Evidence on Self-Fitting Hearing Aids

Abstract

The research on self-fitting hearing aids is reviewed using evidence-based principles. The evaluation begins with a definition of the research questions followed by a detailed search of the literature and then a review of the relevant studies. Four features of self-fitting hearing aids are reviewed: in-situ threshold measurement, whether an initial fitting prescribed using standard prescription formulae will approximate user preferences, outcomes with training of hearing aids for preferred responses, and assembly and use of the aids. There is at least good quality evidence suggesting that in-situ thresholds can be reliably obtained, that prescribed initial fittings approximate preferred responses, and that users are able to train the hearing aids and would prefer the trained responses. However, evidence on other outcomes and the ability of users to assemble and use such instruments is limited. Gaps in research with self-fitting hearing aids are identified.

Introduction

Self-fitting hearing aids are a relatively new concept and should be distinguished from trainable hearing aids. While self-fitting amplification could encompass a trainable hearing aid so that the user would train the hearing instrument to preferred settings, the whole fitting process of a self-fitting hearing aid is conducted without the direct supervision of a professional. This would be ideal for those who have difficulty accessing hearing health care professional services (e.g., clients who have physical disability or live in remote areas and third world countries, where professional hearing health care services are not readily available). Although a fully self-fitted hearing aid is not yet commercially available, various features of this type of aid have been the focus of some recent research. To examine evidence on self-fitting hearing aids, this article will start with a description of the main features of this type of hearing aid, followed by a literature search for evidence and a discussion of the evidence related to these features.

What Is a Self-Fitting Hearing Aid?

A self-fitting hearing aid should exhibit four features. First, it would allow automated evaluation of hearing thresholds to yield an initial hearing aid response. One way to fulfill this is by using an automated function to measure in-situ hearing thresholds and apply the information to derive a prescription based on a conventional fitting method such as the NAL-NL1 (Dillon, 1999). The assumption is that the initial fitting would approximate user preferences and serve as the starting point for the user to train the hearing aids (second feature).

The third feature involves training of the self-fitting instrument to meet the listening needs of a user without the assistance of a professional. This feature is related to the development of a “trainable” hearing aid, as described by Dillon et al. (2006). Hearing aid users would start with a prescriptive fitting. The users would then train the hearing aids by making changes to the parameters to their preferred settings in various real-life listening situations. During the process, the hearing aids would identify important acoustic characteristics of the environments and the corresponding preferred settings selected by the user. This is a training process because the hearing aids would continue to accumulate the data and make refinement to the predictions about the preferred settings in specific environments, at least for a certain period of time. It is anticipated that with sufficient training, the device would be able to automatically adapt to subsequent listening environments to maximize self-perceived benefit.

The fourth feature would allow physical fitting and use of the aid without the supervision of a professional. A set of instructions and self-fit kit could be used for the user or the caregiver to assemble the aid and its accessories (e.g., self-fit tips and tubing in various sizes, batteries) and for the user to wear the instrument in daily situations.

While the concept of self-fitting hearing aids is attractive and sensible, the issues related to these features are (1) whether automated in-situ audiometry would result in accurate measurement and thus data could be accurately applied to achieve an initial fitting; (2) whether parameters obtained using conventional prescription methods are close to individual preferred initial fitting settings; (3) whether there is evidence to show that indeed users are able to train the hearing aids to achieve good outcomes, compared to untrained responses; and (4) whether users could successfully follow instructions to achieve the best physical fit and use the hearing aids. The literature that addresses these issues is reviewed below.

Defining Research Questions and How Evidence Will Be Evaluated

According to evidence-based practice principles, the first step is to formulate a PICO question that would govern the search of literature (Wong & Hickson, 2012). This question should address the Problem/Person (P), Intervention (I), Comparison intervention (C), and desired Outcomes (O) in an analytic study. In a descriptive study, there may not be a comparative intervention. To help focus the search, four PICO questions were formulated to examine the literature on self-fitting hearing aid feature in adult users with sensorineural hearing impairment (Table 1). Internal validity of each study was first evaluated to determine whether the findings from the studies are trustworthy and were used to evaluate the evidence related to each PICO question. Each study that exhibited good internal validity was assigned a level of evidence. The Oxford 2011 Levels of Evidence for diagnostic tests (The OCBEM Level of Evidence Working Group, 2011) was used to evaluate the overall quality of research evidence for in-situ audiometry (PICO Question No. 1). Levels of evidence are used to indicate whether the evidence has yielded consistent results and how the outcomes obtained using an assessment tool compared to a well-recognized standard reference.

Table 1.

The Four PICO Questions Used to Guide the Search of Evidence

1. Among adults tested for hearing thresholds (P), would in-situ audiometry (I), compared to conventional audiometry (C), yield at least as accurate audiometric information for initial fitting of hearing aids (O)?

2. For adults with hearing impairment (P), would initial prescriptive fitting (I) approximate user preferred response (O)?

3. Could hearing aid users (P) train hearing aids (I) to yield better outcomes (O) compared to untrained responses (C)?

4. Could hearing aid users (P) follow written or illustrated instructions (I) to manage the assembly, physical fit and use of a self-fitting hearing aid (O)?

Note: The letters in brackets indicate the PICO elements: P refers to the problem or person, I refers to the intervention condition, C refers to the Comparison Condition, and O refers to the desired outcomes.

For evidence related to other PICO questions, each study was assigned a level of evidence, based on the type of study. The level of evidence of the studies will be assigned using the criteria stated in Cox (2005) and Wong and Hickson (2012). There are six levels of evidence: Level 1 refers to systematic reviews and meta-analysis that are of high-level studies; Level 2 studies consist of well-designed randomized control trials; Level 3 studies involve intervention conditions that are not randomized; Level 4 studies do not consist of an implementation of an intervention; Level 5 research refer to case studies; and Level 6 evidence comprise comments from experts only. An overall grade of recommendation (Cox, 2005; Wong & Hickson, 2012) will then be assigned based on the level of studies evaluated and whether there are consistent findings across studies. A Grade “A” recommendation refers to evidence that is of Level 1 and Level 2 and is consistent across studies. A Grade “B” recommendation results when studies are primarily of Level 3 and Level 4 or when evidence is based on higher level studies that are not directly addressing the PICO question. A Grade “C” recommendation is used to describe consistent results from Level 5 studies or generalizations are made from Level 3 and Level 4 studies that are not directly relevant to the PICO question. A Grade “D” recommendation refers to Level 6 studies, situations where most studies do not provide consistent evidence in relation to the PICO question, or situations where there is a high likelihood of bias.

Search of the Literature

To identify studies related to the three features of self-fitting hearing aids, the following keywords were used: “hearing aid instructions,” “in-situ audiometry,” “in-situ thresholds,” “self-adjustment hearing aids,” “self-adjusted hearing aids,” “self-adjusting hearing aids,” “self-adjustable hearing aids,” “self-fitted hearing aids,” “self-fitting hearing aids,” “trainable hearing aids,” and “trainability.”

Databases searched included PubMed, Medline, Google Scholar, PsychInfo, and DisCom. Ideally, the search for evidence should only include those that provide the highest level of evidence (e.g., systematic review, randomized controlled trials). Because there are not many articles in this area, articles from trade and online journals such as Audiology Online and Hearing Review, were included initially in the search. Reference lists of all articles were also reviewed to identify more relevant evidence. The abstracts of these studies were reviewed for relevance and those articles that did not allow evaluation of quality of research were eliminated. Such articles did not provide sufficient information on subject demographics, methodology, and statistical analyses. Research studies that did not possess good internal validity (Wong & Hickson, 2012) were not included. The search was performed in May, 2011.

• PICO Question No. 1: Among adults tested for hearing thresholds, would in-situ audiometry, compared to conventional audiometry, yield at least as accurate audiometric information for initial fitting of hearing aids?

For the purpose of hearing aid fitting only, in-situ measurement is most efficiently conducted using the hearing aid itself. Assuming that the hearing aid transducer is appropriately calibrated, in-situ audiometry would eliminate the need to use a conventional audiometer and the time to input threshold data in the hearing aid fitting program. In-situ audiometry, if properly applied, would account for variations in residual ear canal volume, eardrum impedance, and other characteristics of the hearing aid fitting (e.g., venting). In-situ thresholds, however, are not equivalent to thresholds obtained using conventional audiometers; real-ear-to-dial correction factors must be applied for comparison.

There was no study on automated in-situ audiometry and three peer-reviewed studies (DiGiovanni & Pratt, 2010; O’Brien, Keidser, Yeend, Hartley, & Dillon, 2010; Smith-Olinde, Nicholson, Chivers, & Highley, 2006) were reviewed to examine reliability and validity of in-situ audiometry measured by a clinician. These studies are listed in Table 2; they compared in-situ thresholds to those obtained using conventional audiometry as a reference standard and therefore are considered as Level 2 evidence using the Oxford 2011 Levels of Evidence for diagnostic tests. That is, these are cross-sectional studies that had applied reference standards to examine whether results obtained are valid. Overall, these studies showed that in-situ thresholds could be reliably obtained with ER-3A earphones or instant-fit eartips and commensurate with thresholds measured using conventional audiometry. Precautions are discussed at the end of this section to ensure that accurate in-situ thresholds could be obtained.

Table 2.

Types of Evidence Provided by Studies Reviewed

	PICO No. 1	PICO No. 2			PICO No. 3				PICO No. 4
Author(s)/Year	In-situ Thresholds Are Reliablea	Initial Prescriptions Slightly Higher Than Preferred Gain	Preferred Gain & Frequency Response Vary With Situations	New Users Prefer Less Gain	Users Able to Discriminate Responses	Initial Responses Affect Preferred Response Selection	Trained Responses Result in Positive Outcomes	Algorithms Could Classify Sounds	Users Could Assemble HA
Alexandre et al. (2006)								Yes
Bentler et al. (1993)				No
Chalupper (2006)								Yes
Chalupper et al. (2009)								Yes
Convery et al. (2005)			Yes	No
Convery et al. (2011)									Yes
Cox and Alexander (1992)		Yes
DiGiovanni and Pratt (2010)	Yes
Dijkstra et al. (2007)								Yes
Dreschler et al. (2008)			Yes		Yes	Yes	Yes
Hornsby and Mueller (2008)		Yes
Horwitz and Turner (1997)		Yes
Humes et al. (2002)		Yes
Keidser et al. (2005)			Yes
Keidser, Dillon, et al. (2008)			Yes		Yes	Yes
Keidser, O’Brien, et al. (2008)		Yes		Yesb
Lamarche et al. (2010)								Yes
Marriage et al (2004)		Yes		Yes
Mueller and Bentler (2005)		Yes
Mueller (2005)
Mueller et al. (2008)			Yes		Yes	Yes	Yes
O’Brien et al. (2010)	Yes
Polonenko et al. (2010)		Yes
Rahal (2010)
Smeds et al. (2006)			Yes	No
Smith-Olinde et al. (2006)	Yes
Zakis et al. (2007)					Yes		Yes	Yes

Note: The PICO questions are listed in Table 1.

^aIn situ thresholds are accurate measures if individual REDD corrections are applied, the transducer is properly calibrated, and ambient noise is controlled.

^bNew users with more than a mild impairment prefer more gain reduction from prescribed settings, compared to experienced users; but overall, there was no significant difference across users with all levels of impairment.

These studies reported results on participants with normal or mild to severe hearing impairment. O’Brien et al. (2010) evaluated in-situ thresholds in listeners with mild to moderately severe sloping hearing impairment, while participants in Smith-Olinde et al. (2006) exhibited normal hearing. In DiGiovanni and Pratt (2010), half of the participants were normal-hearing listeners and the other half had mild to severe sloping sensorineural hearing impairment. ER-3A insert earphones were used to obtain in-situ thresholds in two studies (DiGiovanni & Pratt, 2010; Smith-Olinde et al., 2006) and instant-fit eartips were used in O’Brien et al. (2010). Calibration of in-situ test stimuli was reported by O’Brien et al. (2010) and Smith-Olinde et al. (2006). All three studies reported that insertion depth of the transducers was carefully controlled across participants. Real-ear-to-dial differences (REDDs) were measured in the O’Brien et al. (2010) study to convert thresholds measured to dB SPL at the eardrum.

When the American Speech-Language-Hearing Association (1978) recommended procedures for pure-tone audiometric testing and a 5-dB step size was used, Smith-Olinde et al. (2006) found test-retest reliability was within ± 5 dB in more than 93% of the in-situ thresholds obtained at 500, 1000, 2000, and 4000 Hz. In-situ thresholds (in dB HL) were within ± 10 dB of thresholds measured using conventional audiometry. DiGiovanni and Pratt (2010) found that in-situ thresholds (in dB HL) at 500, 1000, and 2000 Hz were about 10 dB worse in listeners with hearing impairment, compared to thresholds obtained using conventional audiometry. Interestingly, the differences in thresholds obtained in normal-hearing listeners were much smaller and it was unclear from the article what factors could have affected these results and whether these between-group differences were significant. Because calibration of signals was not reported in the study, whether this factor affected the results could not be verified.

Low-frequency leakage may occur with a poorly fitted earmold, an earmold with a vent, and from open domes in particular. Leakage of low-frequency sounds from the ear tip could cause elevation of thresholds. ER-3A insert earphones, with carefully controlled depth of insertion as reported above, probably provide the most consistent in-situ measurement. Self-fitting hearing aids are expected to be fitted without the assistance of a clinician and are more likely to be fitted with an instant-fit closed or open dome and therefore low-frequency leakage would be a concern. O’Brien et al. (2010) reported that while the in-situ thresholds could be reliably measured with the open-fit tip, some degrees of test-retest variability were observed with closed domes, because of inconsistent seal around the ear tips. However, the variability was no worse than the variability measured under headphones with conventional audiometry, except at 250 Hz, and variability was in fact much better at high frequencies. In their study, they also found that for about one third of hearing aid users, the effects of low-frequency leakage could cause prescribed gain from uncorrected in-situ thresholds (i.e., in dB HL) to differ by more than 3 dB RMS across frequencies from the targets based on conventionally measured thresholds. The gain prescribed from the uncorrected in-situ thresholds could be perceptually different from gain prescribed from the corrected thresholds (Keidser, Yeend, O’Brien, & Hartley, 2011). However, when real-ear-to-dial differences (REDDs) related to transducer and hearing aid coupling were used to convert threshold levels measured in dB HL to dB SPL at the eardrum, accurate estimation of hearing thresholds could be achieved. These REDD corrections were about 10 to 29 dB at 250 to 1500 Hz for the open instant-fit tip, and were reduced with the closed instant-fit tip (9 and 7 dB at 250 and 500 Hz, respectively). At other frequencies, in-situ thresholds were found to be within 10 dB of thresholds measured using conventional methods, which is consistent with reports from other authors (e.g., Smith-Olinde et al., 2006) using ER-3A coupled to hearing aids. The O’Brien et al. study also pointed out that noise levels must be controlled during the test, to reduce the effect of ambient noise on lower level threshold measurements.

Conclusion: There is Level 2 quality evidence (The OCBEM Level of Evidence Working Group) that shows test-retest reliability of in-situ thresholds measured using standard audiometric techniques will not differ significantly from those obtained using conventional audiometry (DiGiovanni & Pratt, 2010; O’Brien et al., 2010; Smith-Olinde et al., 2006). In other words, the evidence presented in this article is based on studies that have compared in-situ threshold measurement to a recognized reference standard and consistent results have been noted. However, when validity of measurement with instant-fit ear tips is considered, the evidence is that accurate in-situ thresholds could be obtained when the transducer is properly calibrated, ambient noise is controlled, and individual REDD corrections are applied (Keidser et al., 2011). Whether average REDDs would achieve the same accuracy should be evaluated. Further research is also needed if in-situ audiometry is to be applied in situations where ambient noise level cannot be controlled or when automated in-situ audiometry is used. In-situ audiometry, at present, is constrained to the measurement of air conduction thresholds (Keidser et al., 2011) and has not been evaluated in listeners with asymmetrical hearing impairment where masking may be necessary (O’Brien et al., 2010). This evidence is limited to adults with up to a severe hearing impairment.

• PICO Question No. 2: For adults with hearing impairment, would initial prescriptive fitting approximate user preferred response?

For self-fitting hearing aids, it is important that clients are being prescribed an initial response that approximates user preference, before they would go out to real-life situations to train the hearing aids. This is not only to ensure comfort and good speech intelligibility but we will also learn in the next section that initial response could bias the training process. The literature was reviewed to learn whether prescriptive methods such as the NAL-NL1 or various versions of the Desired Sensation Levels (DSL) are able to achieve an initial fitting that approximates preferred responses. Key evidence includes systematic reviews, randomized controlled trials, and simultaneous crossover designs (Convery, Keidser, & Dillon, 2005; Keidser & Dillon, 2007; Marriage, Moore, & Alcantara, 2004; Mueller, 2005; Mueller & Bentler, 2005). The evidence is of Levels 1 and 2 (Cox, 2005; Wong & Hickson, 2012a). Other studies are of Levels 3 to 4 evidence. While most research was conducted with conventional digital hearing aids, trainable hearing aids were used in a few studies. In general, overall prescribed gains have been found to be slightly higher than average preferred responses. There are substantial variations in individually preferred responses, and preferred frequency responses also vary depending on the listening situations.

Various prescription methods result in prescription targets that differ from individually preferred gain by up to 10 dB or more, particularly at the high frequencies (Hornsby & Mueller, 2008; Keidser & Dillon, 2007; Keidser, Dillon, & Convery, 2008; Keidser et al., 2005; Keidser, O’Brien, Carter, McLelland, & Yeend, 2008; Mueller, 2005; Polonenko, Scollie, & Moodie, 2010; Smeds et al., 2006; Zakis, Dillon, & McDermott, 2007). Other studies have reported that mean preferred overall gain is about 3 to 4 dB less than the prescriptive targets (Hornsby & Mueller, 2008; Horwitz & Turner, 1997; Humes, Wilson, Barlow, & Garner, 2002; Keidser, O’Brien, et al., 2008; Marriage et al., 2004; Mueller & Bentler, 2005; Polonenko et al., 2010; Zakis et al., 2007).

Keidser and Dillon (2007) reviewed data from five research studies conducted at the National Acoustic Laboratories and found that 49% of hearing aid users’ preferred gain was within ± 3 dB of the NAL-NL1 prescribed targets for an input of 65 dB SPL. They also reported 46% of users preferred gain to be more than 3 dB below NAL-NL1 prescribed targets and a very small proportion preferred gain to be more than 3 dB above targets. Another study by Keidser, O’Brien, et al. (2008) showed that mean preferred gain was 3.9 dB less than prescribed gain. Similarly, Smeds et al. (2006) found that in a laboratory setting where loudness was rated, listeners preferred loudness less than NAL-NL1 calculated loudness. Hornsby and Mueller (2008) reported an average preferred gain of about 1 to 2 dB less than the NAL-NL1 prescribed gain and the difference was not statistically different. Preferred gain settings were adjusted to optimize listening comfort and clarity of speech presented at 65 dBA in quiet.

When the DSL v5.0a adult algorithm was used, the prescribed settings were found to deviate from preferred listening levels (PLLs) by an average of 2.6 dB across the frequencies from 250 to 4000 Hz (Polonenko et al., 2010). These PLLs were obtained in hearing aid users listening to running speech at an overall level of 60 dBA in a sound-treated room. The 95% confidence interval of these deviations ranged from 5.8 to 8.4 dB. However, individual hearing users’ PLLs were as much as 11 dB below DSL v4.1 prescribed targets.

With common use of compression hearing aids, changes in gain settings will result in variations of other compression parameters when the changes are input-level dependent. For example, Zakis et al. (2007) found that responses trained in real-life situations yielded higher compression ratios (CRs) and lower gain than NAL-NL1 prescription at typical speech levels as relatively lower gain was preferred for high than for low input levels. Depending on the frequency channel, mean trained gains were reduced by up to about 5 dB, compared to untrained gains when the inputs were at 65 dB SPL or at 80 dB SPL. Mean trained CR was greater than the untrained CR in the three channels. Mean trained noise suppression strength was significantly different from the untrained setting in the middle frequency channel.

Preferred gain and frequency response vary with listening conditions (Keidser et al., 2005; Mueller, Hornsby, & Weber, 2008). In a laboratory experiment, Keidser et al. (2005) found that less gain is preferred in frequency bands where noise is more annoying and the preferred frequency response slopes depend on the slope of the audiogram. Convery et al. (2005), Keidser, Dillon, et al. (2008) and Smeds et al. (2006) also found that the preferred gain-frequency response varied with listening conditions. Dreschler, Keidser, Convery, and Dillon (2008) found overall gain and response slopes across six test conditions, including speech in quiet and in speech noise and nonspeech noise, varied significantly from NAL-RP settings. They found preferred gain to be lower for louder sounds or in conditions where only noise was present. Flatter frequency slopes were selected when relatively greater energy is present at mid and high frequencies, compared to conditions with more low-frequency energy. Using trainable hearing aids, Mueller et al. (2008) also found that preferred gain in various real-world situations could differ significantly from prescribed NAL-NL1 targets.

There is a conventional belief that new users should be prescribed with lesser gain (Mueller & Powers, 2001). However, the majority of current evidence does not reveal a significant difference in preferred gain between new and experienced users (e.g., Bentler, Niebuhr, Getta, & Anderson, 1993; Convery et al., 2005; Keidser, O’Brien, et al., 2008; Smeds et al., 2006). For example, Keidser, O’Brien, et al. (2008) found a nonsignificant difference of 2.7 dB, while Convery et al. (2005) found a smaller and also nonsignificant difference of 2 dB. On the contrary, Marriage et al. (2004) reported a 2.6 dB significant difference in preferred gain between new and experienced users. Despite a lack of significant difference in average preferred gain between new and experienced users, Keidser, O’Brien, et al. (2008) reported that new users with more than a mild hearing impairment tend to prefer more gain reduction from the target settings, compared to experienced users with the same amount of hearing impairment. These new users seem to require more than 1 year to accept prescribed overall gain. Therefore, further study is required to examine whether new users with greater impairment would require less overall gain than prescribed. This issue should be considered in fitting trainable hearing aids because as discussed below, initial settings could bias the users in the selection of preferred gain.

Conclusion: Overall, relatively consistent findings across Level 1 and 2 studies yielded an overall Grade “A” recommendation. The evidence showed that on average, prescriptive fittings yield initial responses that are about 3 to 4 dB higher than user preferences (Hornsby & Mueller, 2008; Horwitz & Turner, 1997; Humes et al., 2002; Keidser, O’Brien, et al., 2008; Marriage et al., 2004; Mueller & Bentler, 2005; Polonenko et al., 2010; Zakis et al., 2007). Consequently, prescribed gain for both the DSL and NAL-NL methods have been reduced in their latest version (Polonenko et al., 2010). However, there could be significant individual variations and preferred responses may vary with listening situations. Therefore, fine-tuning or training of hearing aids is needed to ensure frequency responses meet individual users’ requirements.

• PICO Question No. 3: Could hearing aid users manage training and reliably select their preferred responses?

For the training phase to be successful, clients must be able to understand the training procedures, manipulate the control(s) well enough and within a relatively short time, adjusting the aid to provide appropriate parameter(s) for the particular type of listening situation in hand. With training, the aid would analyze the data and perform some form of auditory scene analysis to learn what the user prefers in various listening situations. Whether the hearing aid algorithm is able to derive preferred settings to suit a situation would depend on how consistently the user is able to train the aid and the design of algorithms; that is, how robust the relationship is between the manipulated amplification parameters and the analysis of the acoustic environment.

Could hearing aid users manage training and reliably select their preferred responses?

There are four studies that showed adult hearing aid users with sensorineural hearing impairment are able to perform training tasks with relative ease and that training yields preferred responses (Dreschler et al., 2008; Keidser, Dillon, et al., 2008; Mueller et al., 2008; Zakis et al., 2007). The latter two studies included take-home trials and therefore provide information on whether hearing aid users are able to manipulate the adjustment controls in real-life situations. Table 2 lists these studies and the types of evidence these studies were reviewed in this article.

This evidence is mostly quasi-experimental studies, with a single sample pre-test/post-test design, which were conducted in laboratory conditions. The search did not reveal any study that is of higher level (e.g., randomized controlled trials). If the classification of level of studies from Cox (2005) and Wong and Hickson (2012) is used, these are regarded as Level 3 and 4 studies. In addition, there was very little information on verification and validation of the initial hearing aid settings. While Zakis et al. (2007) reported on adjustment of hearing aids according to prescribed real-ear insertion gains (REIGs) prior to training, the majority of the studies did not report how well the REIGs match the targets and the outcomes with these initial settings. In the Mueller et al. (2008) study, participants did not start from prescribed REIGs, but from responses that provided 6 dB more or less overall gain than prescribed.

Across the reviewed articles, participants exhibited symmetrical mild to moderate or mild to moderately severe sloping hearing impairment and varied in age; most participants were experienced hearing aid users (e.g., Dreschler et al., 2008; Keidser, Dillon, et al., 2008; Mueller et al., 2008). While most studies were conducted in laboratory settings, participants in two peer-reviewed studies (Mueller et al., 2008; Zakis et al., 2007) trialed the hearing aids in real-life situations. The majority of the studies had no more than 30 participants and power analysis was not conducted to determine whether the sample size was sufficient.

In a laboratory experiment, Dreschler et al. (2008) found that, using four different control configurations, gain-frequency responses were reproducible and a test-retest standard deviation of less than 3 dB was reported. Most participants were reliable at discriminating these responses and a small proportion (2 out of the 22 participants) was less reliable. Hearing aid adjustment seemed quite easy particularly with steps that result in more noticeable changes. Therefore, Dreschler et al. concluded that users could reliably use different controls to make gain-frequency adjustments.

Keidser, Dillon, et al. (2008) also reported that participants were able to discriminate hearing aid responses. They found that 83% of participants were able to discriminate between responses that deviate somewhat from the baseline response, but only 25% of participants reliably preferred some responses over others. In a field trial, Mueller et al. (2008) also found that hearing aid users were able to train their hearing aids to preferred settings to improve speech perception as well as satisfaction with aided loudness. In another take-home trial, Zakis et al. (2007) found that users were able to manipulate a portable hearing aid prototype to alter compression threshold (CT), gain below CT, compression ratio (CR), and noise suppression responses for each of the three channels.

Overall, evidence shows that listeners are able to consistently select responses close to preferred settings. However, selection of preferred response could be biased by the starting point. Keidser, Dillon, et al. (2008) reported that the trained responses were influenced by the initial response especially at the low-frequencies and those with flat and gently sloping hearing impairment seemed to experience greater influence for the baseline response. In addition, listeners who were able to discriminate between responses were the ones who were less biased by the baseline response. When the listeners were fitted with responses that exhibited greater deviation from prescribed settings, more varied selections were made than when starting with prescribed settings. Dreschler et al. (2008) also found that listeners could be biased toward lower gain if the initial gain was lower than prescribed targets and steeper slope if the baseline response slope was steeper. In the study, preferred gain was reduced further by 2.5 dB when the baseline was reduced by 6 dB compared to NAL-NL1 targets, compared with what was preferred when the baseline was the target. Test-retest reliability was better when the baseline response had a flatter slope. In another crossover study where listeners were initially fitted with hearing aids that were 6 dB below or 6 dB above NAL-NL1 targets and the volume control allowing ± 8 dB of change around the initial volume setting, Mueller et al. (2008) also found that baseline settings could influence the final chosen response after training. When listeners started with lower gain, the final preferred gain settings were reduced by an average of 5 dB compared to the NAL-NL1 target. However, when the initial gain was higher than the targets, their final preferred gain averaged about 4 dB greater than NAL-NL1 prescribed gain. It must be noted here that the design of the Mueller et al. study as described above might have biased the participants, so that they might not have been able to consistently reach the same response from the two baseline settings.

Based on their findings, Dreschler et al. (2008) and Keidser, Dillon, et al. (2008) recommended that training should start with an appropriately prescribed baseline response and the alternative responses should be perceptually different to ensure consistent selection of preferred responses.

Do the Trained Parameters Result in Positive Outcomes, Compared to Untrained Responses?

There are very few peer-reviewed studies that had examined outcomes with trainable hearing aids. Three studies (Dreschler et al., 2008; Mueller et al., 2008; Zakis et al., 2007) are of Levels 3 and 4 evidence (Cox, 2005; Wong & Hickson, 2012) and preference for trained responses was shown. Improvement in speech perception and satisfaction with trained responses were reported by Mueller et al. (2008).

Zakis et al. (2007) found that, in a laboratory setting, more listeners preferred the trained hearing aid response. In their study, 9 listeners preferred the trained responses, 3 did not show significant preference while 1 preferred the untrained response and the authors believed that this was due to measurement errors made during fitting. In a later double-blind take-home study, these participants also preferred the trained response to the prescribed settings that were adjusted in the clinic. In a crossover blinded take-home trial, Mueller et al. (2008) also found that listeners are able to train their hearing aids to the preferred settings to improve speech perception as well as satisfaction with aided loudness although no comparison was made to see if the (two) trained responses were preferred to the prescribed response.

How Well Does a Hearing Aid Algorithm Converge to the Users’ Preferred Responses?

There is very little information on whether hearing aid algorithms would converge to users’ preferred responses. Various algorithms for categorizing acoustic environments or statistical models for deriving the preferred responses have been described for use in hearing aids (e.g., Alexandre, Cauadra, Alvarez, Rosa-Zurera, & Lopez-Ferreras, 2006; Chalupper, 2006; Chalupper, Junius, & Powers, 2009; Dijkstra, Ypma, de Vries, & Leenen, 2007; Lamarche, Giguere, Gueaieb, Aboulnasr, & Othman, 2010; Rahal, 2010; Zakis et al., 2007; Zakis, McDermott, & Fisher, 2001). As mentioned earlier, Zakis et al. (2007) was a Level 4 study. Level of study was not assigned to other studies reviewed in this section because some studies did not provide sufficient information on the methodologies that were used to evaluate the outcomes with these algorithms (e.g., Chalupper 2006; Chalupper et al., 2009) and Rahl (2010) trialed the algorithm on simulated users only. Other studies have only described the algorithms without any outcome data (e.g., Alexandre et al., 2006; Büchler et al., 2005; Dijkstra et al., 2007; Lamarche et al., 2010; Zakis et al., 2001). Trainable parameters could include compression threshold, compression ratio, gain below compression threshold, frequency shaping, noise suppression, microphone mode, and spectral enhancement in addition to gain adjustments (Chalupper et al., 2009; Dreschler et al., 2008; Taylor, 2011; Zakis et al., 2007).

Most training algorithms are sound classification based (e.g., Alexandre et al., 2006; Lamarche et al., 2010; Rahal, 2010) and therefore a relevant question is whether sounds are successfully classified using these algorithms. Alexandre et al. (2006) reported an algorithm with an accuracy of up to 95% in classifying sounds into four different categories: speech in quiet, speech in noise, stationary noise, and nonstationary noise. Rahal (2010) found accurate identification of acoustic environment using an algorithm with simulated users with moderate, severe, and profound hearing impairment. There is scarce information on the accuracy of sound classification with commercial instruments. In addition, it is not clear how these algorithms will work in real-life situations and whether these classifications would result in preferred hearing aid responses. Theoretically speaking, extensive training is needed for these instruments to respond accurately to a large variety of listening situations.

The NAL group has been using the processor designed by Zakis et al. (2001) and their research (e.g., Zakis et al., 2007) revealed that trained responses were preferred by users. The algorithm analyses the listening environments as a continuum based on some selected parameters that describe the acoustic characteristics of the environments. In theory, the instruments could thus analyze any auditory scene and apply all collected trained data to any listening situation. Their research also suggests that training of hearing aids does not need to rely on sound classification.

Overall, there is little evidence on how accurate hearing aid algorithms are in arriving at preferred responses. The effectiveness of these algorithms is implied when research shows that users who used these algorithms preferred the trained responses to the untrained responses.

Conclusion: Overall, there is evidence that users are able to discriminate between various responses and manipulate up to four parameters (Dreschler et al., 2008) to train the hearing aids to preferred settings and that hearing aid users would prefer trained responses to untrained responses (Dreschler et al., 2008; Keidser, Dillon, et al., 2008; Mueller et al., 2008; Zakis et al., 2007). These are Level 3 and Level 4 evidence, resulting in an overall Grade “B” recommendation (Wong & Hickson, 2012). The recommendation is to start with an initial setting that is close to the preferred response, but clinicians should be aware that the initial setting could bias the listeners’ final choice (Dreschler et al., 2008; Keiders, Dillon, et al., 2008). There is little evidence available in terms of other outcomes (e.g., speech perception, satisfaction) with trainable hearing aids and how well algorithms are able to adjust hearing aids to optimize performance in various real-life situations.

• PICO Question No. 4: Could hearing aid users follow written or illustrated instructions to manage the assembly, physical fit, and use of a self-fitting hearing aid?

If a self-fitting hearing aid is to be introduced into a market where there is limited provision of audiological services, the user or the caretaker could be assembling and using the hearing aid without the supervision of a professional. By following written or illustrated instructions, users would be selecting an appropriate size instant-fit tip and tubing and connecting them to the hearing aid. Users would also have to learn to insert and change battery and insert the hearing aid into the ear as well as how to use the aid in daily situations. These instructions should be easy to follow and self-explanatory and whether these instructions could achieve the above purposes should be evaluated. Unfortunately, the literature search did not result in any published article in this area.

However, at the time of writing this review, only one relevant paper that is of Level 4 evidence, was found (Convery et al., 2012). Convery et al. report a laboratory experiment where the majority of participants with hearing impairment were able to assemble a pair of conventional hearing aids without the coaching of a professional. Illustrated instructions with text written at a Grade 3 level were given to 80 urban dwellers with ages ranging from 45 to 90 years. Convery et al. showed that male participants with higher health literacy, and to lesser extent higher cognitive functions, were better able to handle the assembly of the aid without assistance. However, female participants with higher health literacy, and to lesser extent with higher age and shorter hearing aid use experience, were better able to assemble the aid without errors. The study showed that more than 70% of the 80 participants were able to independently follow written instructions and illustrations to complete individual assembly steps and 20% of participants were able to complete all tasks accurately and without assistance. Some task steps may be apparent to an audiologist (e.g., identification of the tubing with the correct length) but not to a client and therefore clear instructions are needed. As expected, insertion of the aid into the ear was one of the more difficult tasks. Overall, this study suggests that female users, those who are more willing to seek help from their partners, and those who have higher health literacy are more likely to be able to assemble and use the hearing aid more effectively. Table 2 also lists this study.

Much research is still needed to verify that hearing aid users or their communication partners from various cultural and literacy backgrounds are able to assemble and manage the self-fit procedure of these hearing aids, and to train and use the aids in daily situation. With health literacy being an issue in third world countries—regions that self-fitting hearing aids target, instructions must be designed to avoid ambiguity and to maximize success in hearing aid use. These instructions must be evaluated in the respective country where they will be used, to increase usage success rate. Research from health literacy could help guide the writing of these instructions as discussed in Caposecco et al. (2012).

Summary of Evidence on Self-Fitting Hearing Aids

According to the evidence presented, the author concludes the following:

1. There is Level 2 evidence that in-situ thresholds could be reliably obtained (PICO Question No. 1). To ensure validity of measurement transducer, specific REDD corrections must be applied, while reliable insertion of the ear-tips and ambient noise levels should be considered.

2. There is a grade A recommendation that prescriptive fittings would provide adequate initial amplification fitting (PICO Question No. 2).

3. There is a Grade “B” recommendation that users are able to train the hearing aids to preferred settings (PICO Question No. 3). However, initial responses could affect the selection of preferred settings. Users prefer trained responses in real-life situations as well as in laboratory settings.

There is, however, very little evidence in terms of other outcomes with self-fitting or trainable hearing aids (PICO Question No. 3) or regarding the successful assembly and use of a fully self-fitted hearing aid without the supervision of a clinician (PICO Question No. 4).

Although much more research is still needed, the potential benefit that self-fitting hearing aids could provide should not be underestimated. Self-fitting hearing aids, if successfully implemented, would essentially allow clients to achieve the best fitting in the comfort of their home and allow instrument fitting in locations where there are limited professional hearing health care services. Keidser, Convery, and Dillon (2007) found that in a laboratory setting, 91% of clients they interviewed were positive about the idea of a trainable aid. In addition, about three quarters of their participants were willing to spend up to a few weeks in training the hearing aids. According to Kochkin (2007), among the nonusers who did not use hearing aids because of low benefit and background noise, 19% of them would consider purchasing hearing aids if the aids could be personally adjusted. Because the hearing aid is customized in a user’s own listening environments, it gives the users a sense of ownership of their hearing problems as well as control. The need to adjust hearing aids and therefore the stigma associated with adjusting the aid in public are reduced. When a trade-off between comfort, sound quality, and speech intelligibility is needed, the users could make their own decisions.

Gaps in Research

During the review of the literature for evidence, the author found significant variations in research methodologies in the evaluation of trainable hearing aids (PICO Question No. 3); these could significantly impact on the findings but their effects have not been evaluated. Several factors were identified. First, there were variations in the type of control for training hearing aids. For example, in Dreschler et al. (2008), participants trialed four controls that manipulated different combinations of volume and tone balance (i.e., overall tone balance, bass, mid, and treble) and preference was shown for three types of control that involve volume adjustment. In Taylor (2011), user control allowed programming of four separate memories by adjustment of a single algorithm that combined gain, digital noise reduction, microphone mode, and spectral enhancement.

Second, the allowable changes in level differed across studies. For example, in Mueller et al. (2008), a gain change allowance of ± 8 dB was implemented. To identify participants whose selected response might have been limited by floor and ceiling effects, a ± 6 dB criterion was used. This criterion resulted in 10 of the 22 participants being excluded. In contrast, in Keidser, Dillon, et al. (2008), the majority of the listeners exhibited selection of responses within the ± 16 dB limit. Thus, when the equipment allows, a larger allowance for gain change would facilitate selection of trained responses.

Third, more research is needed in examining how consistency in response selection affects the training and whether a final preferred response could be reached. Across studies, there are variations in the degree of consistency of response selection. More consistent choice of responses is noted when the initial response is closer to the preferred responses. There are also discussions on whether hearing aid users with steeper hearing impairment would be more consistent in selecting preferred response. While Dreschler et al. (2008) suggested that those with a gently sloping hearing loss were less consistent than those with a flat or steeply sloping loss, Keidser, Dillon, et al. (2008) found that listeners with sloping loss were more consistent in selecting high-frequency gain than those with flat hearing configuration. More consistency in judging high-frequency gain has been attributed to the narrow dynamic range at the high frequencies, thus causing changes in high frequency gain to be perceptually larger (Keidser, Dillon, et al., 2008). However, there is also large variability across participants, and cochlear dead regions (Moore, 2004) could cause some listeners to fail in picking up subtle differences across various responses. While one could argue that inconsistency in adjustment could prevent the identification of the most preferred response, there is no evidence to rule out the possibility that listeners could accept a wide range of responses without detrimental effects on outcomes.

Finally, when training is performed that is nonsynchronized in two ears, mismatch in hearing aid response could result and whether these mismatches would affect outcomes is unclear. Hornsby and Mueller (2008) found that on average, trained gains were similar in both ears. Very few listeners demonstrated mismatches between ears, and the mismatches did not result in significant differences in speech understanding ability measured at 50 or 65 dBA. Further research is needed to evaluate whether mismatches would reduce effectiveness or preference of the trained responses or outcomes.

Besides the issues identified above for trainable hearing aids, further evidence on the use of self-fitting hearing aids is needed in several areas. First, there is no research on candidacy for self-fitting hearing aids. There is however one study that has examined candidacy for trainable hearing aids and it reported a discriminant analysis that showed potential users of trainable hearing aids to be younger, better educated males with milder, more sloping, less symmetrical hearing impairment, perceived greater aided handicap and with a positive attitude toward the aid (Keidser et al., 2007). This study was conducted in a laboratory where potential users were introduced to the concept of a trainable hearing aid and manipulated the controls. These findings have not been verified in a field trial or in other studies. In addition, this profile may not be applicable to self-fitting hearing aids that do not include the training feature.

Second, how in-situ threshold measurements could be conducted with reliability and validity using instant-fit ear tips and without the direct supervision of a professional should be evaluated.

Third, whether self-fitting hearing aids would produce equivalent or better outcomes, compared to when the whole process of hearing aid fitting is conducted under the supervision of a clinician, should be examined. While many components of the hearing aid fitting process could potentially be replaced by automation or self-training, the interaction between the clinician, the user, and their communication partners could not be substituted.

Conclusion

Self-fitting hearing aids have the potential to be used in areas where access to hearing health care service is not possible. Although there is good quality evidence that users are able to train the hearing aids to preferred settings and these settings are preferred over untrained responses, how in-situ thresholds could be measured and how hearing aids could be assembled and used without the direct supervision of a clinician requires further examination. Outcomes with these hearing aids also require further documentation.