Auditory Brain Stem Response for Objective
Measures of Hearing
Yvonne S. Sininger, PhD
Children's Auditory Research and Evaluation Center, House Ear
Institute, Los Angeles, California
ABSTRACT
The auditory brain stem response (ABR) is felt to be an
objective technique for predicting hearing thresholds be-
cause a voluntary response is not required from the sub-
ject. However, determination of ABR threshold can be a
subjective process. This article discusses a technique,
termed F,, which adds objectivity to ABR threshold de-
tection by creating a ratio of signal plus averaged back-
ground noise over an estimate of the averaged back-
ground noise for any given averaged ABR. F, values have
an F distribution. Consequently, the confidence of true
detection for a given ABR can be determined by compar-
ing its calculated F, value to statistical tables. Using a
technique such as F, not only adds objectivity to ABR
threshold detection, but also optimizes test time by allow-
ing the averaging process to stop as soon as the back-
ground noise has been reduced and the true neural poten-
tial can be judged to be present. The estimate of the
background noise can be used as a weighting factor to
reduce the influence of noisy segments during the aver-
aging process as well. Using this technique, we have
found ABR threshold to be within 5 or 6 dB of psycho-
physical threshold for like (click) stimuli and, in our pedi-
atric clinic, ABR click thresholds are within 10 dB of pure-
tone average for children with losses ranging from mild to
profound. (Ear Hear 14 1:23-30)
THE ABR IS AN AVERAGED, scalp recording of
neural activity initiated by sound and generated in the
auditory nerve and brain stem. This response is reliably
recorded from adults and infants (Hecox & Galambos,
1974)and is not contaminated by state of consciousness
(Starr, 1976) or attention (Picton & Hillyard, 1974). It
must be noted that the ABR does not measure hearing.
ABR and hearing are two distinct processes; ABR re-
quires detection of sound-initiated neural activity from
the scalp, whereas hearing is the conscious perception
of sound. However, ABR detection threshold is close
Ear and Hearing, Vol. 14, No. 1,1993
to psychophysical threshold (Elberling & Don, 1987a;
Sininger & Don, 1989), and screening-level ABRs have
been shown to be reliable, early indicators of hearing
impairment in newborns (Hyde, Malizia, Riko, & Al-
berti, 1991).
ABR has dramatically improved the accuracy of
hearing assessment of young children and lowered the
age at which a child can be evaluated. However, the
technique, as applied in most audiology clinics today,
is not completely objective. Because it requires no
active response from the patient, it is less subjective
than behavioral measures of hearing; however, ABR
threshold, in most cases, requires that the examiner
make a subjective decision about the presence or ab-
sence of averaged neural activity in background noise.
The difficulty in subjective interpretation of presence
or absence of ABR is due to the inherently poor signal
to noise ratio of the recording. Threshold-level ABR
wave V amplitude can be as low as 50 nV (Sininger &
Masuda, 1990), whereas fluctuations in background
noise during averaging can easily be 20 pV.
Time-locked averaging reduces noise contaminants
of the ABR. The neural response to the stimulus can
be considered deterministic; it is stable in amplitude
and time for a given stimulus. The exception to this
principle is found in adaptation of the neural response
to the stimulus. However, adaptation is complete after
the first few stimuli are presented (Don, Allen, & Starr,
1977), and adaptation should not significantly impact
upon the basic assumption that the response is deter-
ministic.
In contrast to the neural signal, background noise is
random, independent of auditory stimulus, and will be
reduced by partial cancellation after a sufficient number
of sweeps have been averaged. Time-locked averaging
enhances the signal (neural response) in relation to the
noise by reducing the background noise.
Predicting psychophysical threshold with the ABR
involves determining the threshold of ABR, that is, the
lowest stimulus level that will produce a detectable
response. The most common method for determining
ABR threshold is visual inspection of an averaged wave-
form. In most instances, two waveforms are generated
by repeated averaging of a fixed number of sweeps, for
0196/0202/93/1401-0023$03.00/0 EARAND HEARING
Copyright 0 1993 by Williams 8, Wilkins Printed in the U.S.A.
ABR for Objective Measures of Hearing 23
 example, 2048 (Jerger, Oliver, & Stach, 1985). These
responses are compared visually or by calculation of a
cross-correlation product. A clear replication indicates
a response. This judgment is based on the assumption
that peaks that remain after averaging represent neural
response because the averaging process had adequately
reduced background noise.
However, because the amplitude of background noise
can fluctuate widely over time, it is not possible to
determine, a priori, how much averaging will ade-
quately reduce noise in a given average. A visual or
correlational comparison of two waveforms can evalu-
ate responses with very different signal to noise ratios,
leading to errors in interpretation. Figure 1 shows ABRs
from a patient in response to a 100 dB nHL click (top)
and to an 80 dB nHL click (bottom), each with 2000
sweeps averaged. In the top example, waveforms repli-
cate adequately and there is little doubt that an ABR is
present. The situation with the 80 dB recordings is not
as clear. The waveform morphology in one trace sug-
gests a response, but the replication is poor. If a response
is present at 80 dB, it has been obscured by noise that
was not adequately reduced by the averaging.
The examiner’s judgment of the presence or absence
of a response depends on the quality (signal to noise
ratio) of the response, which is determined by the
combination of signal strength and level of averaged
background noise. Once recordings have been obtained,
there is no objective standard for determining how well
two responses must replicate to be judged a true re-
Repeat ABRs from Noisy Patient
2000 SweepslAverage
Figure 1. ABRs elicited by 100 (top) and 80 dB nHL clicks (bottom).
Each trace represents the average of 2000 stimuli. Waveform mor-
phology and visual replicationare good in the 100 dB example, and
a response is clearly present. The 80 dB responses do not replicate
well, and it is not clear if these averages contain neural response or
noise.
24 Sininger
sponse. The subjective nature of the ABR lies in the
examiner’s responsibility to adequately acquire re-
sponses and later to judge the distinction between an
ABR and a recording of noise.
Elberling and Don (1984) and Don, Elberling, and
Waring (1984) have described a technique for deter-
mining the quality of an ABR based on a background
noise level estimate and the overall level of the response.
This technique determines a value known as Fspfor
each block of 256 average sweeps. This value is closely
related to the signal to noise ratio of the recording and
has statistical properties that allow the examiner to
determine the level of confidence that an average con-
tains a true ABR.
The derivation of and rationale for the Fsphas been
described in detail elsewhere (Elberling & Don, 1984).
Fsp is based on the concept that any averaged evoked
potential recording includes some amount of neural
activity and some background noise. For any given
digitized point in an averaged ABR waveform, the
neural or evoked potential contribution to amplitude is
constant from sweep to sweep, whereas the noise con-
tribution varies. The greater the background noise in a
response, the greater the sweep to sweep variance of the
amplitude at any point in the array. In the calculation
of Fspl variance of a single digitized point within the
array is used to estimate the noise level in the recording
that can be accurately characterized after 250 sweeps
(Elberling & Don, 1984).
Fspis calculated according to the following formula:
Fsp= VAR(S)/VAR(SP)
where S is the averaged signal that includes the evoked
potential plus averaged background noise. VAR(S) is
the variance of digitized amplitude values across an
appropriate time window of the averaged response as
in Figure 2 (right). SP is the digitized voltage of a single
point in the array, and VAR(SP) is the variance of that
value calculated over at least 250 sweeps (Fig. 2, left).
VAR(S) should be equal to the RMS2 of the averaged
evoked potential plus the averaged noise, and the
VAR(SP) is equal to the RMS2 of the estimated back-
ground noise.
When VAR(S) is divided by the estimate of the noise
in the response, or VAR(SP), the resulting number is
closely related to the squared signal to noise ratio in
the response. This value estimates the quality of the
ABR or strength of the neural potential. When no
response exists, the expected Fspvalue is close to 1 (zero
signal + noise divided by noise). Fspvalues are calcu-
lated on-line and updated every 250 sweeps during the
averaging process. As averaging reduces the background
noise, the Fspvalue grows. Figure 3 plots the growth of
Fspwith averaging for several signal levels. Fspgrows
more rapidly with high sensation-level stimuli than low
because the overall amplitude of the response is larger
relative to the background noise.
Because Fspis a ratio of variances, these values have
a known (F) distribution with, conservatively, 5 and
Ear and Hearing, Vol. 14, No. 1, 1993
 Variance of a Single P o i n t
from S w e e p to Sweep
250 df (Elberling & Don, 1984). Based on the F table,
the examiner can be 99% confident of a true response
(1 % false-positive rate), and the averaging process can
be terminated when an ABR average reaches an F,,
value of 3.1. Lower criterion levels can be used if
2 confidence requirements for obtaining a true response
can be relaxed. For example, in the clinic, a 2.25 F,,
criterion gives a 5% false-positive detection rate, which
3 is acceptable, but has the advantage of missing fewer
Average O f N sweeps
responses by hiving a better true positive detection
than the 3.1 criterion (Elberling & Don, 1987a). When
combined with other measures of reliability, such as
seeing the expected latency and amplitude changes with
stimulus level, the confidence level, using an Fspcrite-
Y o r , o n c c 0 1 succ.,siv.
in Window 0 1 *".Fop.
Points
rion of 2.25, can actually be greater than 95%. The
NLh
name Fspcomes from F for the distribution and SP for
the single point used to estimate the background noise.
Figure 2. F,, = VAR(S)/VAR(SP). At right is an illustration showing The duration and placement of the time window
the calculation of the numerator, or VAR(S), in the F,, equation. S
stands for the averaged signal. Variance is calculated for points within
used to calculate VAR(S) are critical. The duration of
a window of the averaged response after each block of 250 sweeps.
the window must be sufficient to allow adequate reso-
This variance will indicate the amplitude of the neural response plus lution of noise contributions. If a high-pass recording
averaged noise. At left, the single point variance, or VAR(SP), calcu- filter of 100 Hz is used, a 10 msec window is adequate.
lation is depicted. This is the denominator in the F, equation and However, a longer window must be used if this filter is
estimates the level of background noise in the average. Any ABR lower to allow for adequate amplitude resolution of
contributionto the amplitude at the single point should not vary from low-frequency components. Also, because Fspwill not
sweep to sweep. Noise will cause sweep to sweep fluctuations, and distinguish between neural and other time-locked con-
the variance of the amplitude at the single point will estimate the level tributors, placement of the window must allow visual-
of background noise. ization of all major components of the ABR, but must
avoid any other time-locked, non-neural components
such as stimulus artifact. If stimulus artifact is sus-
pected, the window used to calculate VAR(S) should
Fsp GROWTH WITH STIMULUS LEVEL start after any such components.
AND NUMBER OF SWEEPS Because the contribution from neural components is
equal for each sweep, the absolute value of the ABR at
any given point in time will not contribute to the sweep
"F ? to sweep single-point variance calculation. Conse-

t quently, whether the noise point is selected on top of

i

25 i wave V or at any other point is irrelevant as long as it

+ is within the same window used to measure VAR(S)

20 i 0

0
+
(Elberling & Don, 1984).
The Fsptechnique has many advantages over simple
visual detection of peaks. Because confidence of true
+ response detection is gained from reaching a significant

$110
0
4
0
F,, rather than from seeing a duplication of waveforms,
F,, can be used as a response criterion as well as a
stopping point for the averaging process. Thus, testing
time is minimized. Averaging will continue only until
an adequate signal to noise ratio is obtained. A robust
response from a quiet subject can be obtained with a
small number of sweeps. More time will be spent av-
eraging a greater number of sweeps when the signal to
noise ratio is poor, such as in a noisy subject or when
0 1 2 3 4 5 6 7 8 9 1 0 the stimulus level is close to threshold. Thus, test time
Sweeps in Thousands can be reduced in some situations and can be used
Figure 3. Average F,, growth with the number of sweeps averaged
more efficiently in others. In addition, further time is
using click stimuli ranging from 5 dB below to 20 dB above psycho- saved by eliminating the need to replicate responses.
physical threshold. The horizontal line indicates the 99% confidence The level of background noise in the average, esti-
point for true detection of a response. Data represent the average of mated from the single point variance, can also be used
six subjects (from Don et al, 1984).0 ,20 dB SL; *,
15 dB SL; W, 10 as a stopping point for the averaging process. For ex-
dB SL; A s-5 dB SL. ample, when the averaging process reduces the back-

Ear and Hearing, vol. 14, No. 1,1993 ABR for Objective Measures of Hearing 25
 ground noise sufficiently to resolve a low-amplitude
ABR, but F,, detects no signal, it is logical to stop the
averaging and assume no response. Thus, the quiet
subject with no response will require evaluation only
until the background noise is reduced. Again, overall
test time is reduced.
Another use of the background noise estimate is in
weighting of blocks of sweeps in the average, based on
the relative noise in each block. Such weighting uses
Bayesian estimation principles (Elberling & Wahlgreen,
1985). Each block of 256 sweeps receives a weighting
factor before being entered into the average. Blocks
with a low noise variance will receive a high weight in
the overall average, and noisy segments will be weighted
to reduce their contribution. This technique minimizes
the detrimental effects of episodic, large bursts of noise
which are characteristic of recordings from infants. In
a patient who has episodic bursts of noise, the total
number of sweeps presented to obtain an adequate
response (and subsequently, the total time for averag-
ing) can be reduced using this technique. If noise is
constant, the weighting for each block of sweeps will
also be constant and the process will be the same as
traditional averaging.
The main advantage of the Bayesian weighting
scheme is that, once the averaging process has revealed
an ABR, a large burst of noise will not destroy the
signal to noise ratio. Instead, the influence of noisy
sweeps will be reduced. The Fspwill increase smoothly
as averaging proceeds when Bayesian weighting is used
and an ABR is present (Elberling & Wahlgreen, 1985).
However, unlike traditional techniques of artifact rejec-
tion, the Bayesian weighting technique discards no
information.
STATISTICAL CRITERIA
The value of a technique with a known statistical
framework, such as Fsp, is that for a desired probability
of detecting a true response, the appropriate F,, value
and the operating characteristics of that criterion can
be determined (Elberling & Don, 1987a). For example,
to evaluate F,, as a response criterion for an infant
screening program, the distribution of F,, values for
256-sweep blocks of no-stimulus ABR was determined
for newborn infants. This distribution should indicate
the expected F,, .from infants with hearing impairment
greater than the level of the screening stimulus. Another
distribution of F,, values was found from normal new-
born ABRs after 2000 sweeps in response to 30 dB
nHL clicks. Recording band pass was 30 to 3000 Hz
and the F,, analysis window was 35 msec for both
conditions. Figure 4 shows these distributions.
There is almost no overlap in these two distributions.
The resulting receiver operating characteristic curve
(Fig. 4 inset) shows excellent operating characteristics.
An Fsp criteria of approximately 2.4 appears to be
optimal for distinguishing a 30 dB response from no-
response ABRs. Using this criterion, the false-positive
26 Sininger
DISTRIBUTIONS OF Fsp VALUES
FROM NEONATE ABRS
140
120
100
80
M
60
40
20
0 Y
-2 0 2 4 6 8 10 12 14 16 18 20
Fsp Value
Figure 4. Distributions of F, values from ABRs of newborns using a
recording band pass of 30 to 3000 Hz and a F, analysis window of
35 msec. On the left is the distribution with no stimulus used to
simulate the distribution of values in a "no response" condition. The
number of F, values is shown on the left axis. The distribution of F,,
values obtained from newborn ABRs elicited by 30 dB nHL clicks
after 2000 sweeps is shown at the right. The number of F, values
for 30 dB responses is shown on the right axis. Inset: Receiver
operating characteristic (ROC) based on the two distributions. The
ability to distinguish ABRs from neonates elicited by 30 dB clicks
from a no-stimulus ABR is evaluated based on varying F,, criteria
from 1.O to 4.0 in steps of 0.5.
rate from these data is 1%, the true positive detection
rate is 98.5%, and A' calculated at this point is 0.994.
F., USE IN RESEARCH AND THE CLINIC
ABR threshold using Fsp criterion is an accurate
predictor of hearing sensitivity in the laboratory as well
as the clinic. Comparisons have been made of ABR
threshold determined with F,, and psychophysical
threshold in normal-hearing subjects. These studies re-
vealed that ABR click threshold determined with a F,,
criterion is within 5 to 6 dB of psychophysical threshold
for like stimuli (Elberling & Don, 1987b; Sininger &
Don, 1989). Recording and stimulus parameters that
affect ABR threshold have been evaluated with F,, as a
threshold determiner. Sininger and Don (1989) have
shown that increasing the click stimuli up to 48/sec
does not affect threshold determination with the ABR,
but that a vertical recording channel, Cz to the seventh
cervical vertebra, will give a lower threshold than a
horizontal (ear to ear) or anterior-posterior (nasion to
inion) recording. Click polarity does not influence ABR
threshold levels (Sininger & Masuda, 1990).
ABR with Fsphas also been used to objectively deter-
mine ABR threshold for children in the clinical audiol-
ogy program at the House Ear Institute for the last 5
years. ABR is used routinely to predict hearing level of
infants less than 1 year old and of any young child who
Ear and Hearing, Vol. 14, No. 1, 1993
is inconsistent or difficult to evaluate by standard be-
havioral measures. It is important to note that children
older than 6 months are sedated with chloral hydrate
under medical supervision for ABR testing. Using se-
dation dramatically reduces noise contaminants in the
recordings, consequently reducing ABR threshold levels
somewhat.
ABR procedures in our pediatric clinic are specifi-
cally selected to emphasize threshold detection based
on wave V of the ABR. Responses are recorded from
electrodes on the vertex referenced to the nape of the
neck. This montage records wave V in response to
clicks presented to either ear (King & Sininger, 1992).
For older babies and young children, clicks are pre-
sented up to 48/sec, and a recording band pass of 100
to 3000 Hz and a 15 msec recording epoch are used.
For newborns, clicks are restricted to 20/sec, and a
band pass of 30 to 3000 Hz and a 40 msec recording
window are used. Click stimuli are used first and, when
time permits, high-pass masking is mixed with clicks to
create frequency-specific derived bands.
Figure 5 shows a click-evoked ABR evaluation on a
16 month old child with a family history of hearing
RIGHT
dB
Fsp/Swesps
nHL
LEFT
6o
50 -
-5632
2.25/8448
50
3.1 0/7424
40 - 1 . 1 8/10240
40- - .67/10240
s
5 70 15m.
10
ocoumrn
ABR
me3hdd
RE 6 LE
c o x
I+Sound Field Thresholds I
Figure 5. ABR threshold evaluation with click stimuli. Averaging con-
tinued until an F,, of 2.25 (95% confidence) was obtained or until
10,240 sweeps were presented. A small response is seen in response
to clicks of 45 dB nHL in both ears, with no response seen up to 40
dB clicks. Sound field thresholds indicate good agreement with ABR
thresholds. ABRs indicate that the hearing is reasonably symmetrical.
Ear and Hearing, Vol. 14, No. 1,1993
loss. Responses were averaged until a Fspof 2.25 (95%
confidence) was reached or until 10,240 total sweeps
were presented. An adaptive threshold procedure was
used. A response to a 60 dB nHL click in the right ear
reached criterion after 5632 sweeps. The next presen-
tation was 40 dB, where no response was found, fol-
lowed by 50 and then 45 dB. Whenever possible, thresh-
old is determined with 5 dB steps. The right ear thresh-
old was determined to be 45 dB. The threshold search
began at 50 dB in the left ear, followed by 40 and 45
dB, with threshold found again at 45 dB. These results
are compared to a sound field audiogram (Fig. 5, bot-
tom) obtained a few months after the ABR evaluation.
In general, the number of sweeps needed to obtain a
given signal to noise ratio (Fspcriterion) increases as the
sensation level of the click decreases. Figure 6 shows
the relationship between the number of sweeps needed
to reach criteria and click level relative to ABR thresh-
old. Error bars indicate the maximum number of
sweeps needed (based on a group of 20 subjects) and
the minimum. For these cases, if no response (Fspof
2.25) was found by 6144 sweeps (24 blocks), then the
averaging process was discontinued. The relationship
between stimulus level and amount of averaging needed
to reach a given signal to noise ratio is clear. To obtain
a quiet response with low-level stimuli, more averaging
is needed than with higher sensation level clicks. The
sensation level of the stimulus, the level of the back-
RELATIONSHIP OF STIMULUS LEVEL
AND AMOUNT OF AVERAGING NEEDED
5
._
6144
._ 5120
1Sma
is
c
o 4096
K
I
3072
B
0,
2048
*
v)
1024
0 5 10 15 20 25 30 35 40 45
dB re ABR Threshold
Figure 6. Twenty recent cases were evaluated to determine the
number of sweeps needed to reach an F,, criterion of 2.25 relative
to the stimulus level above ABR threshold. The maximum number of
sweeps in this series of cases was 6144. The error bars indicate the
range of the data. Subjects varied a great deal on background noise
and response amplitude. At threshold, the average number of sweeps
required was over 4300, whereas some subjects required the maxi-
mum number and others reached criterion with as few as 1024. At
40 dB above ABR threshold, the average number of sweeps needed
is less than 1000 and the range is much smaller.
ABR for Objective Measures of Hearing 27
ground noise, and the level of the neural signal are all
unknown during an ABR evaluation. Using F,, to
characterize the signal to noise ratio of the response
allows the appropriate amount of averaging to be used
in each condition.
Based on the information in Figure 6, we use the
number of sweeps needed to reach criterion to indicate
the approximate sensation level of the stimulus. This
can help to reduce the evaluation time by indicating
the next level to present in the threshold search. For
example, when criterion is reached in a small number
of sweeps, for example, less than 1000, the next pres-
entation level will be 20 to 30 dB lower. If a response
is seen only after 5000 or 6000 sweeps, the next pres-
entation level will be reduced by no more than 10 dB.
Figure 7 shows the audiogram and click-evoked ABR
thresholds for a 14 month old child with a severe loss
in the right ear and a profound loss in the left. Click
thresholds agree with pure-tone responses. This case,
typical of many in our clinic, demonstrates the ability
of ABR to distinguish between losses with differences
in pure-tone averages as small as 10 dB. This case also
shows that hearing predictions for severe and profound
losses are as accurate as for mild or moderate losses
with use of Fspcriteria and click-evoked ABRs.
ABR threshold predictions with the F,, criterion have
been compared to subsequent behavioral audiometrics
RIGHT LEFT
dB
nHL
Fsp/Sweeps
It
for a group of children 6 months to 4 years of age.
Figure 8 compares 19 instances of audiometrics (ear
specific or sound field) to best-ear click-evoked ABR
thresholds. Audiometric values used for this compari-
son were those obtained when they were felt to be
accurate and reliable. Pure-tone averages represent the
results of the most recent audiometric results for an
average of 500, 1000, and 2000 Hz, or two frequencies
when one threshold differs significantly from the others.
Several factors can contaminate this kind of compar-
ison. ABR results can influence the audiologist's inter-
pretation of the behavioral evaluation, which will make
the correlation appear better than it actually is. Also,
accurate behavioral results may not be obtained until
many months after the ABR test; in some cases, the
behavioral results judged to be most accurate were
obtained more than 1 year after ABR. A long time
between evaluations tends to make predictions look less
accurate.
Regardless of possible contaminations, the accuracy
of the prediction of behavioral pure-tone average from
click-evoked ABR is within 10 dB, as shown on the
histogram in Figure 8. The two measures appear to
correspond equally well for mild losses as for profound.
Thresholds found using the F,, criterion do not de-
pend upon visual detection of a typical waveform mor-
phology. As Figure 6 illustrates, ABR at threshold often
only displays a low-amplitude baseline shift, indicative
of wave V. It has been our finding that, in cases of
sloping hearing loss through the speech frequencies, the
ABR threshold corresponds to the level of the best
threshold above 500 Hz. When only low-frequency

IJ----- .40/10240

-
90
110 0 *'

1.62,' 10240 0 ye*

8ol------I
.
I,,
,,.
,,,
,,,
,,,
,
10
1.34/10240

15mr
5 10 15ma
100 r=.979 ..,'o

0cou.is

I l l

100 c -
250 500 i o O 0 2 a x ) ~ 8 w o
Frequency (Hz)
30 40 50 60 70 80
ABR CLICK THRESHOLD
90 100.110

Figure 8. Comparison of ABR click thresholds to pure-tone average

Figure 7. ABR threshold search evaluation indicating click thresholds
of 95 dB in the right and no response at 103 dB nHL in the left ear.
hearing thresholds in children. Pure-tone averages are based on
thresholds at 500, 1000, and 2000 Hz, or the best two frequencies
when one threshold was significantly different than the other two.
The dashed line is the regression for the function revealing a corre-
lation of 0.979. The distribution of the differences between the two
measures is shown at the lower right.

28 Sininger Ear and Hearing, Vol. 14, No. 1,1993

regions of the cochlea are available to respond, the
latency of wave V is prolonged, but the click-evoked
ABR is still present. Figure 9 gives an example of such
a case. This illustrates that click-evoked ABR can be
generated by frequency regions below 2000 Hz. This is
expected because the click has energy in a wide band
of frequencies. ABRs generated by low-frequency re-
gions may have lower amplitudes and longer latencies,
and are apparent only at increased sensation levels
compared to ABRs generated by higher frequency re-
gions, but these are still detected with adequate aver-
aging.
When time permits, we utilize the high-pass masked,
derived band technique to obtain frequency-specific
ABR thresholds as described by Don, Eggermont, and
Brackmann ( 1979). This technique mixes high-pass
filtered noise of varying cutoff frequencies with a click
stimulus. The masking eliminates ABR from those
sections of the cochlea responding to the noise. By a
series of subtractions, the ABR generated by octave-
wide cochlear regions (derived band ABRs) can be
distinguished and a threshold for each band can be
determined.
Figure 10 shows an example of such an evaluation
on an 18 month old child compared to her audiogram
obtained at age 4.A correction factor of 15 dB is applied
to 500 Hz derived responses to adjust for the slightly
higher sensation level of low-frequency ABR threshold.
Agreement between derived ABRs and standard pure-
tone thresholds is quite good. Unfortunately, we do not
have a large sample of derived responses from children
with varying levels of hearing impairment. It is not
known how well low-frequency sensitivity can be pre-
dicted by this method when hearing levels fall in the
severe to profound range.
Although we have traditionally used derived re-
sponses, we are also evaluating frequency-specific ABRs
Right Ear Left Ear
Hx of poor speech development
Audiogram at 3 yr 9 mo.
ABR at 18 mo.
Figure 9. Relationship of ABR click thresholds to pure-tone thresh-
olds from a child with a Severely sloping, high-frequency hearing loss.
Based on response detection alone, low-frequency contributors to
click-evoked ABRs will be seen when the high-frequency response is
eliminated by hearing impairment.
Ear and Hearing, Vol. 14, No. 1,1993
Right Ear Left Ear
,251.5 1 2 4 8k Hz .25 .5 1 2 4 8
Family hx of hearing loss.
ABR at 1 year, audiogram at
4 years of age.
Figure 10. Pure-tone audiogram and click and high-pass masked,
derived band ABR thresholds (Don et al, 1979) from a child with a
congenitalhearing impairment.
using tone bursts in notched noise, as described by
Stapells, Picton, Perez-Abalo, Read, and Smith (1989,
and feel that these should also give an excellent descrip-
tion of the pure-tone audiogram.
CONCLUSIONS
The key to accurate prediction of hearing levels using
ABR is obtaining recordings with adequate signal to
noise ratio. Techniques such as Fspfacilitate this process
by characterizing the signal to noise ratio, thus allowing
the examiner to determine the appropriate procedures
for obtaining quiet recordings, such as increased aver-
aging. Calculation of Fspis certainly not the only ade-
quate method for estimating the signal to noise ratio of
the averaged ABR. However, careful study with this
technique has shown that auditory psychophysical
thresholds can be predicted within 10 dB for normal-
hearing persons and for those with a degree of hearing
loss ranging from mild to profound. It is also clear that
ABR can predict tonal auditory threshold for frequen-
cies as low as 500 Hz.
Clearly, ABR is not a measure of hearing, and hearing
assessment is never complete until a behavioral audi-
ogram is obtained. However, the relationship between
ABR threshold and pure-tone average allows the au-
diologist to obtain accurate predictions of sensitivity
many months before reliable behavioral measures are
feasible. When carefully performed, ABR evaluation is
a reliable indicator of presence or absence of hearing
impairment (screening)as well as actual hearing thresh-
olds.
REFERENCES
Don M, Allen AR, and Starr A. Effect of click rate on the latency of
the auditory brainstem response in humans. Ann Otol Rhino1
Laryngol 1977;86:186-195.
ABR for Objective Measures of Hearing 29
Don M, Eggermont JJ, and Brackmann DE. Reconstruction of the
audiogram using brain stem responses and high-pass noise masking.
Ann Otol Rhino1 Laryngol 1979;88(Suppl 57)3: 1-20.
Don M, Elberling C, and Waring M. Objective detection of averaged
auditory brainstem responses. Scand Audiol 1984;13:2 19-228.
Elberling C and Don M. Quality estimation of averaged auditory
brainstem responses. Scand Audiol 1984;I3(3): 187- 197.
Elberling C and Don M. Detection functions for the human auditory
brainstem response. Scand Audiol 1987a;16:89-92.
Elberling C and Don M. Threshold characteristics of the human
auditory brain stem response. J Acoust S o c Am 1987b;81 :I 15- 12I .
Elberling C and Wahlgreen D. Estimation of auditory brainstem
responses, ABR, by means of Bayesian reference. %and Audiol
1985;14~89-96.
Hecox K and Galambos R. Brainstem auditory evoked response in
human infants and adults. Arch Otolaryngol 1974;99:30-33.
Hyde ML, Malizia K, Riko K, and Alberti PW. Audiometric esti-
mation error with ABR in high risk infants. Acta Otolaryngol
(Stockh) 1991;111:212-219.
Jerger J, Oliver T, and Stach B. Auditory brainstem response testing
strategies. In Jacobson JT, Ed. The Auditory Brainstem Response.
San Diego: College-Hill Press, Inc., 1985.
King AJ and Sininger YS. Electrode configuration for auditory brain-
stem response audiometry. Am J Audiol 1992drch:63-67.
Picton TW and Hillyard SA. Human auditory evoked potentials. 11.
30 Sininger
Effects of attention. Electroencephalogr Clin Neurophysiol
1974;36:I9 1-1 99.
Sininger YS and Don M. Effects of click rate and electrodeorientation
on threshold detectability of the auditory brainstem response. J
Speech Hear Res 1989;32:880-886.
Sininger YS and Masuda A. Effect of click polarity on ABR threshold.
Ear Hear 1990;1 1(3):206-209.
Stapells DR, Picton TW, Perez-Abalo M, Read D, and Smith A.
Frequency specificity in evoked potential audiometry. In Jacobson
JT, Ed. The Auditory Brainstem Response. San Diego: College-
Hill Press, 1985:147-177.
Acknowledgments: Portions of this work were supported by NIH R-
29 DC00021.
The author wishes to thank Dr. Manuel Don for Figures 2 and 3 and
for continued support in the use of F
, in the clinic. Dr. Barbara Cone
Wesson provided valuable editorial comments on the manuscript as
well as collaboration and support for the infant work mentioned. The
author also wishes to thank Dr. Ellen Ma and Ann Masuda, who
assisted in data gathering and analysis.
Address reprint requests to Y. Sininger, Ph.D., Director, Children's
Auditory Research 8, Evaluation Center, House Ear Institute, Fifth
Floor, 2100 W. Third Street, Los Angeles, CA 90057.
Ear and Hearing, Vol. 14, No. 1, 1993