Ricketts, Todd A.; Mueller, H. Gustav
Probe-microphone verification is an important part of the hearing aid fitting process, and it has been for over 25 years. Like hearing aids themselves, the procedure has evolved, and today it is common to use real speech or speech-like input signals, and use ear canal SPL fitting targets rather than desired “gain.” We welcome the changes, but with the changes come new issues to consider.
The format of Page Ten was altered this month, so that Todd Ricketts and I can report some of our preliminary findings related to these critical issues.
Page Ten Editor
There are many factors to consider in selecting hearing aid gain and output for a patient. Do you choose the fitting that maximizes audibility? The fitting that optimizes speech understanding? The fitting that “sounds the best” to the patient? The fitting recommended by the manufacturer?
Using these or other possible “gold standards” can result in very different gain and output selections. Therefore, it is generally accepted that the best starting point is a validated prescriptive fitting method, and this is what is recommended by best practice guidelines.1,2
Prescriptive fitting methods have evolved over the years, and today we essentially have only two to choose from: the DSLv5.0 and the NAL-NL1. With adults, the most common method has been the NAL, and there is substantial research evidence to support the effectiveness of this algorithm.3,4 It is important to emphasize that, ideally, these prescriptive targets relate to the gain and output delivered in a given patient's ear canal. Therefore, to actually use the method, it is necessary to conduct some type of real-ear verification or use real-ear correction factors during the programming and verification process. With infants and young children, it may be necessary to measure the RECD and then use these values to derive estimated real-ear output. With adults, equipment and protocols are available to conduct real-ear verification measures directly.
A SHORT HISTORY OF PROBE-MIC VERIFICATION
In the early 1980s, a clinically friendly method of real-ear verification, referred to as probe-microphone measures, first became available. By means of a small silicone tube the output of the hearing aid could be measured in the ear canal. When compared to the open-ear gain (REUG) or the input level at the reference microphone, both REIG and REAG values could be calculated.
At the time, the match to target was only for a single input level, the input signal often was a swept pure tone, and the hearing aids being tested usually were single-channel with linear processing. The calculation typically used for verification was the REIG, as this corresponded to the targets provided by the prescriptive methods that were popular for adults at the time (e.g., Berger, Libby, POGO, NAL-R, etc.).
By the mid-1980s, several different probe-mic systems were available. Several studies were conducted at this time to examine the accuracy of this new verification technique.5-7 Other studies focused on whether or not differences would emerge when testing was conducted with probe-mic equipment from different manufacturers.8,9 For example, would a match to REIG NAL-R target using the Madsen IGO also be a match to target when the Rastronics or Fonix system was used for testing? Given that the different systems used somewhat different test protocols and equalization methods, test-retest accuracy of the different probe-mic measures also was of interest.10,11
In general, the research results from this era revealed that for verification of prescriptive targets (e.g., REIG compared to prescribed gain-for-average inputs), probe-mic measures were valid and reliable, and no significant differences were present among the various equipment and test protocols that were available.
A lot has changed since the 1980s, including the hearing aids, the verification process, and the probe-mic equipment itself. Multiple channels of compression are now the standard fitting, making it necessary to have fitting targets for multiple inputs and sometimes to adjust targets for channel summation. A significant change in verification procedures has been the gradual shift away from using the REIG to using the REAR (absolute ear canal values) to obtain the match to target. Additionally, the input signal used today is typically a shaped speech-like signal, or shaped real speech, and this verification process often is referred to as “speech mapping.”Endnote 1
When one conducts probe-mic verification with different equipment using speech mapping techniques, it is tempting to think that NAL targets are NAL targets are NAL targets. Unfortunately, it is not that simple.
Typically, we do not use the NAL-NL1 stand-alone software, but rather the targets generated by the probe-mic equipment. At least five different factors can affect the accuracy of this verification process: target adjustments for the hearing aid/fitting type, conversions used to display hearing loss and targets in ear canal SPL, hearing aid-specific interactions with the measurement signal, the level and shape of the input signal, and the analysis of the measured signal. We'll briefly explain each of these.
Target adjustment factors
First, there are hearing aid factors that can affect the fitting targets. These include bilateral versus unilateral fittings, the number of compression channels, and the type of output limiting that is assumed.
Perhaps the most straightforward of these is bilateral versus monaural fittings; we have observed that all implementations reduce prescribed output for a bilateral fitting. The NAL-NL1 accounts for up to four compression channels, but many products today have more than four. It's important to know if the algorithm used by the probe-mic manufacturer accounts for this.
The NAL-NL1 software (v1.28) also allows for selection of multi-channel, wideband (single-channel), or no output limiting. When probe-microphone equipment manufacturers implement this software, some default to one type of limitation, others default to a different type, while still others allow for the selection of the type of limitation. The specific output limiting selected can have a large effect on the REAR targets. For example, the selection of “multi-channel limiting” will often result in essentially the same output for 50-, 65-, and 80-dB SPL input levels at 2000 and 3000 Hz. This, of course, also affects REIG targets, but is more easily visualized when REAR (speech mapping) verification is employed. We find the use of these multi-channel limiting assumptions and a high number of channels curious, as it can result in target values that are typically unachievable in modern hearing aids (the hearing aid would essentially have to have hard limiting for inputs above 50 dB SPL).
Conversion factors also influence the speech mapping probe-mic fitting targets. Values that once were expressed in HL or in 2-cc coupler are now converted and expressed on the fitting screen in ear canal SPL. Conversion from dB HL to dB ear canal SPL involves using the REDD, which is often calculated by adding the RETSPL and the RECD.
While RETSPLs are standardized values (ANSI S3.6, 1996), they differ depending on whether the audiometric thresholds were determined using insert or supra-aural earphones. There are no standardized RECD values. Hence, a NAL fitting target for a given hearing loss, which likely would be the same in REIG-speak for different probe-mic equipment, could easily end up different when expressed as an ear canal SPL target, simply because a different REDD was used for conversion.Endnote 2 The clinical measure of the REDD (or the RECD) would reduce these inter-system differences (if the system allows you to enter these values), but this is not a common practice.
Input signal and hearing aid interaction factors
There also are issues to consider related to the equipment itself. The input signal is critical, in overall level, shape, and possible interactions with specific hearing aid processing.
Consider this potential problem with two signals that both have an overall level of 65 dB SPL, but different shapes. Let's say a NAL-NL1 target is correctly displayed in ear canal SPL on the fitting screen, indicating that the desired output should be 90 dB SPL at 3000 Hz for a specifically shaped 65-dB input signal. But, what if the shape of the input test signal used was 8 dB lower at 3000 Hz than the signal used by the NAL to derive that fitting target? The hearing aid would appear to be “under fit” in this frequency region, when in fact it may be a perfect fit to target.
Now, if the displayed targets were readjusted to account for the difference in the input signal spectrum, all would be well (or at least mostly well), as the patient would obtain the same REAG. Such are the things we need to consider.
Commercially available probe-microphone equipment varies considerably in the shape of the test signal used. Further, while some systems use real-speech signals (albeit sometimes shaped to a specific spectrum), others use a signal that is expected to be processed by the hearing aid in a way similar to speech. In all cases, probe-microphone manufacturers must consider if fittings completed with their specific test signals lead to similar gain values as those derived when the NAL-NL1 method was validated. Clearly, these potential interactions can be problematic in probe-microphone equipment that allows fitting to the same NAL-NL1 targets with a variety of signals and a variety of shapes. In these cases, you may not be fitting to NAL-NL1 targets at all, or at best you are not going to be “right” all the time.
Differences in compression processing in individual hearing aids can further complicate this issue. Specifically, in hearing aids that use compression, as with any non-linear system, you cannot predict the output for one signal from measurements done with another. For example, while it is possible to quantify the average difference between a match to targets for a real-speech signal and a match to targets for another (non-speech) test signal, individual hearing aids are expected to deviate from this average measurement based on their specific compression settings.
Signal analysis factors
A final issue that can affect the accuracy of the verification is the method used to analyze the measured output. For example, current systems measure the sound level in the ear using different filter shapes ranging from a constant bandwidth (e.g., a group of filters all 86 Hz wide) to partial octaves (e.g., 1/3 octaves, which will increase in bandwidth with increasing frequency). The effect of these different filter shapes is to make the same signal appear to have a different shape when displayed. Specifically, the shape of the signal measured using partial octaves will fall less steeply with increasing frequency than when the signal is measured using constant bandwidth filters. Since the NAL-NL1 originally was derived as an REIG prescription, it is necessary to consider the effect of output analysis when defining REAR prescriptive targets, since they must be displayed in output rather than in gain.
DO DIFFERENT MODELS GIVE DIFFERENT RESULTS?
Given the many changes in hearing aids, fitting philosophy, procedures, and equipment, we concluded that some of the questions asked back in the 1980s should be revisited. In a perfect world (or at least a perfect office or clinic), a probe-mic match to prescriptive target should be either “bad,” “okay,” or “good,” independent of what equipment is used. Yet, we've heard anecdotal reports from people working in clinics with probe-mic equipment from different manufacturers that this isn't true.
We know that conditions in a busy clinic can lead to procedural variances that could result in measurement errors and the appearance of different findings. But, could it be that the match-to-target really is different? That would not be good. Considering the status woe of hearing aid verification in general, we certainly do not want to add another reason not to do it. So, we thought it was time to conduct some controlled comparative measures of speech mapping, carefully using the procedures recommended by different manufacturers with their equipment, to determine if significant differences really exist.
METHODS OF COMPARISON
We compared hearing aids fitted to equipment-specific NAL-NL1 prescriptive targets across three different probe-microphone systems: Audioscan Verifit, Frye Fonix 7000, and MedRx Avant. Testing was conducted on two adult listeners across two different ear canal coupling configurations: open and closed. One BTE hearing aid was evaluated for the closed condition using the targets generated by each probe-mic system for a flat 50-dB-HL hearing loss. A second BTE hearing aid was evaluated for the open condition using the targets generated for a downward sloping hearing loss, 20 to 65 dB HL. Each system's NAL-NL1 targets (displayed in the speech map mode re: ear canal SPL) were recorded for each audiogram, and the equipment's default test signals were used for testing each probe-microphone system.
All system defaults were used except one: wideband compression limiting was chosen in lieu of multichannel compression limiting for the Fonix system, since the other two systems default to this condition. The number of compression channels was also adjusted in both the MedRx and Fonix system to reflect the channels in the hearing aid tested, either 8 or 16. This option is not available with the Verifit. However, this adjustment did not appear to affect targets for average speech input levels for either the MedRx or the Fonix equipment, as long as wideband compression limiting was selected. Real shaped speech (the “carrot passage”) was used for the Audioscan Verifit; the interrupted composite noise (DSP-ANSI) was used for the Fonix 7000; and the “Speech Noise” signal was used for the MedRx Avant.
Prior to experimental testing, each system was calibrated according to the manufacturer's recommendations. The test participants were positioned relative to the loudspeaker(s) as recommended by each manufacturer. Head movement during testing was minimized. Probe-tube depth was assured to be within 2-3 mm of each subject's TM during each test procedure using the “bump and pull” method.
Each hearing aid was first fitted (as closely as possible) to the NAL-NL1 65-dB input targets for one of the three randomly selected probe-microphone systems. The manufacturer's recommended open-test procedure was used for the open fittings for each system. The system-specific NAL-NL1 target (for the default test signal) displayed in the software, the measured REAR, and the signal level (unaided) at the reference microphone position of the probe-microphone system was then recorded. Once these measures had been obtained, no changes were made to the hearing aid, and the same three measurements were then recorded from the two remaining probe-mic systems. After evaluation using all three probe-mic systems, the entire test sequence was repeated to provide a two-run average for a more stable measure for each system.
After completing these procedures, we re-fitted the hearing aids to the NAL-NL1 targets for the second probe-microphone system, and the entire process was repeated. Finally this same procedure was completed beginning with a re-fitting on the third probe-microphone system. In all, six REAR measures were made on each system for each hearing loss configuration for each subject.
Following these measures, we used the REAR and system-specific NAL-NL1 targets to calculate the deviation from NAL-NL1 targets for each system (e.g., different systems could have different targets, different input signals, and different REARs, but the same deviation from target).
Our primary interest was to determine the consistency of a “fit-to-target” (or close fit to target) for three different models of probe-mic equipment. For example, assume that we first used the Verifit equipment and after careful programming our final fitting resulted in a +2-dB deviation from target at 4000 Hz. If we now move the patient to the Fonix equipment, do not change the programming of the hearing aid, conduct an REAR measure, and compare the measured output to the Fonix target for 4000 Hz, what do we get? If the REAR is now 8 dB below target, the deviation between these two pieces of equipment for 4000 Hz would be 10 dB. A Verifit fitting would appear under fit at 4000 Hz on the Fonix, while a Fonix fitting would appear over fit at 4000 Hz on the Verifit.
As we described earlier, we conducted these types of comparisons, with each system serving as the “baseline” for a portion of the testing. The average intersystem deviation from NAL-NL1 targets across the three probe-mic systems for the open fittings are shown in Figure 1.
These results reveal the following trends. The NAL-NL1 fit-to-target obtained using the MedRx and Verifit systems appear roughly equivalent. That is, it appears that in using either of these systems you would program the hearing aid in a similar manner. In contrast, the Fonix system resulted in a desired fitting that was approximately 3-4 dB lower in the low frequencies, and up to 8-10 dB higher in the high frequencies. These results clearly suggest that a good match to NAL-NL1 targets with the Verifit or MedRx systems will not yield a good match to targets using the Fonix system.
CLOSED-EAR CANAL COUPLING SYSTEM
When conducting testing for the open-canal fittings, we used each system's stored equalization method. To examine the consistency of these results, the intersystem deviation from NAL-NL1 targets across the three probe-mic systems were also measured for closed ear canal fittings, using concurrent equalization. These data were collected for different hearing aids and different adult ears. These findings are shown in Figure 2.
While not identical, the inter-system differences for the closed canal condition, seen in Figure 2, clearly show the same pattern as demonstrated in Figure 1 for open-canal fittings. Again, it appears that you would program a hearing aid differently (less gain in lows, more gain in highs) if you were using the Fonix system than if you were using either the Verifit or the MedRx. As before, the differences between the Verifit and the MedRx were very small.
IS COMPRESSION A FACTOR?
As we noted in our introduction, there are a number of reasons for the general pattern of differences across the various probe-microphone systems' implementations of NAL-NL1. One factor, which could account for at least some of the intersystem differences, relates to how compression could interact with the different systems' test signals. That is, for the results we have shown, the hearing aids were evaluated with real-speech signals via the Verifit, but with non-speech signals via the Fonix and MedRx systems.
To examine this issue, the intersystem variation between the Fonix and Verifit systems was evaluated for two subjects for two different compression settings. Specifically, the same commercial hearing aid was evaluated using slow-acting, low-threshold compression (RT∼1800 ms) and fast-acting (syllabic), low-threshold compression (RT∼80 ms); compression kneepoints remained constant. To really push the envelope of the compression question, we also dug out an old analog single-channel linear peak-clipping hearing aid. All testing for these conditions was conducted with a closed coupling configuration (Comply foam earmold). For these comparisons we employed the same round-robin procedure of fitting each hearing aid to the NAL-NL1 targets for a 65-dB input for each of the two probe-mic systems in turn. Intersystem deviations resulting from this testing are shown in Figure 3.
While the compression setting did (as expected) affect the measured gain within each system, the pattern of intersystem deviation across the three amplitude processing configurations was similar. This suggests that the across-system differences are not generally due to the interaction between the specific probe-microphone system test signals and modern non-linear hearing aids, but instead, reflect the individual probe-microphone systems' implementation of the NAL-NL1 prescriptive method.
So whose fitting is correct? What does “correct” mean? Is there a “correct?” At the moment we aren't sure, but we are conducting additional testing in an attempt to determine the underlying factors for these differences.
However, for now it is clear that real and potentially clinically significant differences across systems exist. While these differences appear to be generally small (less than 5 dB) at most frequencies, intersystem differences of 5-10 dB appear at 3000 and 4000 Hz. We speculate that the magnitude of this difference could lead a patient fitted with these different systems to have differing comments related to how “bright,” “sharp,” “clear,” or “harsh” signals may sound. Consequently, until further evidence is available, we would encourage clinicians completing probe-microphone measures to consider adjusting these frequencies (3000-4000 Hz) separately when comments or complaints related to high-frequency sounds are made. Or, in some cases, a cross-check using REIG verification might provide helpful information.
We strongly support the use of probe-microphone measures as part of the verification process of hearing aid fittings, and we believe that probe-mic results can help differentiate a good fitting from a bad one. The grading of the fitting, however, should not be influenced by the equipment that is used to conduct the testing.
BTE: Behind the ear
DSL: Desired Sensation Level
MLE: Microphone location effect
NAL-NL: National Acoustic Laboratories-Non Linear
NAL-R: National Acoustic Laboratories-Revised
POGO: Prescription of gain and output
REAG: Real-ear aided gain
REAR: Real-ear aided response
RECD: Real-ear-to-coupler difference
REIG: Real-ear insertion gain
RETSPL: Reference equivalent threshold in SPL
REUG: Real-ear unaided gain
REUR: Real-ear unaided response
Hawkins and Mueller describe three types of test-retest for probe-mic measures: immediate (repeat testing at the same patient visit without removing the probe tube or the hearing aid), short-term (repeat testing at the same patient visit, but with both the hearing aid and probe tube removed and then replaced), and long-term (repeat testing at a later patient visit).7
Previous research has shown relatively good short-term test-retest for probe-mic measures. For example, Hawkins et al. reported standard deviations ranging from 1.4 to 2.7 dB for key frequencies between 1000 and 4000 Hz, for the REAR measure.11 These data, however, were collected using the modified-pressure concurrent equalization method—that is, the at-ear regulating microphone was active during testing.
One of our test conditions was the comparative evaluation of an open fitting. When evaluating open fittings, it is necessary to use stored equalization—the reference microphone is inactive during the test condition.12 This has the potential to reduce reliability, as minor head movement during the test period is not accounted for. There are few data available regarding the test-retest reliability for modern equipment when stored equalization is used.
As part of our protocol, each participant was tested twice (hearing aid and probe tube removed and replaced) for each of the three different hearing aid settings, allowing for the calculation of test-retest for each probe-mic system. Our results showed standard deviations ranging from 0.7 to 2.9 dB for the Verifit, 0.5 to 2.5 dB for the Fonix, and 0.7 to 2.9 dB for the MedRx for the range of 1000 to 4000 Hz. For all systems, the greatest variance was for 4000 Hz.
In our round-robin testing, for all three systems the test-retest standard deviations that we obtained for the stored equalization conditions were essentially the same as those for concurrent equalization. And, in agreement, our values are very similar to those reported by Hawkins et al.11 when concurrent equalization was employed. It appears, therefore, that if the patient (“participant,” in our case) is instructed not to move his or her head during testing, probe-mic measures will be quite reliable even when stored equalization is used.
© 2009 Lippincott Williams & Wilkins, Inc.