What'S new from NAL in hearing aid prescriptions?
2006; Lippincott Williams & Wilkins; Volume: 59; Issue: 10 Linguagem: Inglês
10.1097/01.hj.0000286003.14036.36
ISSN2333-6218
Autores Tópico(s)Speech and Audio Processing
Resumo1 Let'S me start with a basic question: What are you attempting to achieve with your prescriptive fitting? We see a hearing aid fitting prescription as a way of pulling together, in a very concrete and practical manner, the conclusions and implications of lots of research. I could answer your question on one level by saying that we try to make it as easy as possible for the clinician to get each hearing aid adjusted as right as possible as quickly as possible. The formulas we come up with attempt to summarize and quantify our best understanding of the amplification requirements of people with hearing loss.Figure: Dillon.On a more technical and specific level, the new non-linear prescription that we are now working on, called NAL-NL2, aims to maximize predicted speech intelligibility while making overall loudness sound normal. 2 I heard that an “NL2” version was being birthed, but your answer to my question sounds the same as what you gave me 7 years ago when I asked about NAL-NL1. Why the new version? Well, as you might know, since our first NAL version was developed in the 1970s by Denis Byrne and Will Tonnisson, there has been a continued evolution. The NAL prescriptions have always been theoretically calculated procedures, with the calculations based on experimentally measured characteristics of hearing-impaired people and of the speech signal. If new knowledge about the underlying characteristics comes along, which it has, the numbers going into our derivation process change, and so does the formula. 3 Makes sense, but what has changed in the things you mix into the brew out of which the procedure comes? I like the way you put that. The most fundamental set of empirical data is the ability of hearing-impaired people to extract information from speech. In the years since we launched NAL-NL1, there has been much interesting research investigating the impact of dead regions in the cochlea on the ability to understand speech.1-3 It caused us to do a comprehensive study investigating the ability of a large group of subjects with a wide range of losses to understand speech filtered in different ways, in quiet and in noise, and at low and high sensation levels. That research has given us new data on the degree to which people are able to understand speech (when it is made audible by amplification) for different degrees of hearing loss. Most importantly, it also gives us new data showing how the inevitable loss of ability varies with frequency. These data give us the ability to estimate speech intelligibility for any particular hearing loss, at least for the average person with that loss. It is largely the expected contribution to speech intelligibility in a particular frequency region that determines the amount of gain that is optimal for that region. 4 So you're saying that a decreased ability to understand amplified speech in some particular frequency region affects the amount of gain we should apply in that region. Does a poor ability mean that we should give more gain, or less? Let'S first draw an analogy. Optimizing gain is like going into a candy store with a limited amount of money, and trying to maximize the taste sensation by spreading our budget across the best range of candies. We want to come out with a variety of tastes, but more of the best-tasting ones than the others. In our case, every extra decibel of gain at one frequency causes loudness to increase by some amount depending on the loss and the input level, and hence on how loud this region already was. These separate loudness contributions from each frequency region add together to give the total loudness. But if we exceed our total loudness budget, the hearing aid wearer will say the hearing aid is too loud, and turn it down, or maybe even cease wearing it completely. Consequently, we want to spend our gain/loudness budget where we can get the biggest contributions to intelligibility. If someone has a reduced ability to understand speech in one frequency region then that becomes a less attractive region in which to use up our precious loudness budget. Consequently, we will prescribe less gain for that frequency region than we would if the person were able to make better use of audibility in that frequency region. Unless the hearing loss is very great or the input level is very low, we will generally still try to achieve some audibility across the entire speech frequency range, as low levels of audibility make only low contributions to total loudness. 5 Does hearing loss ever get so great, and the ability to extract information so poor, that the NAL formula prescribes no gain at all? Definitely. Our earlier speech intelligibility results, measured with variously filtered bands of speech, indicated that for very severe (and profound) hearing losses in the high frequencies, there was such a small contribution to intelligibility that it was impossible to calculate an optimal gain at these frequencies. This can occur whether the hearing loss is flat or sloping. For this degree of hearing loss, it is often the case that perception of signals at these frequencies is occurring at places in the cochlea tuned to lower frequencies. This is referred to as off-frequency hearing, or, more simply, it is that there is a “dead region” in the high frequencies. Our new results also show that the ability to extract information decreases as hearing loss increases. 6 Does that mean I will need to do a separate test for dead regions before prescribing a hearing aid whenever my patients have a severe hearing loss at any frequency? There is no doubt that we learn more about the characteristics of remaining hearing if we so some supra-threshold test as well as measuring hearing thresholds. Good candidates include the TEN test4 and psychoacoustic tuning curves.5 However, it is not yet clear to us whether or not the extra information we gain is sufficient to justify the time needed to conduct this testing. It may well be, but we will certainly be bringing out the NAL-NL2 prescription in a form that will enable it to be used with just hearing thresholds. It is quite possible that we will have a supplementary formula that can be used when the clinician has tested for dead regions. Relative to the standard formula, the presence of dead regions will cause reduced (or no) gain in the dead region, whereas the absence of dead regions will cause increased gain. 7 I understand that there may be times when amplification in a given frequency region will not be productive, but it bothers me when your prescriptive method just shows an “empty space” where the prescribed gain should be. Can you do anything about that? Sometimes what seem like unimportant details to us researchers can become major problems to the people who take up our work, so there are a few lessons we learnt from our experience with NAL-NL1. In NAL-NL1, whenever the optimizer could not find any value of gain that contributed to speech intelligibility, it simply made no prescription at all for that combination of conditions. This happened most strongly for very low and very high frequencies, for greater degrees of loss, and for softer input levels. It'S very believable that under this combination of circumstances, there is no amount of gain that produces a better outcome than no gain at all, so our final formula did the honest thing and simply made no prescription for these combinations. Unfortunately, the clinician, or the manufacturer writing the fitting software, has to make the hearing aid do something, so we ended up with a bunch of adhoc rules to ensure that these unimportant conditions did not adversely affect the fitting (e.g., by providing so much gain that loudness was adversely increased, or providing so little gain that the gain was also reduced for other combinations of hearing loss, frequency, and input level where the gain was important). In NAL-NL2, we are using a mathematical technique that ensures the gain always rolls off smoothly to 0 dB for very, very low frequencies, for very, very high frequencies, for very low degrees of loss, and for very high input levels. That way, we can make a practical gain prescription for every combination of loss, frequency, and input level, even if the actual gain still makes no contribution to intelligibility for some particular combinations of these variables. The principle is to at least do no harm, and hopefully the fitting screen will be more visually acceptable to you. 8 Earlier you mentioned a “total loudness budget.” I presume the “budget” is that we should not exceed normal loudness. Is this unchanged from NAL-NL1? The concept is the same, but the numbers are different. We have several pieces of evidence, direct and indirect, suggesting that NAL-NL1 prescribes slightly too much gain. First, a review of experiments in which subjects indicated their preferred gain showed that, on average, subjects preferred a gain 4 dB lower than that prescribed by NAL-RP, the linear predecessor of NAL-NL1, and which has a very similar gain for mid-level inputs.6 Second, experiments using a master hearing aid that logged the acoustic environment and the subjects' preferred gain in each environment also indicated that the preferred gain was lower than the NAL-NL1 prescribed gain.7-9 The gain was too high at all input levels, but the discrepancy was greater for high input levels than for low input levels. We suspect that the root cause of the gain being slightly too high is that the loudness model we use in our calculations slightly underestimates loudness for people with hearing loss. Consequently, the optimizer thinks it can allow the gain to be higher than it really should if the total loudness is not to exceed that experienced by normal-hearing people. Higher gains nearly always increase predicted intelligibility, so if the optimizer thinks it can get away with increasing gain, it just about always does. Consequently, what you'll see in NAL-NL2 is a reduction of gain of about 4 dB at mid-input levels and probably a slightly bigger reduction at high input levels. 9 Is there anything that is not changing in the new method? Yes, quite a lot. We are not aware of any new evidence about binaural loudness summation, so the difference in gain between unilateral and bilateral fittings remains the same. That is, the difference varies with hearing loss (greatest for symmetrical losses) and with input level (greatest for high input levels) in just the same way. Second, the formula for maximum output level is not changing. Because the gain, particularly for high input levels, is coming down, limiting will control gain less often in NAL-NL2 prescriptions than in NAL-NL1. 10 What about gain for new versus experienced hearing aid wearers? This is an area where we have long had a different opinion from the vast majority of clinicians, and with manufacturers too, as many of them use acclimatization managers and adaptation managers in their software. I know that nearly all clinicians believe that new hearing aid wearers will not accept as much gain as experienced wearers with the same hearing loss. However, the difference in preferred gain between the two groups is so small that it is barely worth considering. A review of the literature indicated that, across a range of experiments, there was only a 2-dB difference between the gains preferred by the two groups.6 As this conclusion came from experiments that were mostly intended to investigate something else, we specifically designed an experiment to test whether preferred gain changes with experience. It is now nearly complete, and seems to be headed for exactly the same conclusion, at least for patients with mild or moderate hearing loss. 11 If new hearing aid wearers prefer much the same gain as experienced wearers, why do most clinicians believe their preferences are different, and why have many manufacturers included some sort of adaptation manager within their fitting software. I can only guess that it is the indirect result of prescriptive procedures prescribing too much gain. I have already mentioned that NAL-NL1 prescribes slightly too much, and other generic procedures prescribe more than it does. Perhaps the patient'S preference for less than the prescribed gain has been ascribed to the patient being inexperienced rather than the prescription procedure being wrong. If a new user says the hearing aid sounds too loud, and the clinician can solve the problem by using an “adaptation manager,” I can see how this success might reinforce the belief that experience strongly affects preferred gain. 12 Something that is around now that wasn't 7 years ago are open canal fittings. Don't you need a different prescription for these than for conventional closed fittings? There'S so much to say about that I don't know where to start, but the short answer is “NO.” First, open fittings are not fundamentally new. They are great, and the best fitting for many, but electroacoustically they are just a very large vent, and that option has been available to clinicians, and an important part of our prescription considerations, for at least 20 years. What makes open fittings more achievable more often now is the digital feedback-cancellation systems they employ. Second, there is no reasons why changing the size of the vent should change the ear canal SPL of speech that is optimal for a person at any frequency. In the low frequencies, this optimal gain is 0 dB for many people with mild to moderate loss.10 This is lucky, because with a large vent, it is exceedingly difficult to achieve an insertion gain of more than 0 dB at 250 Hz or more than about 5 dB at 500 Hz, no matter what you do with the controls on the hearing aid. Of course, changing the vent size (including using an open-ear fitting) does change the coupler gain (measured with no vent) needed to achieve any particular real-ear gain. Finally, even for hearing aids with feedback cancellation, we can end up having to compromise: Do we insist on avoiding the occlusion effect by having the fitting very open or do we insist on achieving the target gain in the high frequencies, because we often can't do both? The best fitting may well be one that has less high-frequency gain than the target. That'S not because the target should be different for open fittings. It'S just a technology-dependent limitation on our ability to achieve both goals. We have long had to accept such limitations, but the best trade-off may well change as the technology changes. 13 You mention achieving target, but does the actual gain on a hearing aid really matter? If it'S wrong won't people just change the volume control until it'S right? We use to think that the accuracy of the overall gain was less important than the accuracy of the balance across frequencies (that is, the gain-frequency response shape), precisely because people had volume controls that they expected to use in different situations. However, non-linear amplification (by which I mean compression, adaptive noise suppression, and different degrees of directionality in different settings) opens up the possibility of never having to use a volume control, as long as the hearing aid is smart enough to adapt to the conditions, and as long as we get the prescription of overall gain right. Clinicians would like to achieve this right from the start as often as possible, rather than rely on a trial-and-error process. And of course, there are many hearing aids fitted that don't employ a volume control. 14 Let'S talk about gain for high inputs. NL1 didn't use loudness discomfort measures for determining high-level gain. Are you sticking with that philosophy for NL2? Yes, we are, and here are the reasons why, though I admit the case is arguable. First, our direct experiment on the use of measured loudness discomfort level indicated that it contributed only very slightly to the accuracy with which we could prescribe the optimal maximum output settings for wideband compression limiting and wideband peak clipping. One reason for this unexpected finding is that there is no reason why the optimum maximum output level should be equal to the loudness discomfort level, though we certainly don't want the output to actually cause loudness discomfort. The second reason is that the loudness discomfort level measured on one occasion in the clinic with one specific stimulus may be a poor estimate of the level that people will happily tolerate for more complex sounds in real-life situations. Third, with wide dynamic range compression hearing aids in which the gain gradually lessens as the input level rises, there is greatly reduced need for limiting compared with linear hearing aids. The only reason for our doubt about basing limiting level solely on hearing thresholds is that frequency-specific loudness discomfort levels measured in the clinic probably give a useful guide to how limiting should vary across frequency in a multi-channel hearing aid. If a clinician wished to supplement the prescription by measuring discomfort level, that would be fine, though my guess is it would be a better use of clinical time to instead commence with the threshold-based prescription and then, immediately after fitting, empirically test that high-level sounds do not cause discomfort. 15 How about compression settings? Many manufacturers recommend a fairly low kneepoint, but, as I recall, NAL-NL1 usually calls for a kneepoint around 50 dB SPL or even higher? Any expected changes here? Setting the compression threshold is something we do right at the end of the process of deriving the formula, which we have not arrived at yet. If we simply aimed to maximize intelligibility while keeping loudness normal, we would apply compression right down to the level of the weakest sound in the softest level speech that anyone ever encounters, which is possibly around 30 dB SPL. This would have the advantage of allowing hearing aid wearers to hear virtually all the soft sounds around that normally hearing people hear, and the disadvantage that the gain prescribed at low levels would often cause the hearing aid to whistle whenever people went into a quiet place. The need to hear very soft sounds is arguable; if a person is having no trouble understanding those around them and associates each separate sound with a unique source, it seems likely that hearing all the everyday sounds around would add to their quality of life. On the other hand, some hearing aid wearers have trouble separating the many sounds they hear, so many situations seem like a confusing jumble of sounds. For them, restoring all these sounds to normal loudness may not be optimal. Given the lack of evidence about optimal compression thresholds, we still don't have a strong recommendation about where they should be set. As for NAL-Nl1, our tentative recommendation will be that speech with a wideband rms level of around 50 dB SPL should just be entering compression. This translates into lower thresholds within each channel, with the actual amount lower depending on the number and bandwidth of the channels. Many hearing aids will apply expansion at lower levels still, and this is certainly an effective way of ensuring that internal noise of the hearing aid is inaudible. 16 Staying with the compression theme, will you be adding anything re-lated to compression time constants? We will still not be making a prescription about compression speed. There is no doubt that compression is beneficial, and there is no doubt that this applies whether the compression is fast (compared with syllables) or slow (compared with words or sentences). My personal bias is that an adaptive compression speed, where the speed of compression depends on the dynamics and timing of the signal, captures the advantage of both fast and slow. However, we don't think that the speed of compression carries any significant implication for what the average level of speech in each frequency region should be. Consequently, the prescription will be usable for any speed. Obviously the dynamic range at the output of the hearing aid will depend on the compression speed chosen. 17 I fit a lot of pediatric patients. Are you head-ed toward the same prescription for children as for adults? The short answer is that we are headed toward a different prescription, but to explain the reasoning behind it requires a much longer answer. The easy part is that if you refer to a prescription expressed as a coupler gain, then of course the prescriptions will differ, as they always have. The smaller ear canals of children require us to use less coupler gain if we are to achieve the same real-ear aided gain as for adults. The much harder issue is whether or not a child and an adult require the same real-ear aided gain if both have the same hearing loss from the eardrum inwards. We can approach this question from more than one perspective. One common argument is that because children are still acquiring language, they are less able to use context. Therefore, they require more acoustic information than adults to understand speech and therefore require a different gain-frequency response. While the premise is certain, I don't believe the conclusion necessarily follows. Just like children, adults wearing hearing aids frequently struggle to understand the conversation, despite their knowledge of language, and so they also want the gain-frequency response that maximizes speech intelligibility. However, there is some evidence that loudness preferences of the two groups differ, so if we follow the principle I gave in the answer to your very first question, we do end up with a higher gain for children. 18 How did you reach this conclusion? There are now three experiments where we have compared NAL-NL1 to DSL[i/o], and the first two of these gave us some insights into differences between the two types of listeners. In the first experiment, school-aged children compared the two procedures in a crossover design, in both Australia and Canada. Although there were marked differences in average gain between the two prescriptions, children indicated very small loudness differences between the two procedures after they had worn each for 2 months, indicating a marked flexibility in their perception of what was most comfortable for them. The average preferred gain was between the two prescriptions, and therefore was a higher gain than prescribed by NAL-NL1. In the second experiment, adults were prescribed with both responses in a crossover design. There was unanimous preference for the NAL-NL1 response, driven strongly by DSL[i/o] prescription being judged as too loud. In the third experiment, infants were randomly assigned to receive either NAL-NL1 or DSL[i/o], right from their first fitting. We have only preliminary data, but are so far seeing no difference in performance between the two prescriptions. It'S not that the measures are insensitive, because we do have significant differences between those who received their hearing aids before 6 months of age relative to those who received them later. Taken together, it seems that congenitally impaired children prefer more gain than adventitiously impaired adults, though the differences in gain do not appear to result in any differences in performance. Theoretically, an increased gain is most likely to lead to improved intelligibility at low input levels, so the child and adult versions of NAL-NL2 will be most different at low input levels. At these levels, there are no adverse safety implications from using a higher gain. 19 A quick question concerning verification: Is there a continuing role for real-ear measurement to ensure that the prescription has been achieved? Yes, though there are changes coming that might do away with the need for real-ear measurements in some cases. First, I think that manufacturers are getting better at making software automatically set up the hearing aid to achieve a reasonable approximation of the prescription, though individual ear differences will never allow this to be a perfect process. Second, trainable hearing aids will be the best solution for some clients. With a fully trainable hearing aid, the user can train the device to have the gain, frequency response, compression ratio, and compression thresholds that the individual prefers.9,11 For such devices, there seems little point in slavishly matching a prescription if one knows that the client is about to train the device, potentially away from the prescribed response, which at best is optimal only for the average client. So, for a trainable device that has its initial response automatically set to a reasonable approximation of the prescription, I think the time needed for real-ear measurements could be better spent talking to the client. 20 Finally, you have talked about several issues that affect the new NAL-NL2 hearing aid prescription. But, is it available? Not yet, but soon. We don't actually have the new formula yet, and when we do, it will have to go through a cross-checking process to ensure that the prescriptions are consistent with the changes that the accumulated evidence suggest are necessary. We are keen to get it out as quickly as possible, so that clinicians have easy access to the latest knowledge in this area. Back in 1976, a paper from Australia was published in Scandinavian Audiology describing a new prescriptive fitting method for hearing aids. While the paper was highly regarded, the method itself had little clinical impact in the U.S. At the time, many practitioners still believed it was necessary to use speech audiometry to select the best hearing aid, and the few who did favor prescriptive fittings quickly latched onto the Berger method, also introduced about this time. When the revised version of the Australian method, the NAL-R, came out 10 years later, the U.S. climate was quite different and the cleverly designed multi-colored slide rules started showing up in clinics everywhere. Soon, the NAL-R method was included on probe-mic equipment, and the rest is history. But hearing aids change, and so does people'S thinking about fitting them. It'S only natural, then, that prescriptive fitting methods must also evolve. Now, 30 years after the launch of the original NAL, that'S what this month'S Page Ten is about. Our author is Harvey Dillon, PhD, director of research at the National Acou-stic Laboratories (NAL) of Australia. Dr. Dillon is internationally known in the areas of acoustics, psychoacoustics, and hearing aids, and his book, Hearing Aids, has become the standard hearing aid text for students and practicing professionals. Harvey'S research interests are diverse, ranging from electrophysiology to APD in children to hearing protection. This month, however, he'S here to talk about what he is known for best: selecting and fitting hearing aids, and, specifically, the birth of the new NAL-NL2 prescriptive fitting method. In April 1999, my introductory comments for Page Ten went something like this: “If you believe the television commercials, ‘Foster'S is Australian for beer.’ A rather debatable statement if you regularly sample the barley beverages from down under. What'S not debatable is that ‘the NAL is Australian for good hearing aid research.’ This definition is known from Beijing to Bitburger to Bud.” Those comments referred to a Page Ten written by none other than Harvey Dillon, in which he described the new NAL-NL1 fitting procedure—an article that quickly became one of the most referenced in the 60-year history of the Journal. While those Foster'S commercials seem to have faded over the past 7 years, it'S good to know that Harvey and the folks from the NAL just keep making things better, providing us with a contemporary hand-crafted brew for selecting and fitting hearing aids. Gus Mueller Page Ten Editor
Referência(s)