in the latency of the three waves studied. This is generally consistent with other investigations, which show the same degree of loss required to produce an effect on wave V latency (Coats & Martin, 1977; Gorga et al, 1985b; Selters & Brackman, 1977; Sohmer, Kinarti, & Gafni, 1981). Agreement is found across other studies with the result that the 4 kHz or 2 to 8 kHz regions provided the maximum predictability of wave V latency variation for HFHL when threshold is used as the available characterization of the status of the cochlea. For the FLATHL configuration, our results demonstrated no superior hearing test frequency for prediction of wave V latency behavior. It is suggested that this may reflect a broad locus of damage in the cochlea leaving intact the relative contributions to the responses based on synchronous activity of the VIIIth nerve. This would be in contrast to the HFHL configuration, where the typical dominating influence of the basal cochlear region is lessened as either click intensity is decreased or the magnitude of the loss is increased, thus causing a disruption in the Ear and Hearing, Vol. 13, No. 3,1992
synchronousfiring of the high-frequency region relative to the undisturbed activity of the apical regions.
Comparison to Normal System Studies The present results are consistent with predictions from studies of normal ears where wideband noise or derived narrowband responses are used to restrict the regions of the cochlea available to participate in the generation of auditory evoked potentials. As wideband noise is introduced with a click stimulus, the slope of the latency intensity function for the VIIIth nerve compound action potential (intracanicular component corresponding to wave I of the ABR complex) is found to be unchanged from that without the noise, even though the absolute latencies are prolonged (Montandon et al, 1985a; Montandon, Shepard, Marr, Peake, & Gang, 1985b). As contribution of the basalward cochlear regions are selectively removed by use of filtered clicks or derived responses, wave latencies progressively increase (Coats et al, 1979; Don & Eggermont, 1978; Eggermont & Don, 1980, 1982). For filtered clicks, the rate of Cochlear Hearing Loss and ABR
177
change in the latency intensity function for waves I, 111, and V appears to increase as the cochlear region of dominant contribution decreases below 2 kHz. This effect is not shown as clearly for the derived response paradigm using high-pass noise, but is suggested by the data (Eggermont & Don, 1980). Using these results, it could be hypothesized that a selective loss of basal contribution (HFHL) to the ABR would produce an increase in latency beyond that seen for a normal system. Also, the slope of the latency intensity function would be greater than that for normal hearing or a flat loss, as was shown with the present results. From the filtered click studies, the slope of the latency intensity function for wave I is greater than that for waves 111 and V because the contribution from the basal end of the cochlea is reduced (as intensity is lowered), resulting in reduced neural synchrony (Coats et al, 1979). This would predict the present result that indicates a decrease in the I-V and 1-111 intervals as the degree of the HFHL is increased. It may be hypothesized that this effect is, in part, related to the decreasing dependence on synchronous firing of the respective generating sites for waves I, 111, and V. Therefore, in the case of the FLATHL group with some preservation of the relative synchronous contributions, we find that the 1-111 and I-V intervals are independent of degree of loss. Predicting Effects of Hearing Loss on Wave V Latency Wave V normative range for HFHL and FLATHL groups can be developed by defining 2 SD above the predicted mean latency for a given degree of loss as the upper boundary for normal. Latency beyond this range would be defined as retrocochlear. Using degree of loss at 4 kHz of 60, 80, and 100 dB HL, a plot of 2 SD above the predicted mean for wave V latency versus click intensity was developed for each degree of loss for both HFHL and FLATHL groups. These plots, along with the upper limit for the normal-hearing male and female norms, are shown in Figure 3a for the HFHL and Figure 3b for the FLATHL. Ideally, normative data of this type are designed to reduce the false-positive rate without a reduction in hit rate. Our clinical experience over a 2Y2 yr period using hearing-impaired norms shows this to be correct (Shepard, Turner, & Burtka, 1986). Obviously, only clinical environments with a referral base with a substantial population of severe hearing loss patients stand to benefit from the use of norms of this type. It is exactly clinical situations of this type that have the need for this process. We suggest that the rate of false positives is proportional to the percentage of patients with significant loss of sensitivity. This may provide an explanation for the wide range of false-positive results (0-38%) reported by Turner, Shepard, and Frazer (1984) after a review of 15 yr of literature. Other measures of cochlear function (e.g., compound action potential tuning curves and psychophysical tun178
Shepard et al
High F r e q u e n c y H e a r i n g L o s s N o r m s
a
WAVE
v
UTENCY
10.0
3.0
8.0 v)
E
7.0
6 0 I
.
1 .
25
45
"
I
T
'
I I
I
7
~
; 1
- -
i
I
1
1
65
I
.
I
1 :
I
II
85
105
d 8 Attenuator Setting
Flat Hearing Loss N o r m s
WAVE V L A W
k Y
10.0
9.0
8.0 In E
7.0 6.0
5.0 25
45
65
85
105
dE Attenuator Setting Figure 3. Normal-hearing subject upper limit of the normal range for females (A-A) and males @€I) for wave V latency versus click intensity in dB attenuator setting. In panel a, the upper limit of the acceptable range for a cochlear loss of sensitivity for a 4 kHz loss of 60,80, or 100 dB HL is given by the dashed lines for HFHL. Panel b shows the same information for a cochlear hearing loss of a flat configuration.
ing curves) have been shown to accurately reflect the status of the cochlea (Shepard & Abbas, 1983). These measures have a higher probability of increased correlation and, therefore, increased predictability of wave latency performance. These measures provide for both loss of sensitivity information as well as frequency resolution capabilities of the cochlea. This may be of Ear and Hearing, Vol. 13, NO. 3,1992
use especially in cases where the cochlear lesion does not manifest itself through significant change in behavioral threshold. The utility of the compound action potential tuning curve or the psychophysical tuning curve for a clinically usable model is limited by the time required to obtain this information. Therefore, the pragmatic compromise is to trade reduction in accuracy for availability of the hearing threshold information. The decision to use hearing thresholds may well be a major reason for the inability to completely account for the increased false-positive test results seen with a severely hearing-impaired population. As stated under “Methods,” the FLATHL group was comprised of only those patients diagnosed as having endolymphatic cochlear hydrops or a specific form of endolymphatic hydrops, Meniirss disease. The possibility that the specific results obtained for the FLATHL group may be pathognomonic to this disorder classification cannot be ruled out, but is unlikely. This group of patients constitutes the largest percentage with cochlear lesions that are suspect for retrocochlear involvement. Their choice for the present study was, therefore, appropriate. CONCLUSIONS
1. We have demonstrated a differential effect of cochlear hearing loss configuration on ABR waves I, 111, and V latencies. 2. For FLATHL configuration, interpeak intervals show no effect with increasing loss of sensitivity. HFHL configuration shows a reduction in the intervals of I-V and 1-111 with increasing degree of loss. The 111-V interval showed no effect. 3. It is hypothesized that an explanation for the differential effects on latency across configuration and within the HFHL configuration for wave I versus waves I11 and V reflect the locus of the lesion within the cochlea and relative disruption of synchronous neural activity of wave sources, respectively. 4. Although use of hearing threshold information is not optimal for development of a model of this type, it is the only practical measure to use as a predictor in a clinical setting. 5 . Normative data were developed for ABR interpretation that would be predicted (and shown by clinical trial under separate reporting) to reduce the false alarm rate without a significant effect on the hit rate in a clinical population when losses of auditory sensitivity are prevalent. REFERENCES Coats CA and Martin JL. Human auditory nerve action potentials and brain stem evoked responses effects of audiogram shape and lesion location. Arch Otolaryngol 1977;103:605-622. Coats CA, Martin JL, and Kidder HR. Normal short-latency electrophysiological filtered click responses recorded from vertex and external auditory meatus. J Acoust SOCAm 1979;65(3):747-758. Don M and Eggermont JJ. Analysis of the click-evoked brainstem potentials in man using high-pass noise masking. J Acoust SOCAm
Ear and Hearing, Vol. 13, No. 3,1992
1978;63(4): 10984-1092. Eggermont JJ and Don M. Analysis of click-evoked brainstem potentials in human using high-pass noise masking. 11. Effect of click intensity. J Acoust SOCAm 1980;68(6):1671-1675. Eggermont JJ and Don M. Analysis of click-evoked brainstem auditory electric potentials using high-pass noise masking and its clinical application. Ann NY Acad Sci 1982;82:471-486. Gorga MP, Reiland JK and Beauchaine KA. Auditory brainstem response in a case of high-frequency conductive hearing loss. J Speech Hear Disord 1985a;50:346-350. Gorga MP, Worthington DW, Reiland JK, Beauchaine KA, and Goldgar DE. Some comparisons between auditory brainstem response thresholds, latencies and the pure-tone audiogram. Ear Hear 1985bl6:105- 1 12. Montandon PB, Megill ND, Kahn AR, Peake WT, and Kiang NYS. Recording auditory-nerve potentials as an ofice procedure. Ann Otol Rhinol Laryngol 1975a;84:2-11. Montandon PB, Shepard NT, Marr EM, Peake WT, and Kiang NYS. Auditory-nerve potentials from ear canals of patients with otologic problems. Ann Otol Rhinol Laryngol 1975b;84:164-173. Selters WA and Brackmann DE. Acoustic tumor detection with brainstem electric response audiometry. Arch Otolaryngol 1977;103:181-187. Shepard NT and Abbas PJ. Compound action-potential tuning curves in normal and acoustically traumatized cats. Ann Otol Rhinol Laryngol 1983;92:496-503. Shepard NT, Turner RG, and Burtka MJ. BAER experience using cochlear hearing impaired normative data. J Acoust SOCAm 1986;81:S2. Sohmer K, Kinarti R and Gafni M. The latency of auditory nervebrainstem responses in sensorineural hearing loss. Arch Otolaryngol 1981;230:189-199. Turner RG, Shepard NT, and Frazer GJ. Clinical performance of audiological and related diagnostic tests. Ear Hear 1984;5:187-194. ~~
Acknowledgment: We would like to express our gratitude to Dr. Robert Turner for his critical review of the manuscript and his helpful suggestions. Address reprint requests to Neil T. Shepard, Vestibular Testing Center, Dept. of Otolaryngology, Box 0816, Rm. C166A, Med-Inn Bldg., University of Michigan, 1500 E. Medical Center Dr., Ann Arbor, MI 48109. Received June 23,1989; accepted February 10,1992.
APPENDIX
Linear Model
Y=a+bx where Y = predicted wave latency or interpeak interval; x = measured hearing loss thresholds in dB HL; and a and b = calculated from the regression analysis for best fit to empirical data by least squares methods.
b = [C(xi - X U i - y)I/[Z(xi- x)’I a=y-bx where x = mean of the hearing loss thresholds in db HL; y = mean of the observed wave latencies for a given wave or interpeak interval across thresholds; (xi, yi) = datum point of hearing loss and wave latency for a given wave I, 111, or V, or hearing loss and interpeak interval; and n = number of data points for each wave latency or interpeak interval. Second-Order Polynomial Model Y = bo + bix
+ b2X2
Cochlear Hearing Loss and ABR
179
where Y = predicted wave latency or interpeak interval; x = measured hearing loss threshold; and bo, bl, bz = calculated coefficients from the regression analysis. Transform Model Y‘ = a’
+ b‘x
where Y’ = predicted latency or interpeak interval after transforming of observed latency or interpeak interval values; x = measured hearing loss thresholds in dB HL; and a’, b’ = calculated by regression analysis after first transforming the observed latency and interpeak interval values as follows:
where x = mean of the hearing loss thresholds in dB HL for a given wave or interpeak interval; y‘ = mean of the transformed observed latency or interpeak interval values; and (x,,y,’) = datum point of hearing loss threshold and transformed latency or interpeak interval value for a given wave or interpeak interval.
Third-Order Polynomial Model Y = ho -C hlx + bzx2 + b s 3
Y’ = logloY b’
=
[C(xi - 2 ) ( ~-’ ,jj’)l/[Cx,- n)’] a’ = j j ‘ - b’2
where y = predicted wave latency or interpeak interval; x = measured hearing loss threshold; and bo, bl, b2, and b3 = calculated coefficients from the regression analysis.
~~
180
Shepard et al
Ear and Hearing, VOI. 13, NO. 3,1992
