Comparison of 3 ABR Methods for Diagnosis of Retrocochlear Hearing Impairment

Background

Modern otoneurosurgical techniques and methods of radiotherapy (such as the use of γ-rays, the ‘gamma knife’) make it possible to remove small acoustic neuromas without affecting hearing thresholds, even when the patient’s hearing sensitivity is normal [1–5]. However, to effectively use these techniques, one must detect acoustic neuromas or tumors of the cerebellopontine angle at a very early phase, when their size does not exceed 1 cm in diameter. For this reason, the key issue is to use screening methods with high sensitivity and specificity, and diagnostic methods that allow for early detection. Electrophysiological methods satisfy these requirements, and are often based on recording auditory brainstem responses (ABRs) or using magnetic resonance imaging with gadolinium (MRI-Gd). The latter is considered the gold standard in diagnosis of acoustic neuroma [6].

Because of the increased accessibility of MRI examinations and its high sensitivity for diagnosis of small acoustic neuromas, some researchers have in recent years developed the view that to detect small acoustic neuromas (smaller than 1 cm) one may give up ABR testing altogether and rely only on MRI [7]. However, taking economic considerations into account, many authors still believe that a combination of ABR and MRI methods remains a better option [6,8–10]

At the same time, clinical symptoms in retrocochlear pathologies are not very specific, so it is difficult to accept that one should directly refer all patients with a suspected retrocochlear pathology for an MRI examination – because MRIs are much more costly than ABR tests (in Polish conditions approximately 6 times more expensive). Such a process would lead to a large number of false-negatives, low predictive value, and significantly increased diagnostic cost. Additionally, one cannot overlook the problem of claustrophobia in some patients, who are reluctant to subject themselves to MRI examination. On this basis, it still seems reasonable to adopt a two-stage procedure in which all patients suspected of having a retrocochlear pathology are first subjected to an ABR examination and then followed up with an MRI examination at a second stage if necessary [2,10–12]. Obviously, the ABR method used must satisfy the criteria for an efficient screening method: it must have high sensitivity and specificity, and be inexpensive and easy to apply on a large scale.

In the case of acoustic neuromas and tumors exceeding 1 cm, the standard ABR method (ABR STD), which relies on click-evoked responses, has very high sensitivity (over 95%) and high specificity [10,13]. However, if the size of the tumor is less than 1 cm, sensitivity of the ABR STD method drops quickly, according to some authors to 85%, or even to 63% [7,11,14]. Because of this limitation, researchers have for many years been seeking an ABR method which, with appropriate stimulus and analysis, might give enough sensitivity to detect small tumors.

One successful approach has come from the investigations of Don, who has developed the so-called stacked ABR method. Its most important feature is very high sensitivity in detecting small tumors, comparable to that of the MRI method [15–18] The method is based on the recording and registration of so-called derived band responses that originate from narrow segments of the basilar membrane, and which can be obtained by applying a masking technique employing high-pass filtered noise. Such an approach provides an integrated response, the stacked ABR, which represents activity of the whole cochlea as generated by nearly all auditory nerve fibers. Deterioration of transmission through any group of fibers (e.g., that resulting from pressure from an acoustic neuroma) significantly decreases the amplitude of the integrated response. However, popularity of the stacked ABR method is limited by the fact that it is more troublesome and time-consuming than a standard ABR examination, and because the availability of commercial ABR-measuring devices equipped with a stacked ABR option is poor. Yet another disadvantage of the method is that it is based on measuring wave V amplitudes, which have high variability and low repeatability [19,20]. The result is that the stacked ABR method, despite its obvious advantages, is very rarely used in clinical practice.

An attempt to replace the time-consuming registration of derived-band responses used in the stacked ABR method by much more easily obtained responses evoked by tone-bursts was undertaken by Philibert and co-workers [21]. Although tone-burst stimuli are not as often used as click stimuli, their usefulness in electrophysiology is well documented [22–25]. The comparison done by Philibert showed that the basic characteristics of the stacked ABR waveforms are similar, regardless of whether one uses the derived-band technique or responses evoked by tone-bursts. This conclusion applies to both normal ears and ears with sensorineural impairments. However, in the approaches used by both Don and Philibert, [15,21], the responses must be analyzed using the Stacked-ABR software, which is implemented in only one commercially-available device produced by Bio-Logic – Natus [26].

The results achieved by Don and Philibert have encouraged other authors to develop their own methods of stimulation and analysis, free of the drawbacks of the Stacked-ABR method, but which could still provide early detection of small retrocochlear lesions in different groups of auditory nerve fibers [27–30]. As this paper will attempt to show, there are reasons to believe that one satisfactory approach is a method developed by Kochanek et al. [29] called the ABR TP technique. The method is based on registration of responses evoked by Gaussian-shaped tone-pips with frequencies of 1000, 2000, and 4000 Hz. The rise and fall times of these pips are twice as long as those used in standard ABRs for objectively evaluating hearing threshold: for tone pips of 1000 and 2000 Hz, the rise/fall times are 4 ms while at 4000 Hz the figure is 2 ms. The longer rise times mean that the stimulus bandwidth is approximately halved, which improves frequency specificity and the ability to detect retrocochlear lesions.

Results of studies by many authors [31–38] have shown that the increase in rise/fall time of a tone-pip leads to the reduction of wave V amplitude and increase in its latency. This is because the auditory nerve impulses generated by such extended stimuli are averaged over a longer time, and effectiveness of averaging is poorer than in the case of a shorter rise time. This can be explained by a decrease in synchrony of the averaged neural impulses, which in an extreme case may lead to total disappearance of the averaged response – despite the fact that the nerve potentials are still generated in auditory neurons. Then, considering the effectiveness of the averaging process, we may say that the responses to a tone-pip of a longer rise/fall time are poorly synchronized with the stimulus, compared to those evoked by a stimulus of shorter rise/fall times. It makes these responses weaker, as far as response amplitude is concerned, and more sensitive to different factors, such as reduction of stimulus intensity or retrocochlear pathology.

In summary, using tone pips with relatively long rise/fall times ensures, on the one hand, adequate frequency specificity of the response, and, on the other hand, makes it more sensitive to disturbance of synchronization in the auditory nerve caused by the presence of a neuroma or other pathology.

Until recently, direct comparison of the three methods, evaluating their sensitivity, specificity, and predictive values, has not been done. From reports published so far, it is known that both the stacked ABR and ABR TP methods have greater sensitivity than the standard ABR STD method in detecting small acoustic neuromas. Preliminary results comparing the specificity and selectivity of the three methods have shown that the stacked ABR method offered very high sensitivity, approaching 100% [39]. At the same time, however, it yielded the greatest number of false-positives, which made its specificity much lower than that of the other two methods. A full assessment requires a comparative study on the same group of patients. In this way, it is possible to directly compare the clinical usefulness of each method for detecting retrocochlear impairments.

The aim of the study was to compare the sensitivity and specificity of three ABR-based methods for early detection of retrocochlear impairments. Ideally, the comparison should lead to the development of an optimal procedure for screening for these pathologies.

The three ABR methods evaluated were:

Material and Methods

MATERIAL:

The study was carried out on material consisting of 152 persons. The starting point for the study was an evaluation of the results of magnetic resonance imaging (MRI) with gadolinium (GT1W MR images) by 3 independent expert radiologists. On this basis, the examined subjects were divided into 2 groups:

The subjects gave written informed consent prior to participation in the study. The research procedures were approved by the Ethics Committee of the Institute of Physiology and Pathology of Hearing, Poland. Among the subjects in group NR, there were 47 persons with normal hearing, volunteers recruited from employees of the Institute of Physiology and Pathology of Hearing, and 76 patients admitted to the Institute for diagnostic examinations aimed at excluding possible retrocochlear pathology. In this latter group, the majority reported tinnitus, episodes of vertigo, or balance disorders. In 76% of cases, hearing thresholds were no higher than 20 dB HL; in the others, hearing thresholds were elevated but did not exceed 50 dB HL over the frequency range 0.25–8 kHz. Average hearing thresholds in both groups are listed in Table 1 and illustrated in Figure 1.

According to the guidelines for the stacked ABR method by BioLogic [26], and to comply with the conditions required for the ABR STD and ABR TP methods, we excluded from both groups cases of:

EXPERIMENT DESIGN:

In all ABR examinations, we used the system for registration and analysis of auditory evoked potentials AEP NavPro v. 6.2.0 (BioLogic – Natus) which was equipped with hardware and software suitable for applying the stacked ABR method. The stimuli were presented through ER2 BioLogic broadband earphones. Non-disposable silver electrodes were placed at C_Z, M_L (left mastoid), and M_R (right mastoid), as recommended. The tests were carried out in acoustic chambers.

In the stacked ABR method, we used the measuring procedure described in the literature [15–18,26] for implementation in the AEP NavPro system. In the NavPro system, most of the measurements are performed automatically, so there is no option for the user to modify the parameters or interfere with the procedure (such as interrupting the measurement cycle or repeating a single registration). The number of sweeps is determined by the system based on signal to noise ratio, and the averaging process is continued until the residual noise level reaches approx. 20 nV rms; it requires between 1900 and 9720 sweeps (mean approx. 6900) for each trace. Control by the user is limited to maintaining a low level of noise by minimizing disturbances, and correcting waveform fitting (when the stacked ABR response is created by summing derived-band responses). The result of a test (positive or negative) is determined automatically based on the stacked ABR wave V amplitude (A) and/or interaural amplitude difference (IAD) after comparing them with normative values. Normative values are taken to be 800 nV (males) and 875 nV (females), with the normative relative interaural amplitude difference equal to 10% (both genders).

During registration of signals in the stacked ABR method, care was taken to comply with the requirements laid down by the system manufacturer, in particular, maintaining an adequately low level of EEG background noise and eliminating other artifacts such as sensorimotor potentials. To this end, patients were seated in a comfortable, semi-reclining position, and instructed to keep their eyes closed and not to move during the tests (many fell asleep after a few minutes). In cases of excessive noise and/or myogenic artifacts, the protocol was repeated.

Examinations started with the ABR STD test; in cases where we observed excessive disturbance and noise (due to patient fatigue or restlessness) the tests were postponed to the following day, otherwise, we continued with registration of stacked ABRs, and then ABRs evoked by tone pips (the ABR TP method). Registrations of auditory evoked potentials were performed once in every subject.

In measurements with the ABR STD method, we used a click stimulus of alternating polarity presented through TDH 39 (Telephonics) earphones at a level of 90 dB nHL with a repetition rate of 31/s. The amplifier bandwidth was 100–1500 Hz, amplification 100 000 times, and analysis time 20 ms. The number of sweeps required for each averaged response was between 512 and 1024 (average approximately 750 sweeps); 2 traces were registered for each ear to control repeatability of the response. We measured the latencies of waves I, III, and V and the inter-peak time intervals I–III and III–V; interaurally, we measured latency differences of wave V and other inter-peak time intervals. If the wave V interaural latency difference and/or the interaural difference in inter-peak time intervals exceeded 0.2 ms, the result was considered abnormal. The upper normal limit for wave V latency was assumed to be 6.2 ms.

In the ABR TP method, we used tone pips of Gaussian envelope (no plateau), whose rise and fall times were equal to 4 periods at 1000 Hz and 8 periods at 2000 and 4000 Hz. The stimuli had alternating polarity, a repetition rate of 31/s, and were presented through TDH 39 (Telephonics) earphones. The amplifier bandwidth, amplification, and analysis time were the same as those used in the ABR STD method. However, the number of sweeps required to obtain a well-averaged response was greater, ranging from approximately 800 to 1500 (average 1550 sweeps). Two traces were registered for each stimulus in order to control repeatability of the response. In the tone-pip evoked responses we analyzed morphology of the waveforms; latency of wave V, LV; and the interaural latency difference of wave V (IT5). The upper limits of norms for wave V latency are presented in Table 2 [40]. Responses in which 1 of these values exceeded the norm were considered abnormal [28,40].

ANALYSES:

Waveforms of evoked potentials recorded with ABR STD and ABR TP were analyzed by 3 independent experts. The person who evaluated the results had no information about assessments made by the other experts. In a few cases of diverging opinions, the experts met to agree on the assessment.

Statistical analyses were performed with Statistica v.7.1–10.2 (StatSoft Inc., Tulsa, OK, USA) and Matlab (The MathWorks, Natick, MA, USA). For statistical analysis, we used descriptive statistics (scatter plots, t-tests), generalized linear regression, nonparametric statistics (Kolmogorov-Smirnov 2-sample test), and ROC analysis for assessing screening test effectiveness. Generally, we used a confidence level of p=0.05.

Results

To illustrate the usefulness of each method for early diagnosis of retrocochlear impairment, examples of test results are presented in Figures 2 and 3. These results were obtained from the 3 ABR-based methods in otologically normal males (Figure 2) and in patients with confirmed retrocochlear pathologies (Figure 3).

In Figure 2, retrocochlear processes were excluded by MRI-Gd examination. In both cases, results of examination with ABR STD and ABR TP were normal (negative). Interestingly, however, the result of examination with the stacked ABR method was abnormal (positive) in both subjects. In the first case, (a), the stacked ABR wave V amplitude was below the limit of the norm, and in the second case, (b), despite a high stacked ABR wave V amplitude, the result of the test was abnormal (positive) because of an excessively high interaural amplitude difference (IAD >10%).

Figure 3 shows examples of tests done with the 3 ABR methods in patients with radiologically-confirmed small (<1 cm in diameter) acoustic neuroma. In patient a, the result of the ABR STD method was normal (negative), but results obtained with the ABR TP and stacked ABR methods were both abnormal (positive). In patient b, results were abnormal in all 3 methods, but in this patient (a male), stacked ABR wave V amplitudes in both ears were normal (greater than 800 nV) and only the interaural amplitude difference (IAD=42%) exceeded the norm.

These examples indicate that the stacked ABR method might not be a reliable tool for assessing retrocochlear pathologies because of apparent random variability of stacked ABR amplitude observed in both normal-hearing and pathologic ears.

Table 3 shows results of the three ABR methods applied in the two groups of 148 subjects (N=296 ears). Numbers indicate positive results (true positive, TP, and false positive, FP), and negative results (true negative, TN, and false negative, FN). Also shown are calculated values of sensitivity, SEN, specificity, SPE, and positive predictive value, PPV=TP/(TP+FP) of each method. The same data are plotted in graphical form in Figure 4.

In assessing the stacked ABR method, we applied normative values of stacked wave V amplitude as assumed in the AEP NavPro system (800 nV for males and 875 nV for females). The supplementary criterion based on interaural amplitude difference IAD (IAD <10% for the norm) was not taken into account. Because of limited sample size, the values of sensitivity and specificity carry uncertainty. Confidence intervals shown in the table are based on the assumption that the number of TP and TN results are independent random variables that follow a binomial distribution, with probability of success equal to SEN and SPE, respectively, and the number of trials is equal to the number of ears with and without pathology, respectively. We assumed confidence levels of 0.05 and 0.95 [41]. In a similar way, one may estimate the interval of uncertainty for positive predictive value (PPV).

One can see that the stacked ABR method gives the highest sensitivity (well over 90%). Unfortunately, this sensitivity is obtained at the expense of specificity, which is extremely low, and therefore the positive predictive value is also low. This is due, among other things, to the significant variability of stacked ABR amplitudes (illustrated in the examples of Figures 2, 3), which leads to a large number of false positives.

Detailed analysis of the results shows that the performance of the other two ABR methods depends on the kind of retrocochlear pathology and the size of the acoustic neuroma. Unfortunately, the number of acoustic neuromas in the examined population was too low to reliably determine an exact relationship between the tumor size and the test result. Nevertheless, it is clear that ABR TP has an advantage over ABR STD. All false negatives in ABR STD pertain to cases of small tumors (all below 1 cm), but all of them were positive in the ABR TP method. Indeed, we observed that in all cases of tumors, irrespective of their size, the ABR TP test results were positive.

Both the ABR STD and ABR TP methods exhibited worse sensitivity in cases of neuro-vascular conflict. Over 70% of these ears tested positive in the ABR TP method, but the ABR STD method gave much poorer results – about 60% of these ears tested negative (i.e., false negatives).

The stacked ABR method applied with standard normative values of stacked ABR amplitude (the values assumed in the system) had very high sensitivity (above 90%), but its specificity was extremely low, much lower than either of the other 2 methods. This makes its usefulness questionable. One might suspect that the normative values for stacked ABR amplitude, as given by the system manufacturer, were too high, and an improvement in effectiveness might come from modifying the normative values. Table 4 presents test results and characteristics obtained for modified normative values − 700 nV for males and 750 nV for females – derived from previous experiments by the authors. Comparison of the 2 sorts of stacked ABR tests is also illustrated in Figure 5.

As one can see, in this case sensitivity drops to about 70% and specificity remains very low (well below 50%, which is still unsatisfactory). Consequently, the positive predictive value of the method is almost the same as that obtained with the previous normative values. Such a low PPV value means that 80–90% of positive results are false positives. Consequently, if one applied such a screening test, the number of cases referred to MRI examination would be 10 times more than the actual number of pathologies, leading to a substantial increase in the cost of diagnosis, and casting doubt on any screening test based on such a method (instead, it might be better to do an MRI examination on all patients suspected of retrocochlear pathology, as suggested by some authors [7].

Considering the practicality of each method for clinical use, it is of interest to compare the test duration of all 3 methods. Figure 6 plots the examination times of the stacked ABR, ABR TP, and ABR STD methods. The times pertain only to the period of registration; preparation time (e.g., electrode montage and system start-up) is not taken into account.

Discussion

The assessment of the sensitivity and specificity of the stacked ABR method, based on our investigations and presented here, differs from the analyses published by the original authors of the method [15–18]. In their investigations, the reported sensitivity and specificity were much better, and ROC analysis showed much greater effectiveness (area under the curve above 0.9). However, one must take into account that those investigations were carried out in a controlled, research environment, with the use of dedicated instrumentation and software originally created by Don and co-workers. In this way, one might expect very good stability and repeatability. For example, as shown in [18], the scatter of stacked ABR amplitude was reasonably low, and one could easily differentiate between normal and pathology. Our investigation generated a much greater variability of stacked ABR amplitude, and more closely overlapping amplitude distributions for the normal and pathology groups. Another factor that might have contributed to higher sensitivity/specificity values in the work by Don and coworkers was that the reference groups were recruited from normal-hearing, experienced, and cooperative volunteer subjects, who were compared with patients with diagnosed pathology. The situation changes when we deal with a population of actual patients; many have physiological or psychological deficits, for whom lengthy electrophysiological tests are difficult to withstand, and for whom high levels of stimulus and masker noise cause irritation and fatigue. Even in a group without retrocochlear pathology, we can expect more disturbances and instability in ABR waveforms, and this strongly affects the ABR amplitude. On the other hand, there is no evidence that the working parameters of hardware and realization of software algorithms in the commercial AEP Nav-Pro system are identical to those originally used by the method’s authors. In particular, a key parameter is the level of residual noise that remains after the averaging process is terminated (in theory, it should not exceed 20 nV rms). This level is measured and displayed by the system, but there is no way of verifying the correctness of this value, nor can one increase the number of sweeps over the fixed maximum (9000 sweeps) to improve the S/N ratio.

Nonetheless, the investigations performed in this study confirmed the notion, evident from our previous works, that one can obtain high sensitivity and specificity using methods based on the morphology and latency of auditory evoked responses. The standard method based on click-evoked responses (ABR STD) has very high specificity – in our case well over 90% (as it generates a small number of false positives) – but its sensitivity is unsatisfactory. Unfortunately, in cases of small tumors one can barely notice abnormalities in click-evoked ABRs [27,42–44]. Much better results can be obtained using the method based on ABRs evoked by appropriately specified tone-pips (ABR TP). Previous reports by the authors have shown that it can provide very high sensitivity and satisfactory specificity [28,40]. This finding has been confirmed by the present investigation.

Yet another advantage of the ABR STD and ABR TP methods is the short time needed for examination. The first method needs only 4 recordings of ABRs evoked by clicks (recordings are repeated twice in each ear); the second requires 12 recordings (at 3 frequencies, twice for each ear), and the number of sweeps is relatively low. In the stacked ABR method, one must record 12 waveforms, and the registrations usually take much longer because of a lower stimulus level and the presence of masker noise. Consequently, the number of sweeps is much greater (up to 9000) and the whole procedure is several times longer.

We might consider the option of combining the ABR STD and ABR TP methods. The proposed strategy of testing for retrocochlear pathologies would then be as follows: first, one applies the standard ABR, which is simple, quick, and has high specificity. If the result of this test is abnormal (positive), the patient would be referred to MRI examination to confirm the diagnosis. Negative results would be followed up at a second stage using the ABR TP method, which is more sensitive than the ABR STD. Normal (negative) results of both tests would justify not proceeding to an MRI examination and instead referring the patient to clinical observation.

Running the 2 tests in series, according to the above strategy, may significantly increase sensitivity, as expressed by the following formula [45]:

At the same time, the effective specificity is slightly reduced, but it remains relatively high if both tests have sufficient specificity:

where SEN1 and SEN2 denote sensitivities of the first and second tests, and SPE1 and SPE2 denote their specificities.

The above formulae are valid, however, when the tests are carried out in a large population, and true values of sensitivity/specificity are known. Taking into account the results obtained in the examined group presented in Table 2, one might expect that combined strategy of ABR STD and ABR TP would give an effective sensitivity of 94.3% and a specificity of 87.7%. In reality, in our study, we could hardly obtain any improvement in sensitivity applying the 2-stage strategy compared to the application of the ABR TP method only. Sensitivity and specificity values are loaded with uncertainly (Figure 4). Besides, in this particular group of patients, all positive cases detected with ABR STD method also tested positive in the ABR TP method. Nevertheless, the 2-stage strategy is still worth using, because a number of pathologies are detected already in the first stage, so it is not necessary to use the ABR TP method, which is slightly more complicated and time-consuming. In conclusion, using the combination of ABR methods, one may expect a significant reduction in cost of diagnosis in comparison with a strategy that refers all patients with suspected retrocochlear lesions to MRI examination.

Conclusions

The advantages of methods based on assessment of latencies of auditory brainstem responses – ABR STD and ABR TP – are:

A clear disadvantage of the method is that the results must be interpreted visually, which calls for adequate qualifications and experience of the audiologist performing the tests.

The advantage of the stacked ABR method consists in its very high sensitivity (when one applies settings recommended by the manufacturer); however, the disadvantage is very low specificity. An automatic measuring process and an automatic decision about the outcome of the test are also advantages, which avoid errors made by the audiologist. Nevertheless, the stacked ABR method as implemented in the AEP NavPro system has several serious disadvantages, including:

Other disadvantages that one must take into account are:

The results presented above show that the stacked ABR method, in its present version, is not yet an optimal clinical tool for screening for retrocochlear pathologies.

At the same time, the results obtained from the other 2 ABR-based methods indicate that a practical solution for screening for retrocochlear pathologies is to apply ABR STD and ABR TP in series.