in practice
F O R C L I N I C A L A U D I O L O G Y
Febru a r y 2 0 1 0
The N1-P2 Cortical Auditory
Evoked Potential in threshold
estimation
Guy Lightfoot, Ph.D.
Bio: 1960s using analogue computers and later in the 1970s when
digital computers became widely available. Referred to at the
Guy Lightfoot, Ph.D. obtained time as the Slow Vertex Response, Hallowell Davis and colleagues
his first degree in Physics then an pioneered the use of CAEPs in a clinical setting. N1 & P2 are
MSc in Audiological Science in the thought to have multiple generators in Heschl’s gyrus (auditory
1970s before doing his PhD on the cortex), and have latencies of 100-160 ms and 160 – 270 ms
effects of repetition rate on the respectively, depending on the sensation level of the stimulus. The
neurological ABR. Since 1976 he N1-P2 amplitude is typically 10-15 µV for louder stimuli, reducing
has developed and managed an as threshold is approached. It is thought that N1 reflects the
adult diagnostic audiology service detection of a change in the acoustic environment. Like the ABR
at the NHS Royal Liverpool University Hospital, UK, taking a therefore, it can be considered as an onset response.
special interest in electrophysiology, vestibular assessment,
teaching, instrumentation & calibration. For the last 18 years
he has organized and contributed to a highly successful annual Clinical utility
ERA & OAE training course. He is currently active in establishing
national protocols, staff training and quality assurance of ABR The use of the N1-P2 CAEP response in the estimation of hearing
testing for the English newborn hearing screening program. Guy sensitivity is well established, with most studies suggesting that
has an enduring interest in CAEPs and has long campaigned for threshold estimation in adults is accurate within 10 – 20 dB.
manufacturers to provide efficient software for this underused For example, Lightfoot and Kennedy (2006) identified a mean
test. His non-commercial educational resource web site can be audiometric / electrophysiological threshold difference of 6.5
found at: www.corticalera.com dB, after correction of which 94% of threshold estimates were
within 15 dB. That’s better than most estimates of adult ABR
accuracy. Our study revealed that the test can be time-efficient
Background with appropriately designed software that automates predictable
manual tasks such as waveform manipulation: threshold
The N1-P2 complex was the first cortical auditory evoked potential estimation at three frequencies in both ears took an average of
(CAEP) to attract substantial research interest, initially in the mid- 20.6 minutes. With standard evoked potential software the test
does take longer, principally because the user has to manually prediction increases for individuals with a recruiting hearing
control the creation of grand averages, waveform manipulation, loss, the steeper response input-output function seen in such
etc. Probably the most significant clinical limitation of the N1-P2 cases yielding larger amplitude responses than non-recruiting or
response as an audiological tool is its late maturation, extending normally hearing individuals at stimulus levels just above their
into the late teenage years, though the technique remains threshold. In some countries (e.g., the UK) the test is established
viable in older children (Stapells, 2002). It is not appropriate for by legal case law and government military pension schemes as
audiological use in infants, for whom the ABR (in its transient the ultimate objective arbiter of hearing status.
or 80 Hz ASSR guise) is a superior instrument because of earlier
maturation of the relevant response generators. Devotees of the
cortical response will know that the infant P1 (which is probably N1-P2 response characteristics
fused with the P2 response at that age) can be recorded at supra-
threshold stimulus levels but that response cannot be used to
& the practicalities of
reliably estimate auditory sensitivity. recording
The response can be evoked by the relatively abrupt onset (or
Disadvantages over the ABR offset, or change) of an auditory stimulus. Frequency-specific
• Adults & older children only stimuli are desirable for audiological applications and that
• Time & laborsaving software not yet usually means a tone burst. The amplitude of both ABR and
available CAEP responses are larger for stimuli with a more abrupt onset
but a short rise time and duration limits frequency specificity.
For the ABR, a commonly accepted compromise between
these competing requirements is a 2:1:2 cycle (rise/plateau/
So, why is this response fall) tone burst. A rise time substantially longer than a couple of
almost unknown? milliseconds “smears” the ABR in time because ABR latencies
are so short. We can see that effect in the 500 Hz ABR: a rise
The N1-P2 response has been criticized for its variability and time of 2 cycles at 500 Hz is 4 ms which is substantial compared
susceptibility to the effects of subject drowsiness (Näätänen, to the Wave V latency and we usually see a rather ill-defined
1992). Indeed some texts on auditory evoked potentials appear ABR as a result. The much longer latencies of the N1-P2 complex
to dismiss the technique as a viable audiological test (McPherson, gives us the freedom to use longer stimulus rise times; a typical
1966; Hall, 1992) perhaps because less than optimal test value being 10-20 ms. Stimulus plateau is typically 60 ms (see
parameters have led to poor test performance. This is probably Frequency Conversion Table and Trial Settings). By that time the
why the technique has not been widely used or taught, with tone response either will or will not have been evoked so there is
burst ABR usually being the preferred method in many countries.
The N1-P2 response does appear to be poor in a small proportion
of individuals, with results overestimating the threshold by 20 Advantages over the ABR when testing
dB or more. The resulting lack of clinical demand has stifled adults
software development by equipment manufacturers. In contrast, • Better threshold accuracy
Hyde (1997), Stapells (2002) and Martin et al. (2007) considered • More frequency-specific
it to be the test of choice for threshold estimation of most older • Response morphology does not degrade at
children and adults. Interestingly, individuals exaggerating their low stimulus frequencies
audiometric thresholds often give better responses than honest • Potentially quicker than the ABR (with
subjects, especially for stimulus levels that are audible but below appropriate software)
their volunteered thresholds. This is thought to be associated • Tests more of the neural pathway
with heightened cortical arousal level for these stimuli, which • No patient relaxation needed
are seen as a threat. As with the ABR, the accuracy of threshold
2
little merit in extending the duration of the stimulus beyond this Filters of 1 Hz to 15 Hz provide the best signal to noise
value. One of the advantages of the N1-P2 response is therefore compromise when the response is used to estimate threshold;
the almost ideal frequency specificity it provides, allowing steeply this is a much narrower bandwidth than one would use for
sloping or notched audiometric contours to be resolved. Unlike research where waveform fidelity is important. The amplifier gain
the ABR, the morphology of the N1-P2 response is every bit as / artefact rejection setting should be such that incoming signals
clear for low frequencies as for high frequencies. Response greater than about ±50 µV (e.g. 60%) are rejected. The choice
amplitude does decline at frequencies above about 3 kHz (see of repetition rate is an interesting compromise: it takes about
Figure 2a) but we found that threshold prediction at 8 kHz was 10 seconds for the response to fully recover (Davis et al., 1966).
no worse than at lower frequencies (our study used stimuli of 1, 3 A rate above 0.1 Hz will diminish the response amplitude, but
& 8 kHz). faster stimulation allows more averaging and hence greater signal
to noise ratio improvement per unit test time. A repetition rate
In addition to the excellent frequency specificity and good of 0.5 to 1.0 stimuli per second is optimal. However, this means
threshold precision, other advantages over the ABR include testing that the first few stimuli in an averaging sequence (preceded
the integrity of a greater proportion of the auditory nervous by silence) will evoke larger responses and as the averaging
system and the capability to employ speech-based stimuli. run continues the response amplitude will decline. Very long
averaging runs are therefore counterproductive. At each stimulus
For clinical threshold estimation a single recording channel is level, typically two or three waveforms with 10 – 20 sweeps each
sufficient, with a Cz positive electrode site (a high forehead site are sufficient to identify a response at supra-threshold stimulus
will give a significantly attenuated amplitude) and a mastoid levels but 20 – 40 sweeps per waveform are usually needed close
(either one, or linked mastoids) negative electrode site. Ground is to threshold. Recording more than a single waveform allows
conventionally placed on the forehead. us to assess response repeatability (by visual inspection and by
3
computer correlation) and residual noise (as the average gap at supra-threshold test levels can be helpful in the correct
between replicates). For this process to be valid, the response identification of near-threshold responses.
should be stationary, i.e., not change with time. But we know
that this response declines with time during an averaging run and Figure 1: N1-P2 intensity series
can vary with the patient’s level of arousal. These effects can be
minimized in dedicated software by acquiring the 2 or 3 replicates
pseudo-simultaneously or, less ideally, by manually ensuring that a
silent period of at least 10 seconds separates the averaging runs.
A recording epoch (time base) of 500 ms to 1000 ms is usual, the
latter allowing some pre-stimulus activity to be recorded as an
indication of background non-response electrical activity.
Unlike shorter latency responses which place high demands on
muscle relaxation, CAEP subjects simply need to remain awake
and reasonably alert during testing and this can readily be
achieved by asking them to sit upright (not reclined) in a chair
and browse a magazine or watch a silent video. Eye closure is
to be avoided since this is associated with EEG alpha (8 – 12 Hz)
activity which can contaminate the recording as well as increased
drowsiness. Asking the subject to listen to or count the stimuli is
unnecessary.
Response identification
Figure 2: Input-Output functions of the N1-P2 CAEP for (a)
Although some objective assessment tools such as measurement amplitude; (b) latency. From Lightfoot & Kennedy (2006).
of signal to noise ratio and cross-correlation are available on
most evoked potential equipment, N1-P2 response identification
(like ABR response identification) is usually based on subjective
judgment by the audiologist and is therefore vulnerable to
operator error or bias.
Figure 1 illustrates N1-P2 CAEPs at a number of stimulus levels.
The stimulus was a 120 cycle plateau tone burst at 2 kHz with 20
cycle linear rise and fall times. See cortical protocol trial settings at
the end of the document.
In Figure 1 the left panel shows replicated responses; the
right panel shows the grand averages of the replicates. The
manner in which response amplitude and latency change as
stimulus level is changed is, for frequency-specific stimuli, quite
predictable though influenced by the extent of any loudness
recruitment present. Figure 2 (from Lightfoot and Kennedy, 2006)
demonstrates input-output (I/O) functions for the N1-P2 complex.
Knowledge of these functions together with definite responses
4
to be overlooked in clinical practice. Additional averaging, leading
to a lower residual noise, is the most obvious means of resolving
inconclusive waveforms.
Case Study: Medico-legal
noise trauma
Figure 3 is the audiogram of a 36-year-old telephonist
Figure 3
The appropriate identification of the lowest level at which a
response is seen is only half of the threshold estimation process;
the other is the highest level at which a response is absent. For
the latter to be valid, not only must there be no obvious response
but we also need to be sure that the recording conditions are
sufficiently good to ensure that a small response is not obscured
by residual noise. Note that there is an important distinction
between not seeing an obvious response and being able to
demonstrate that a response is absent. A maximum residual
noise level criterion is therefore appropriate when defining
response absence and for the N1-P2 response a value around
1.5 µV is reasonable for this purpose. One way of assessing the who wore her earpiece in the right ear and whose main complaint
residual noise is by superimposing replicates and then visually was that of right-sided continual hissing tinnitus following an
estimating the average gap between them, across the entire incident in which her right ear received a very high level noise for
recording window. At points where the waveforms cross, the gap a few seconds during equipment malfunction. She reported her
is zero (obviously) whereas at other latencies the gap will be a hearing as reasonably good bilaterally (audiometric thresholds
maximum. At stimulus levels where there is no obvious response were ≤20 dB HL on the left). The authenticity of the 6 kHz
present and the noise estimate is acceptably low, one can state threshold on the right (and by inference, the existence of the
with reasonable confidence that this stimulus level is below tinnitus) was questioned by the defendant and CAEPs were
threshold. Of course it is possible that some recordings may fail commissioned.
to meet the acceptance criteria of either the response presence or
response absence status. Examples include “probable” responses Figures 4 & 5 are the CAEP waveforms at 4 kHz & 6 kHz
that have a signal to noise ratio below an acceptable value and respectively. The dotted vertical line denotes stimulus onset. Full
“probably absent” responses where the residual noise level is details of test parameters and procedure are given in Lightfoot
above an acceptable value. Such waveforms must be graded and Kennedy (2006). The residual noise (RN) in each waveform
as inconclusive and though tempting, must not be allowed to is a function of the number of sweeps in each average and was
contribute to the process of threshold definition. This is equally calculated as the mean of the modulus of the differences between
true of ABR responses, though sadly, it is common for this issue the sub-averages across the entire recording window. This is the
5
same as the subjective noise estimation method suggested above. responses at 60, 40 & 30 dB HL whilst at 20 dB HL the objective
At the highest stimulus level in Figures 4 & 5, only 15 sweeps measures suggest there is no response. Subjectively it is tempting
were used (bold lines, showing the grand average of the three to grade this as a “possible response” but we must resist! The
sub-averages which were the result of only 5 sweeps each) since minor bump suggesting a P2 is predominantly from only one of
the response was clear even in the presence of a moderately the three sub-averages and is therefore probably just residual
high residual noise. At lower test levels the declining response noise.
amplitude leads to a declining signal to noise ratio (S/N) and 30
sweeps were used to obtain a lower RN and hence a good S/N. We employ a rule whereby if the amplitude of the lowest level
Here, S/N was calculated as the N1-P2 amplitude divided by RN. response is >3 µV (>5 µV at 1 kHz or below) then the threshold is
At the sub-threshold levels 45 sweeps were needed to obtain a taken as 5 dB below this level. The CAEP result therefore predicts
sufficiently low noise floor to ensure the response was absent. a threshold of 25 dB HL at 4 kHz, validating the audiogram. The
6 kHz responses suggest a threshold of 55 dB HL, in complete
The author’s system calculates the correlation between the sub- agreement with the behavioural audiogram. This is recruitment
averages in the region of a suspected response (50 ms prior to sharpening our precision. I just love recruitment!
where the system detects a potential N1 to 50 ms following a
potential P2) and uses this correlation together with the S/N to In this case, CAEP testing was able to confirm the presence of
derive a p-value (the likelihood of the response being spurious) a fairly tight notch in the audiogram. Recall that a stimulus rise
for the response (Lightfoot, 2009). Note how in Figures 4 & 5 time of 10 ms was used. At 6 kHz that represents 60 cycles, not
the p-values are very low for the supra-threshold responses, the 2 cycles used in ABR. It is unsurprising that this stimulus is
supporting the subjective visual interpretation that at these levels sufficiently frequency-specific to probe such audiometric notches.
there are likely genuine responses. At 4 kHz there are unequivocal
Figure 4: Right ear 4 kHz CAEPs Figure 5: Right ear 6 kHz CAEPs
6
References
Davis, H., Mast, T., Yoshie, N., & Zerlin, S. (1966). The slow response of
the human cortex to auditory stimuli: Recovery process. Electroenceph Clin
Neurophysiol, 21,105-113.
Hall, J.W. (1992). Handbook of auditory evoked responses. Boston: Allyn
and Bacon.
Hyde, M. (1997). The N1 response and its applications. Audiol Neuro-
otol, 2, 281–307.
Lightfoot, G., & Kennedy, V. (2006). Cortical electric response audiometry
hearing threshold estimation: Accuracy, speed and the effects of stimulus
presentation features. Ear Hear, 27(5), 443-456.
Lightfoot, G. (2009). Objective detection of the cortical N1-P2 response
using signal to noise ratio. Oral presentation at the XXI Biennial Sympo-
Artifact rejection set at 60% = +/- 50 uV sium of the International Evoked Response Study Group, Rio de Janeiro,
Display Scale recommend 3.00 uV Brazil.
Martin, B.A., Tremblay, K.L., & Stapells, D.R. (2007). Principles and
applications of cortical auditory evoked potentials. In R.F. Burkard, M.
Frequency Ramp (rise/fall) Ramp (rise/fall) Plateau 60 ms
(Hz) 10 ms 20 ms
Don, & J.J. Eggermont (Eds.), Auditory Evoked Potentials (pp. 482 –
507). Baltimore: Lippincott Williams and Wilkins.
250 3 5 15
500 5 20 30 McPherson, D.L. (1996). Late Potentials of the Auditory System. San
750 8 15 45 Diego: Singular.
1000 10 20 60
Näätänen, R. (1992). Attention and Brain Function. Hillsdale. NJ: Law-
1500 15 30 90
rence Erlbaum Associates.
2000 20 40 120
3000 30 60 180 Stapells, D. (2002). Cortical event-related potentials to auditory stimuli.
4000 40 80 240
In J. Katz (Ed.), Handbook of Clinical Audiology (5th Ed). Philadelphia:
6000 60 120 360
Lippincott Williams & Wilkins.
8000 80 160 480
Milliseconds to Cycles Conversion Table at 10 ms (rise/fall),
20 ms (rise/fall) and 60 ms (plateau)
Closing remarks
The N1-P2 CAEP is not without imperfections and limitations
(most important is its inapplicability to infants) but it is a
valuable component of the audiologist’s toolbox. Its success in
hearing threshold estimation lies in the use of appropriate test
parameters, efficient procedure, and rigorous interpretation. If
there is sufficient demand, I hope that equipment manufacturers
will develop and provide labour & time-saving software – so let
them know!
7
Otometrics is the world’s leading manufacturer of hearing and
balance instrumentation and software – innovative concepts
designed to help healthcare professionals make the best possible
decisions.
Our solutions range from infant screening applications and
audiologic diagnostics, to balance testing and hearing instrument
fitting. As an industry leader, we are committed to developing
innovative, integrated solutions that help healthcare professionals
make the best possible decisions.
GN Otometrics, Europe. +45 45 75 55 55. [email protected]
GN Otometrics, North America. 1-800-289-2150.
[email protected]www.otometrics.com
