Other than the obvious effect of reduced audibility of conversational

(1)

Larry E. Humes Department of Speech and Hearing Sciences Indiana University Bloomington Laurel A. Christensen Department of Communication Disorders LSU Medical Center New Orleans, LA Fred H. Bess Andrea Hedley-Williams Division of Hearing and Speech Sciences Vanderbilt University School of Medicine Nashville, TN

In this clinical study, 110 patients seen at three different clinical facilities were fit binaurally with linear, in-the-canal (ITC) hearing aids. All patients were new hearing aid users. Each of the hearing aids was equipped with an adjustable control that could be set by one of the audiologists (Audiologist A) at each site to convert it from a linear instrument to an experimental nonlinear one with automatic reduction of low-frequency gain at high input levels (or base increase at low levels, BILL). Both the patient and the audiologist performing the outcome testing at each site (Audiologist B) were blind as to the present setting of the hearing aid. Each participant was enrolled in the study for a total of 12 weeks, with the hearing aid set to either the linear or BILL-processing mode of operation for the first 8 weeks and the opposite setting for a subsequent 4-week period. In summary, this was a prospective, double-blind, crossover study of 110 new hearing-aid users. Outcome measures focused on hearing-aid benefit and included both objective and subjective measures. Objective measures were derived from scores on the Northwestern University Auditory Test No. 6 (NU-6) and the Connected Speech Test (CST) obtained for all possible combinations of two speech presentation levels (60 and 75 dB SPL), two types of background noise (cafeteria noise and multitalker babble), and two signal-to-noise ratios (+5 and +10 dB). Subjective outcome measures included magnitude estima-tion of listening effort (MELE), the abbreviated form of the Hearing Aid Performance Inventory (HAPI), and estimations of hearing-aid usage based on daily-use logs kept by the participants. All of these measures were used to evaluate the benefit provided by linear amplification and the benefit resulting from the experimental BILL process-ing. Participant preferences for the experimental BILL-processing scheme or linear processing were also examined by using a paired-comparison task at the end of the study. Results were analyzed separately for three subgroups of patients (mild, moderate, severe) formed on the basis of their average hearing loss at 500, 1000, 2000, and 4000 Hz. In all three subgroups, significant improvement in performance was observed for linear amplification and for BILL processing when compared to unaided performance. There were no significant differences in aided performance, however, between linear processing and the experimental BILL processing.

KEY WORDS: hearing aids, benefit, linear processing, BILL-processing, speech understanding in noise

A Comparison of the Benefit

Provided by Well-Fit Linear

Hearing Aids and Instruments

With Automatic Reductions of

Low-Frequency Gain

O

ther than the obvious effect of reduced audibility of conversa-tional level speech, it has long been observed that sensorineural hearing loss interferes with communication in noisy backgrounds.

(2)

the past several decades, there have been no compre-hensive, large-scale, prospective clinical studies of the benefits provided by such instruments. Such informa-tion is critically important as alternatives to linear am-plification, such as BILL-processing, multichannel com-pression, and wide-dynamic-range comcom-pression, are developed and evaluated. When studies have been con-ducted comparing linear and nonlinear systems, it has typically been the case that the hearing aids under in-vestigation differ in more than just their linearity. In view of the foregoing, the purpose of this study was to evaluate the clinical performance of a linear ITC hear-ing aid with and without activation of an experimental BILL-based processing option. The hearing aid, with and without the BILL processing activated, was evaluated using objective and subjective tests of device perfor-mance and measures of participant preference. As de-scribed in more detail below, a total of 110 persons with sensorineural hearing loss participated in this prospec-tive study conducted at three different clinical sites.

Protocol

Overview of Design

This was a prospective, double-blind, crossover, con-trolled study. The study population was divided into three groups according to their hearing loss: (1) mild, defined as a four-frequency (500, 1000, 2000, and 4000 Hz) pure tone average (4PTA) of 25–40 dB HL; (2) mod-erate, 4PTA of 41–60 dB HL; and (3) severe, 4PTA of 61–85 dB HL. Each of the participants was tested with the same hearing aid in two different circuit configura-tions: BILL option and linear. Randomly, one-half of the participants were dispensed the BILL circuit active first, and the other one-half received the linear circuit active first. After an appropriate adjustment period (discussed below), the participant’s performance with the first cir-cuit configuration was assessed. After that assessment, the participant was switched to the second circuit con-figuration and was given an appropriate adjustment period. The participant then returned for an assessment of performance with the second circuit configuration.

Assignment of initial aided condition (BILL vs. lin-ear) was randomized such that an equal number of par-ticipants was included in each group. Further, the study was double blind. The participants did not know which circuit type was being tested first. At each test site, one audiologist (A) was responsible for fitting the hearing aids. A second audiologist (B), without knowledge of which cir-cuit type a participant was using for any given test ses-sion, performed the performance testing. The participants and Audiologist B were also blind regarding the hearing aid’s manufacturer. This was accomplished by having no manufacturer’s labeling or markings on the hearing aid For the vast majority of listeners with sensorineural

hear-ing loss seen by audiologists and hearhear-ing instrument spe-cialists over the past few decades, the conventional ap-proach to rehabilitation has been to fit the person having sensorineural hearing loss with a linear, head-worn hear-ing aid with output limithear-ing accomplished via peak clip-ping (Bess & Humes, 1995; Hawkins & Naidoo, 1993). To improve the communication performance of persons wear-ing such instruments in noise a variety of modifications to the basic linear, peak-clipping instrument has been investigated. High-pass filtering in a hearing aid has been advanced as one technique to minimize the effect of back-ground noise on speech communication.

High-pass filtering can be accomplished in a hear-ing aid in two ways. First, the hearhear-ing aid can be set to have a static high-pass response. At all times, the hear-ing aid’s frequency response is set to minimize low-fre-quency gain and to maximize high-frelow-fre-quency gain. The principal drawback to this approach is that many lis-teners do not like the sound quality of such an amplify-ing system in quieter environments (Punch & Beck, 1986). Further, there is useful speech information in the lower frequencies that should be provided to the listener, if at all possible. This low-frequency information should be restricted only if there is a drawback to its presence (such as excessive upward spread of masking or unde-sirable sound quality).

The second manner in which low-frequency energy can be attenuated is via adaptive high-pass filtering (Fabry & Walden, 1990). In this approach, the hearing aid is set to have a relatively broad response in quieter environments. As the overall sound level in the envi-ronment increases above a certain threshold level, the hearing aid automatically shifts to a more high-pass response. The assumption behind such an approach is that the listener is likely to prefer a broader response if there is a minimal amount of noise present. If the over-all level of the listening environment rises, there is a high likelihood that a significant amount of this energy is low-frequency noise (Ono, Kanzaki, & Mizoi, 1983). Therefore, the hearing aid automatically reduces gain for this undesirable low-frequency sound and concen-trates the response of the hearing aid to the more im-portant high-frequency region. This automatic low-fre-quency reduction in gain at high input levels is implemented through different methods in several mod-els of commercially available hearing aids. It has been variously termed “automatic signal processing (ASP),” “adaptive frequency response (AFR),” and “bass increase at low levels (BILL).” Some commercial devices incor-porate adaptive high-pass filtering in single-channel analog devices, whereas others make use of single- or multichannel compression to realize BILL processing.

(3)

casing or accompanying hardcopy materials, as well as by channeling all communications between the manufac-turer and test site through Audiologist A.

This study was carried out at three independent audiology research facilities: (1) the Division of Hear-ing and Speech Sciences, Vanderbilt University School of Medicine, Nashville, Tennessee; (2) the Department of Communication Disorders, Louisiana State Univer-sity Medical Center, New Orleans, Louisiana; and (3) the Department of Otolaryngology, University of Iowa Hospitals and Clinics, Iowa City, Iowa. These facilities were all in close physical and administrative proximity to an active clinical hearing aid dispensing program. Investigators at all research facilities obtained the ap-propriate Institutional Review Board approval for the use of human participants at their facilities.

Participants

The participant was eligible for entry into the study if the participant: (a) was at least 18 years of age or older; (b) had a flat or gently sloping sensorineural hearing loss (slope 0–15 dB/octave from 250 Hz to 4000 Hz, no inter-octave change of greater than 20 dB), consistent with the fitting guidelines of the hearing aid; (c) had a hearing loss that was symmetrical (within 10 dB at octave and half-octave frequencies from 250 Hz to 4000 Hz); (d) had thresh-olds that ranged between 25 and 85 dB HL at octave and inter-octave intervals from 250 Hz to 4000 Hz, inclusively; (e) had received medical clearance by a physician to use hearing aids binaurally; (f) had read, understood, and signed the Informed Consent form; (g) had normal tympanograms bilaterally; and (h) resided within a rea-sonable distance of the test facility.

The participant was ineligible for entry into the study if the participant: (a) was a prior hearing aid user, although the participant could have participated in a prior 30-day trial of amplification (but not within the previous 6 months); (b) had any indication of a medi-cally or surgimedi-cally treatable ear disorder; (c) had an ex-ternal auditory canal with a size and shape that was inconsistent with an appropriately sized device; (d) had known fluctuating or rapidly progressive hearing loss; (e) had any cognitive, medical, or language-based con-dition that limited his/her ability to complete all test procedures; (f) was currently participating in any other clinical trials; (g) had any known disease or condition, other than sensorineural hearing loss, that might affect hearing or cognition; or (h) was taking any medication(s) that could affect hearing or cognition.

Participants could also be discontinued from the study for one of the following reasons: (a) development of any of the exclusion criteria during the course of the study, such as medical/surgical ear condition, fluctuating

hear-ing loss, disease/condition affecthear-ing hearhear-ing/cognition, or VIII-nerve tumor; (b) inability or unwillingness to main-tain wearing schedule; (c) inability or unwillingness to make or keep clinic appointments during study; (d) death or disability; (e) relocation of participant; (f) over 3 weeks cumulative off-the-ear time; (g) change in participant’s thresholds at two or more test frequencies by more than 10 dB, as compared to the initial baseline audiogram; or (h) a missed evaluation visit.

For their participation in the study, the participants received their hearing aids at no charge. All clinical ser-vices provided during the study were also provided at no charge to the participant.

The mean air-conduction audiograms for the left (squares) and right (circles) ears for each of the three hearing-loss subgroups are provided in Figure 1. Stan-dard deviations are also given in the upper portion of Fig-ure 1. The mean audiogram for each subgroup shows a bilaterally symmetrical, gently sloping configuration that varies in severity across subgroups. Table 1 shows the distribution of participants by site, severity of hearing loss, and gender. Age ranges are also provided. As can be seen, most of the participants with severe hearing loss were tested at the University of Iowa and were mostly men, whereas those with mild hearing loss were fairly evenly distributed across all three sites and were mostly women.

Figure 1. Mean right (circles) and left (squares) air-conduction thresholds (ANSI, 1989) for the mild, moderate, and severe hearing-loss participant groups. N indicates the number of participants in each group. Corresponding standard deviations for each group are plotted at the top of the audiogram.

(4)

Hearing Aid

All participants were fit with binaural, custom, in-the-canal (ITC) hearing aids manufactured by Dahlberg, Inc. The shells of the hearing aids were unvented (which was considered to be generally appropriate given the degree and configuration of the hearing losses) and were fabricated from standard materials and with medium-length canals. The test hearing aids included an experi-mental BILL circuit that reduced the amount of gain in the low frequencies as the input intensity increased. A manually adjustable potentiometer on the device, ac-cessible only to the audiologist, allowed the BILL cir-cuit to be deactivated. When the circir-cuit was deactivated, the device acted as a linear amplifier—the most com-mon circuit currently used on the hearing aid market. This particular linear circuit, however, made use of a patented Class AB output stage that, according to the manufacturer, incorporated low-distortion peak clipping, reduced low-frequency circuit noise, and low current drain. When the BILL circuit was active, the threshold for activation of the BILL processing was set at the low-est possible level (about 60 dB SPL). This was essen-tially a single-channel analog device with adaptive high-pass filtering with a variable threshold for activation of the adaptive filtering. When the potentiometer was ad-justed to one extreme position, inputs as low as 60 dB SPL would activate the adaptive high-pass filtering and maximum reduction of low frequencies occurred, whereas adjustment to the other extreme position re-sulted in deactivation of the adaptive filtering (linear processing). Figure 2 illustrates the effect of these two extreme potentiometer settings on the frequency re-sponse of the hearing aid as a function of input level. The biggest reductions in low-frequency gain are ob-tained for the BILL circuit for inputs of 70 and 80 dB. Note that activation of the BILL circuit actually reduces the gain in the mid and high frequencies by about 8–10 dB, compared to the linear setting at an input level of Table 1. Detailed descriptions of study participants by clinical site, participant hearing loss severity, and participant gender.

Hearing loss Site # Males # Females Age range (y)

Mild LSU 3 11 27–85 Vanderbilt 3 7 44–84 Iowa 3 14 48–83 Moderate LSU 8 12 34–83 Vanderbilt 8 9 45–83 Iowa 7 9 25–87 Severe LSU 1 1 54–66 Vanderbilt 0 1 90 Iowa 12 1 47–85

Figure 2. Illustration of the effects of hearing-aid circuit setting (linear vs. BILL) on the frequency response measured in accordance with ANSI S3.22-1987. Frequency responses are shown for both the linear (solid lines) and BILL (dashed lines) settings. Each panel depicts the frequency responses for each setting obtained for a different input level (50–90 dB SPL). Results shown are for the highest gain linear circuit included in this study.

(5)

used to listening to an amplified signal. It may not be until several weeks after being fit with a new hearing aid that some users can make maximal use of the new auditory information. Therefore, in order to increase the likelihood of achieving optimal performance with the hearing aids in this study, the user was allowed to adapt to its use. A longer period of acclimitization (8 weeks) was implemented for the trial with the first cir-cuit because it was anticipated that there would be a greater need for adjustment to the use of amplification initially that included familiarization with the insertion/ removal of the device, operation of controls, replacement of batteries, and adjustment to the new sound quality.

During the adjustment period, no changes in the response or casing of the hearing aid were allowed, ex-cept in cases of reported physical or loudness discom-fort or acoustic feedback. In typical clinical practice, the audiologist may make fine-tuning adjustments in the participant’s hearing aid during the first several weeks. These adjustments are typically due to sound-quality preferences on the part of the participant. However, these adjustments are not predictable and will vary from participant to participant depending on factors that are not fully understood. The NAL-R prescriptive rules are well developed, well supported in the literature, and gen-erally accepted by audiologists. Therefore, in an attempt to provide consistent responses from participant to par-ticipant, variations from the NAL-R prescribed hearing aid responses were not permitted during the study.

The first evaluation session took place no more than 2 weeks after the end of the 2-month adjustment period. After the first circuit had been evaluated, the participant’s hearing aids were changed to the alternative circuit. The participant was then given a 1-month adjustment period with the second circuit. The second evaluation session took place no more than 2 weeks after the end of this 1-month adjustment period. Follow-up phone and mail con-tacts were made in cases in which participants did not keep scheduled evaluation sessions.

Speech levels vary across a broad range in typical communicative environments (Pearsons, Bennett, & Fidell, 1977). In this study, testing was performed at two different speech levels, 60 and 75 dB SPL, which represent normal and raised speech (Pearsons et al., 1977). Further, the degree of low-frequency amplifica-tion in the experimental BILL circuit varies inversely with the overall input level. Evaluating nonlinear cir-cuits at more than one level has been recommended by others (e.g., French-St. George, Engebretsen, & O’Connell, 1992). By testing at these two levels, we were able to assess BILL-circuit performance aspects and aided benefit across a range of typical input levels.

A variety of background noise types has been used in past evaluations of automatic low-frequency reduc-60 dB and, to a lesser extent, at 70 dB. Although results

shown in Figure 2 are only for the hearing aid circuit with the highest gain, identical patterns of results were observed for the circuit with lowest gain, although the reduction of gain in the mid- and high-frequency regions was less pronounced (4–6 dB) for this circuit. The rela-tive effects of BILL processing shown in this figure, more-over, are independent of volume-control setting.

During the study, the hearing aids were marked only with a serial number—no brand identification informa-tion. At study completion, they were returned to the manufacturer for proper labeling before they were re-turned to the participants. All hearing aids were tested electroacoustically in accordance with ANSI S3.22-1987 before dispensing and at all post-fit evaluations.

The gain, frequency response, and maximum out-put of each hearing aid was individually selected by the manufacturer on the basis of audiometric characteris-tics of each test ear. The gain and frequency response were selected to meet the targets prescribed by the re-vised National Acoustics Laboratory (NAL-R) fitting pro-cedure (Byrne & Dillon, 1986) when set for the linear mode of operation. The desired frequency response was confirmed via probe-tube measurement techniques at hearing-aid delivery. Consistent with the recommenda-tions of Mueller (1992), the real-ear insertion gain of the hearing aid was within 10 dB at all octave frequen-cies below 4000 Hz and within 15 dB at 4000 Hz. The mean target gain curves based on the NAL-R formula and the mean real-ear insertion responses for the lin-ear setting obtained from the participants for a 60-dB-SPL input are shown for each ear and hearing-loss sub-group in Figure 3. There is excellent agreement between the prescribed (target) gain and the measured insertion gain in all cases from 250 Hz to 4000 Hz, but a consis-tent underamplification of about 8–12 dB at 6000 Hz.

SSPL90 targets were derived from loudness discom-fort levels (LDLs) measured using procedures described by Hawkins, Walden, Montgomery, and Prosek (1987). The maximum output of the device was selected to be no more than 10 dB below the SSPL90 target maximum at octave frequencies from 500 Hz though 4000 Hz.

Hearing Aid Evaluations

No evaluation of hearing aid benefit took place until after a 2-month adjustment period with the first circuit. This adjustment period is consistent with recent evi-dence of the maturation of hearing aid benefit or “acclimitization” (Cox & Alexander, 1992; Gatehouse, 1992, 1993), allowing for plateau performance, al-though there is some debate about the size and sig-nificance of this effect (Turner, Humes, Bentler, & Cox, 1996). Nonetheless, it was assumed that it takes some finite amount of time for a new hearing aid user to get

(6)

tion amplification. The performance of automatic low-frequency reduction amplification has been demon-strated to vary with background noise type (Fabry, 1991). Specifically, as the noise shifts from predominantly low-frequency energy to more broadband energy, the benefi-cial effects of the circuit tend to diminish. Also, the use

of steady-state narrow-band, speech, or white noise does not replicate real-world listening environments. There-fore, in this study, two noise types were used: cafeteria noise (Auditec) and multitalker speech babble (Auditec). Spectrally, cafeteria noise tends to be dominated by low-frequency energy, but continues to have some higher Figure 3. Mean target real-ear insertion gain generated by the NAL-R formula (squares) compared to the

mean real-ear insertion gain measured (circles) for a 60-dB input. Vertical bars above and below circles represent one standard deviation above and below the mean. Left-hand panels and unfilled symbols are for left ear, whereas right-hand panels and filled symbols are for right ear. Top panels present data for group with mild hearing loss, middle panels contain data for group with moderate hearing loss, and bottom panels present data for group with severe hearing loss.

(7)

frequency content because of the presence of some com-peting speech. Babble tends to have the same long-term characteristics as test speech material. By using both caf-eteria noise and speech babble, we assessed the effects of the BILL circuit and the benefit provided by both BILL-processing and linear hearing aids in realistic, yet spec-trally different, background conditions.

Procedures

Before the beginning of this study, investigators from each clinical site met with the study sponsor (the manu-facturer) to review the protocol and data-handling is-sues. Identical materials, protocols, and forms were used at all sites. An independent study monitor familiar with medical-device clinical field studies was retained by the study sponsor to monitor the study procedures at each site; this person made several visits to the sites through-out the duration of the study to assure uniformity and adherence to the protocol.

At the patient’s first visit to the facility, referred to as the recruitment visit, information on the participant’s history, demographics, eligibility, unaided hearing per-formance, and hearing-aid fit was collected and evalu-ated. Ear impressions were obtained at this visit, and the hearing aids were ordered. The investigator also ex-plained the Physician Release Form and the Informed Consent Form to the participant during this visit.

The second visit, the dispensing visit, took place within 3 weeks of the recruitment visit. At this visit, the participant returned the signed Informed Consent and Physician Release Forms, and before the hearing aids were dispensed Audiologist B confirmed the participant’s hearing thresholds measured previously. Next, Audiologist A evaluated the hearing aids in the testbox, measured the real-ear performance, and gave a hearing-aid-use log to the participant with an explana-tion of its use. Participants were instructed to wear their hearing aids at least 5 hours per day, 5 days per week. Audiologist B then obtained unaided speech-recognition scores, along with magnitude estimates of listening ef-fort (see below), from the participant.

The third visit to the facility represented the first aided evaluation of performance and took place no more than 2 weeks after the 2-month “acclimatization” period (time from dispensing visit to the first aided evaluation visit). At this visit, hearing sensitivity was again measured by Audiolo-gist B to confirm the stability of hearing thresholds. Audi-ologist A evaluated the hearing aids in the testbox and collected the hearing-aid-use log from the participant. In addition, the Hearing Aid Performance Inventory (HAPI; Walden, Demorest, & Hepler, 1984) was administered in its abbreviated form (Schum, 1993) using a pencil-and-paper format, and a new hearing-aid-use log was given

to the participant. Audiologist B then obtained aided speech-recognition and listening-effort scores. Once these measures were completed, the hearing aid was switched to the alternate setting (linear or BILL) by Audiologist A.

The fourth visit to the facility represented the second and final aided evaluation of performance and took place no more than 2 weeks after the 1-month adjustment period (time from the first aided evaluation visit to the second aided evaluation visit). At this visit, hearing sensitivity was again measured by Audiologist B to confirm the stability of hear-ing thresholds. Audiologist A evaluated the hearhear-ing aids in the testbox and collected the hearing-aid-use log from the participant. In addition, the abbreviated HAPI was again administered. Audiologist B then obtained aided speech-rec-ognition and listening-effort scores. Finally, a series of paired-comparison measurements was performed by the partici-pant with the assistance of Audiologist A.

Table 2 provides a summary of the sequence of mea-surements peformed during this study. It also indicates which audiologist (A [not blind] or B [blind]) at each site was involved in the measurements. Explanations of hearing-aid function, use, care, and communication strategies were provided by Audiologist A.

Outcome Measures

Taped speech-recognition tests (NU-6; Tillman & Carhart, 1966 / Connected Speech Test, CST; Cox, Alexander, Gilmore, & Pusakulich, 1988) were admin-istered at +5 and +10 dB signal-to-noise ratios (S/N) in babble and cafeteria noise at speech presentation levels of 60 and 75 dB SPL. Oral responses were used for both tests. All scores for both the NU-6 and CST tests were based on 50 items. Speech and noise were presented from a single loudspeaker located at zero-degrees azimuth and one meter from the center of the participant’s head. All sound pressure levels refer to those measured in the sound field using the method of substitution (Skinner, 1988).

A magnitude-estimate of listening effort (MELE) was also developed and used in this study. In this task, the listener rated the difficulty of listening to speech in caf-eteria noise and babble backgrounds. The listener heard a 10-sentence encyclopedia-style passage from the CST. Four passages were presented at 60 and 75 dB SPL in both noise backgrounds at +5 and +10 dB S/N (32 total passages). After each passage, the listener rated “ease of listening” on a 0–100 scale, with 100 representing ex-tremely easy listening. In each listening condition, the ratings from the last three passages were averaged and the first in each condition was discarded as practice.

After completion of the second round of objective and subjective measures of performance, the participant was then placed in the sound field and presented with

(8)

continuous discourse (the Rainbow Passage recording available on the Q/MASS Speech Audiometry, Volume 1 compact disk) at the same levels and signal-to-noise ra-tios described above. The audiologist then switched the circuit back and forth from the BILL to the linear set-ting. The participant heard a 15- to 20-second passage in one circuit setting, followed by a second 15- to 20-second passage with the other circuit active. The order of which circuit was active first was randomly varied. The participant then chose which circuit he/she preferred to use in that noise condition.

The participant used the following rating scale to indicate preference: (1) Circuit A significantly better than circuit B; (2) Circuit A moderately better than circuit B;

(3) Circuit A slightly better than circuit B; (4) Circuit B slightly better than circuit A; (5) Circuit B moderately better than circuit A; (6) Circuit B significantly better than circuit A. The participant made one practice, then three test ratings, at each speech level, in both noise backgrounds and at each signal-to-noise ratio. The re-sults from the three test ratings were averaged in each listening condition.

Result

A total of 115 participants was enrolled in the study, but 1 was not dispensed a hearing aid and 4 Table 2. Summary of visits and forms used in the study.

Session Form/Data description Audiologist

1st visit: Recruitment Participant History Form B

Order Form

Pure tone thresholds B

Loudness discomfort levels B

Speech discrimination B

Tympanometry B

Hearing aid order information A

Impressions A

Eligibility Criteria (reviewed participant eligibility) A Physician Release (explained & gave to participant) A Informed Consent (explained and gave to participant) A 2nd visit: Dispensing Eligibility Criteria (reconfirmed eligibility and signed) A

Performance Testing form B

NU-6 monosyllabic word list scorea _B

Connected Speech Test scorea _B

Listening Effort Test scorea _B

Hearing Inventory (HHIE) A

Hearing Aid ANSI Performance A

Hearing Aid Use Log (explained & gave to participant) A 3rd visit: 1st Evaluation Performance Testing form B

NU-6 monosyllabic word list scoresa _B

Connected Speech Test scoresa _B

Listening Effort Test scoresa _B

Hearing Aid Performance Inventory (HAPI) A

Hearing Aid Use Log (explained & gave to participant) A 4th visit: 2nd Evaluation Performance Testing form B

NU-6 monosyllabic word list scoresa _B

Connected Speech Test scoresa _B

Listening Effort Test scoresa _B

Hearing Aid Performance Inventory (HAPI) A

Paired Comparison Testinga _A

(9)

were discontinued. Of the 4 discontinued participants, 1 was discontinued because closer evaluation revealed that the participant did not meet all of the inclusion criteria and should not have been enrolled, whereas the other 3 were lost to follow-up (did not return and could not be reached). Of the 110 remaining partici-pants, none missed a scheduled visit. The results de-scribed below are for the 110 who were enrolled and completed the entire study.

Group Data—Linear vs. BILL-Processing

NU-6 Test

Recall that the NU-6 test was performed at several signal-to-noise ratios, speech levels, and in two differ-ent types of noise background at the dispensing visit (unaided) and the first and second evaluations (aided). Figure 4 provides a summary of the mean aided NU-6 percent-correct data for the 110 participants with mild, moderate, or severe hearing loss, respectively. Table 3 provides the corresponding standard deviations for the mean values appearing in Figure 4.

Connected Speech Test

The Connected Speech Test (CST) was also per-formed for the same combinations of signal-to-noise ra-tio, speech level, and types of background noise used with the NU-6 at the dispensing visit (unaided) and at the first and second evaluations (aided). Figure 5 pro-vides a summary of the mean aided CST percent-cor-rect data for the 110 participants with mild, moderate, or severe hearing loss, respectively. Table 4 provides the corresponding standard deviations for the mean CST values appearing in Figure 5.

ANOVAs on NU-6 and CST Scores

Percent-correct scores for the NU-6 test and the CST were arcsine transformed to stabilize the error variance (Kirk, 1968), and the difference scores between the linear and BILL settings were computed. An analysis of variance (ANOVA) with a between-subject factor of hearing-loss group and four bivalent repeated-measures or within-sub-ject variables (test materials, signal-to-noise ratio, noise type, and speech level) was performed first and showed no statistically significant (p < .05) effects of hearing-loss group (either main effects or interactions). Consequently, a sec-ond ANOVA was performed on the difference scores (lin-ear vs. BILL) for all 110 participants and the four re-peated-measures, within-participant variables without regard to hearing-loss group. The results of this ANOVA are summarized in Table 5 and indicate that there was no difference in performance between the linear and BILL settings (i.e., the linear-minus-BILL difference scores did not differ significantly from zero). Moreover,

Figure 4. Mean percent-correct scores on the NU-6 test for the linear (cross-hatched bars) and BILL (unfilled bars) settings of the hearing aid. Panels are organized from top to bottom according to hearing loss group such that participants with mild hearing loss are in the top two panels, those with moderate hearing loss are in the middle two panels, and those with severe hearing loss are in the bottom two panels. All left-hand panels contain data for the 60-dB-SPL presentation level, whereas all right-hand panels contain data for the 75-dB-SPL presentation level. Within each panel, results for babble background at +5 and +10 dB signal-to-noise ratio, followed by cafeteria-noise background at +5 and +10 dB signal-to-noise ratio, are plotted along the x-axis as one moves from left to right.

(10)

none of the other variables (test materials, noise type, signal-to-noise ratio, or speech level) significantly af-fected the linear-minus-BILL difference scores, either individually or in combination.

Magnitude Estimation of Listening Effort

Test (MELE)

The MELE test required the listener to rate, from 0 to 100, the ease of listening to speech in cafeteria noise and in babble backgrounds. The same connected-speech passages used in the CST served as the speech stimu-lus. As with the NU-6 and CST tests, the MELE test was performed at +5 and +10 dB S/N in babble and caf-eteria noise, with the speech signal presented at 60 and 75 dB SPL at the dispensing visit (unaided) and the first and second evaluation visits (aided).

Figure 6 provides a summary of the MELE data for the 110 participants in this study with mild, moderate, or severe hearing loss, respectively. Table 6 provides the standard deviations corresponding to the mean MELE data appearing in Figure 6. Ratings, which ranged from 0 to 100, were repeated a total of four times per condi-tion, with the first rating discarded and the remaining three averaged for each subject. Preliminary analyses conducted on the linear-minus-BILL difference scores, as used in the analysis of the NU-6 and CST results, found significant heterogeneity of variance for the MELE difference scores. Consequently, analyses of variance were performed on the raw scores rather than differ-ence measures for the MELE test (variances of raw scores were found to be homogeneous across conditions). An analysis of variance performed on the MELE rat-ings, with hearing-aid setting (linear vs. BILL) as a vari-able, is summarized in Table 7. For the most part, this analysis indicated that there was no significant effect of the hearing-aid setting on MELE ratings. That is, the main effect of circuit setting (linear vs. BILL) was not significant and only one of the 15 interactions involving this variable was significant (hearing loss group x noise type x setting). The presence of this significant interac-tion involving hearing loss group, however, confounds a simple interpretation of the effect of hearing-aid setting. From visual inspection of the data in Figure 6, this three-way interaction involving the hearing-aid setting appears to be due to the higher MELE scores for the subjects with severe hearing loss listening in babble for the BILL hear-ing-aid setting and the reversal of this trend under the same conditions for the listeners with mild hearing loss. The presence of several significant two-way interactions (noise type x signal-to-noise ratio, noise type x presenta-tion level, and signal-to-noise ratio x presentapresenta-tion level), none involving the hearing-aid setting variable, illustrates the importance of assessing MELE across a range of lis-tening conditions.

Figure 5. Mean percent-correct scores on the CST for the linear (cross-hatched bars) and BILL (unfilled bars) settings of the hearing aid. Panels are organized from top to bottom according to hearing loss group such that participants with mild hearing loss are in the top two panels, those with moderate hearing loss are in the middle two panels, and those with severe hearing loss are in the bottom two panels. All left-hand panels contain data for the 60-dB-SPL presentation level, whereas all right-hand panels contain data for the 75-dB-SPL presentation level. Within each panel, results for babble background at +5 and +10 dB signal-to-noise ratio, followed by cafeteria-noise background at +5 and +10 dB signal-to-noise ratio, are plotted along the x-axis as one moves from left to right.

(11)

Table 3. Standard deviations for NU-6 percent-correct scores for participants with mild, moderate, or severe hearing loss.

Processing condition and presentation level

BILL Linear

Stimulus condition 60 dB SPL 75 dB SPL 60 dB SPL 75 dB SPL Mild hearing loss group

+5 dB S/N in Babble 15.75 15.84 14.71 15.51

+10 dB S/N in Babble 20.16 18.08 18.77 14.52

+5 dB S/N in Cafeteria Noise 17.81 16.07 16.15 17.22

+10 dB S/N in Cafeteria Noise 19.66 16.64 19.86 13.67 Moderate hearing loss group

+5 dB S/N in Babble 12.94 13.30 11.41 13.79

+10 dB S/N in Babble 18.00 18.77 18.93 17.84

+10 dB S/N in Cafeteria Noise 19.51 18.75 18.93 19.32 Severe hearing loss group

+5 dB S/N in Babble 4.67 5.96 11.10 6.96

+10 dB S/N in Babble 14.08 13.06 9.25 13.04

Table 4. Standard deviations for CST percent-correct scores for participants with mild, moderate, and severe hearing loss.

BILL Linear

+5 dB S/N in Babble 24.83 26.88 25.27 23.30

+10 dB S/N in Babble 24.62 22.91 21.67 21.10

+5 dB S/N in Babble 26.43 26.52 24.98 28.25

+10 dB S/N in Babble 27.76 28.99 27.79 29.32

+5 dB S/N in Babble 17.50 10.52 13.06 16.31

+10 dB S/N in Babble 19.97 23.30 22.11 26.51

Order Effect

The results were analyzed for effects of the order in which the hearing-aid setting was first dispensed to a participant. One might hypothesize that participants will

perform best with the hearing-aid setting (linear vs. BILL) to which they first became accustomed, in part because it was their first experience, but also because they wore the hearing aid with that setting for 8 weeks, versus 4

(12)

weeks for the second setting. To address this issue, six separate ANOVAs were performed: one for each of the three benefit measures (NU6, CST, and MELE) in each of two noise types (cafeteria and babble). A significant (p <. 01) main effect of the initial hearing-aid setting was not observed, and no significant interactions with the initial setting were obtained in any of these ANOVAs.

Hearing Aid Performance Inventory

(HAPI)

The HAPI questionnaire was used to identify the effect of the hearing aids on the participant’s communi-cation abilities in daily life. Each participant completed this questionnaire at the first and second evaluation visits, ranking the benefit provided by the hearing aid from Very Helpful (1) to Hinders Performance (5).

Table 8 provides a summary of the observed HAPI scores (means and standard deviations) for all hearing loss groups. Analyses using Wilcoxon Signed Ranks Tests for each participant group showed no significant differ-ence in HAPI scores for linear or BILL settings (Wilcoxon Z = –1.10, p = 0.27; Z = –1.59, p = 0.11; Z = –1.53, p = 0.12; for mild, moderate, and severe hearing-loss groups, respectively). The grand mean scores across all hearing loss groups were 2.10 for the BILL setting and 2.02 for the linear setting, indicating that both circuits were gen-erally considered to be “helpful” to the participants. (On the HAPI, a rating of 1 is associated with a label of “very helpful” and a rating of 2 with a label of “helpful.”) The difference in grand mean HAPI scores for the linear and BILL settings, collapsed across all three subject groups, however, approached statistical significance (Z = –2.36, p = 0.02).

Table 5. Summary of ANOVA results for BILL-linear differences for all 110 participants combined (all p values > .05).

Effect F value (df = 1, 109)

Constant (BILL-linear difference > or < 0) 1.07

Material (NU-6, CST) 0.30

Noise (cafeteria noise or babble) 0.10 SNR (signal-to-noise ratio, +5 or +10 dB) 0.66 Sound level (sound level, 60 or 75 dB SPL) 0.14

Material by noise 0.09

Material by SNR 0.48

Material by sound level 0.22

Noise by SNR 0.55

Noise by sound level 0.20

SNR by sound level 0.06

Material by noise by SNR 0.41

Material by noise by sound level 3.52 Material by SNR by sound level 1.43 Noise by SNR by sound level 0.86 Material by noise by SNR by sound level 0.02

Figure 6. Mean ratings on the MELE test for the linear (cross-hatched bars) and BILL (unfilled bars) settings of the hearing aid. Panels are organized from top to bottom according to hearing loss group such that participants with mild hearing loss are in the top two panels, those with moderate hearing loss are in the middle two panels, and those with severe hearing loss are in the bottom two panels. All left-hand panels contain data for the 60-dB-SPL presentation level, whereas all right-hand panels contain data for the 75-dB-SPL presentation level. Within each panel, results for babble background at +5 and +10 dB signal-to-noise ratio, followed by cafeteria-noise background at +5 and +10 dB signal-to-noise ratio, are plotted along the x-axis as one moves from left to right.

(13)

Paired-Comparison Test

After completion of the second round of objective and subjective measures of performance at the second evaluation visit, the participant was placed in the sound field and presented with continuous discourse to deter-mine participant preference for either the BILL or lin-ear setting. The test was performed at +5 and +10 dB S/N in babble and cafeteria noise, with the speech sig-nal presented at 60 and 75 dB SPL. During this test the audiologist switched the circuit back and forth from the BILL setting to the linear setting randomly. For each of the eight listening conditions (2 noise types x 2 S/N values x 2 sound levels), the participant chose the circuit he/she preferred. Each listening condition was repeated four times, and the mean preference rating for the final three ratings was computed (first rating discarded as practice).

For this task, the hearing-aid settings were rated on a scale of 1 to 6 by the participants, with ratings of 1, 2, and 3 representing varying degrees of preference for the BILL setting and ratings of 4, 5, and 6 representing varying degrees of preference for the linear setting. A mean value of 3.49 or below, therefore, meant that the participant preferred the BILL setting, and a value of 3.51 or above meant that the participant preferred the linear setting. The overall mean result for the paired-comparison test (3.57) showed no group preference for either the linear or BILL setting (z = 0.1472, p = 0.442).

Hearing-Aid Use

Table 9 provides a summary of the average reported hearing-aid wear times for participants with mild, mod-erate, and severe hearing loss. Participants with severe hearing loss reported wearing their hearing aids longer (about 9 hours/day) than participants in the mild and moderate groups (about 7 hours/day).

Group Data—Unaided vs. Aided

Performance

Figures 7 and 8 provide summaries of the mean data for the NU-6 and CST with regard to unaided and aided conditions for the 110 participants of this study with mild, moderate, or severe hearing loss, respectively. Table 10 provides the standard deviations for the unaided scores appearing in these two figures. Since aided performance with the linear and BILL circuits was found to be equiva-lent in the previous analyses, the linear setting was se-lected arbitrarily to represent aided listening, and the standard deviations for these data appeared previously in Tables 3 and 4. The less well-established MELE mea-sures are not considered in detail here as meamea-sures of benefit. Rather, the focus is on the more conventional measures of benefit (NU-6, CST, and HAPI).

When a repeated-measures ANOVA was calculated for the aided-minus-unaided differences collapsed across all three subject groups (for arcsine-transformed NU-6 and CST data), differences were found to be significantly Table 6. Standard deviations for MELE ratings for completed participants with mild, moderate, and severe

hearing loss.

BILL Linear

+5 dB S/N in Babble 23.43 23.82 22.92 27.23

+10 dB S/N in Babble 21.93 22.67 16.58 18.39

+5 dB S/N in Babble 23.36 26.15 24.87 26.69

+10 dB S/N in Babble 23.24 22.85 25.12 23.35

+5 dB S/N in Babble 28.17 31.88 31.36 30.71

+10 dB S/N in Babble 29.99 23.18 26.43 21.16

(14)

greater than zero (F = 146.13, df = 1,109, p < .01). In particular, aided scores were higher than unaided scores, which led to statistically significant benefit. To analyze the effects of each independent variable on the aided-minus-unaided difference scores, a second mixed-model ANOVA was performed. The results are summarized in Table 11. As can be seen in this table, main effects of test material (NU-6, CST) and noise type (cafeteria, babble) were significant (p <. 01) and unconfounded by

any interactions with other independent variables. Aided-minus-unaided differences were greater in cafeteria noise than in babble and for the CST than for the NU-6 test. In addition to these main effects, the three-way interaction of hearing-loss group, signal-to-noise ratio, and presen-tation level was also significant. From visual inspection of Figures 7 and 8, this three-way interaction appears to be attributable to the effect of signal-to-noise ratio on the aided-minus-unaided difference at various presentation levels across the three groups. In particular, benefit on both the NU-6 test (Figure 7) and the CST (Figure 8) increases as signal-to-noise ratio increases from +5 to +10 dB, except for the groups with mild or moderate hearing loss at the highest presentation level (75 dB). Finally, it should be noted that, given the lack of any significant differences in aided scores between the linear and BILL circuits described previously, significant benefit (aided scores minus unaided scores) exists for both circuits, even though the linear circuit was arbitrarily selected as the circuit for the aided condition in these analyses of the benefit provided by the hearing aid.

Individual Data

Figure 9 provides scatterplots of the individual data for all 110 participants on the NU-6 test. In the top panel, percent-correct scores for the linear setting of the hearing aid are plotted against the corresponding percent-correct score for the BILL setting. The solid lines represent the diagonal (equality of the two scores) and 95% critical-difference boundaries from Thornton and Raffin (1978) for a test consisting of 50 items. Note that the vast majority of the data points plotted fall within the 95% critical differences, with approximately equal numbers of points falling below the boundaries as above them.

In contrast, the bottom panel of Figure 9 shows the aided NU-6 scores for the linear condition plotted as a function of the unaided scores. Notice that a much greater number of data points falls above the upper 95%-criti-cal-difference boundary than below the lower boundary. Thus, these individual data are consistent with the analy-ses of the group results in that little difference was ob-served between the linear and BILL settings (top panel), Table 7. Summary of ANOVA for MELE results.

Effect F (df)

HL group (mild, moderate, severe) 2.22 (2,107) Noise (cafeteria noise, babble) 109.73 (1,107)*

HL group by noise 0.33 (2,107)

Setting (BILL, linear) 0.09 (1,107) HL group by setting 0.29 (2,107) Snr (+5, +10 dB) 234.66 (1,107)* HL group by snr 1.24 (2,107) Sndlev (60, 75 dB SPL) 5.92 (1,107) HL group by sndlev 3.84 (2,107) Noise by setting 1.44 (1,107)

HL group by noise by setting 7.77 (2,107)*

Noise by snr 38.66 (1,107)*

HL group by noise by snr 0.62 (2,107)

Noise by sndlev 16.50 (1,107)*

HL group by noise by sndlev 1.06 (2,107)

Setting by snr 0.01 (1,107)

HL group by setting by snr 0.36 (2,107)

Setting by sndlev 1.29 (1,107)

HL group by setting by sndlev 0.98 (2,107)

Snr by sndlev 9.45 (1,107)*

HL group by snr by sndlev 1.90 (2,107) Noise by setting by snr 1.86 (1,107) HL group by noise by setting by snr 1.66 (2,107) Noise by setting by sndlev 1.51 (1,107) HL group by noise by setting by sndlev 0.56 (2,107) Noise by snr by sndlev 5.73 (1,107) HL group by noise by snr by sndlev 0.60 (2,107) Setting by snr by sndlev 3.94 (1,107) HL group by setting by snr by sndlev 1.80 (2, 107) Noise by setting by snr by sndlev 0.11 (1,107) HL group by noise by setting by snr by sndlev 0.18 (2,107) *p < .01

Table 8. Summary of means and standard deviations (in parenthe-ses) for the Abbreviated Hearing Aid Performance Inventory (HAPI) for participants with mild, moderate, or severe hearing loss.

Hearing aid mode Hearing-loss group BILL setting Linear setting

Mild 2.21 (0.56) 2.16 (0.55)

Moderate 2.01 (0.53) 1.95 (0.47)

Severe 2.11 (0.50) 1.88 (0.49)

Table 9. Mean reported wearing time (in hours per day) for participants by hearing loss group for BILL and linear hearing-aid settings.

Hearing aid mode

Hearing-loss group BILL setting Linear setting

Mild 6.65 7.03

Moderate 6.77 7.31

(15)

Figure 7. Mean percent-correct scores on the NU-6 test for the unaided (unfilled bars) and aided (linear; cross-hatched bars) settings of the hearing aid. Panels are organized from top to bottom according to hearing loss group such that participants with mild hearing loss are in the top two panels, those with moderate hearing loss are in the middle two panels, and those with severe hearing loss are in the bottom two panels. All left-hand panels contain data for the 60-dB-SPL presentation level, whereas all right-hand panels contain data for the 75-dB-SPL presentation level. Within each panel, results for babble background at +5 and +10 dB signal-to-noise ratio, followed by cafeteria-noise back-ground at +5 and +10 dB signal-to-noise ratio, are plotted along the x-axis as one moves from left to right.

Figure 8. Mean percent-correct scores on the CST for the unaided (unfilled bars) and aided (linear; cross-hatched bars) settings of the hearing aid. Panels are organized from top to bottom according to hearing loss group such that participants with mild hearing loss are in the top two panels, those with moderate hearing loss are in middle the two panels, and those with severe hearing loss are in the bottom two panels. All left-hand panels contain data for the 60-dB-SPL presentation level, whereas all right-hand panels contain data for the 75-dB-SPL presentation level. Within each panel, results for babble background at +5 and +10 dB signal-to-noise ratio, followed by cafeteria-signal-to-noise background at +5 and +10 dB signal-to-noise ratio, are plotted along the x-axis as one moves from left to right.

(16)

but many individuals demonstrated significantly higher aided scores than unaided scores (bottom panel). The latter conclusion appears to be particularly true for un-aided NU-6 scores less than or equal to 40%.

Figure 10 provides a comparable set of scatterplots for the CST percent-correct scores. The top panel again shows comparison of aided scores for the linear and BILL conditions, whereas the bottom panel compares aided (linear) to unaided performance. Although there is more scatter than in the NU-6 data, the same trends are apparent. Specifically, the two aided conditions do not appear to differ systematically (top panel), whereas aided scores are frequently significantly greater than unaided scores (bottom panel). Again, because there were no significant differences between aided scores for the linear and BILL circuits, scatterplots equiva-lent to those in Figures 8 and 9 would have resulted if the results from the BILL circuit, rather than the lin-ear circuit, had been selected to represent aided lis-tening in these figures.

A voluminous amount of data was gathered in the course of this study—not only in terms of sample size, but in terms of the number of outcome measures ob-tained. Speech-recognition performance alone was mea-sured in 16 different conditions. How do these data com-pare with those in the literature? Unfortunately, there are no norms available for these measures under the same combinations of noise type, speech level, and sig-nal-to-noise ratio. To gain some insight into the valid-ity of these results, Speech Intelligibilvalid-ity Index (SII;

Table 10. Standard deviations for unaided NU-6 and unaided CST scores for participants with mild, moderate, and severe hearing loss.

Test material and presentation level

NU-6 NU-6 CST CST

+5 dB S/N in Babble 13.80 15.88 21.25 25.53

+10 dB S/N in Babble 19.49 16.64 24.80 26.73

+5 dB S/N in Babble 9.03 16.78 18.84 25.05

+10 dB S/N in Babble 15.48 20.34 27.77 32.22

+5 dB S/N in Babble 2.50 4.66 6.89 10.83

+10 dB S/N in Babble 2.50 7.26 6.02 20.74

ANSI S3.79—draft V3.1) calculations were performed. The SII is essentially an updated version of the Articu-lation Index (AI) and is sensitive to factors manipu-lated in this study, including noise spectrum, noise level, speech level, hearing loss, and hearing-aid gain. SII calculations were performed only for the babble background—for which it could be reasonably assumed that the speech and babble had the same long-term average spectrum. Moreover, it was assumed that the average speech spectrum and the normal vocal effort included in the SII standard were reasonably repre-sentative of the 60-dB SPL speech level for both the NU-6 and CST talkers. Figures 11 and 12 depict the mean results from this study for the NU-6 test and the CST, respectively, plotted as a function of the SII. Best-fitting polynomials fit to the data in each figure (solid lines) indicate that the data are consistent with SII pre-dictions in that performance increases monotonically with the SII. In the absence of adequate norms, this analysis assists in establishing the validity of these data.

Discussion

No statistically significant differences in perfor-mance were found between the BILL-processing and lin-ear settings for any of the outcome measures used in this clinical investigation. This is consistent with a num-ber of previous reports from smaller scale laboratory stud-ies (see review by Fabry, 1991). Essentially, in situations in which the speech and noise are broadband stimuli, BILL

(17)

processing will provide level-dependent changes in gain that affect both the speech and the noise equivalently and do not improve the signal-to-noise ratio in any frequency region. Were the noise predominantly low-frequency in nature, however, one might expect different results.

Perhaps of greater importance, however, is the con-sistent observation, across several outcome measures, that there was significant benefit provided by these hear-ing aids when operathear-ing in either the linear or BILL-processing mode compared to unaided listening. For both the NU-6 and CST measures, for example, the differ-ence between unaided and aided scores was significantly greater than zero. Moreover, the participants reported using their hearing aids 7–9 hours per day, on average, supporting a self-perceived benefit associated with the use of their hearing aids. That is, they would probably not wear the hearing aids much if they failed to per-ceive significant benefit. Recall, however, that partici-pants were instructed to wear their hearing aids at least 5 hours per day, 5 days per week, as participants in this Table 11. Summary of ANOVA on aided-unaided differences.

Effect F (df)

HL group (mild, moderate, severe) 8.42 (2,107)* Material (NU-6, CST) 14.07 (1,107)*

HL group by material 0.73 (2,107)

Noise (cafeteria noise, babble) 49.40 (1,107)*

HL group by noise 4.17 (2,107) Snr (+5, +10 dB) 72.13 (1,107)* HL group by snr 1.18 (2,107) Sndlev (60, 75 dB SPL) 71.27 (1,107)* HL group by sndlev 12.80 (2,107)* Material by noise 0.28 (1,107)

HL group by material by noise 0.84 (2,107)

Material by snr 0.51 (1,107)

HL group by material by snr 0.34 (2,107)

Material by sndlev 0.83 (1,107)

HL group by material by sndlev 2.62 (2,107)

Noise by snr 4.35 (1,107)

HL group by noise by snr 1.38 (2,107)

Noise by sndlev 2.28 (1,107)

HL group by noise by sndlev 4.45 (2,107)

Snr by sndlev 9.31 (1,107)*

HL group by snr by sndlev 10.80 (2,107)* Material by noise by snr 0.65 (1,107) HL group by material by noise by snr 0.10 (2,107) Material by noise by sndlev 1.68 (1,107) HL group by material by noise by sndlev 1.43 (2,107) Material by snr by sndlev 0.25 (1,107) HL group by material by snr by sndlev 0.48 (2,107) Noise by snr by sndlev 2.39 (1,107) HL group by noise by snr by sndlev 0.09 (2,107) Material by noise by snr by sndlev 0.30 (1,107) HL group by material by noise by snr by sndlev 1.12 (2,107) *p < .01

Figure 9. Scatterplots of NU-6 percent-correct scores for individual participants. The top panel plots the aided score for the linear setting against the aided score for the BILL setting, whereas the bottom panel plots aided score for the linear setting against the unaided score. Triangles represent scores obtained in babble, whereas squares represent those obtained in cafeteria noise. There are a total of 880 datapoints in each panel (440 for babble and 440 for cafeteria noise) for the 110 participants in the study. Solid lines in each panel represent diagonal and 95% critical-difference bound-aries above and below the diagonal (Thornton & Raffin, 1978).

study. The extent to which these instructions could have biased the participants’ estimates of hearing-aid use is unclear. The validity of the estimates of hearing-aid use is reinforced, however, by the fact that participants with severe hearing loss reported greater daily usage than those with milder hearing loss and that many participants

(18)

reported daily usage that was either greater or less than the minimum usage targeted in the audiologist’s instruc-tions. Finally, the HAPI provided additional support for significant benefit in that, on average, participants rated their hearing aids as being “helpful.” Thus, this large-scale study documents significant benefit provided by well-fit Figure 10. Scatterplots of CST percent-correct scores for individual participants. The top panel plots the aided score for the linear setting against the aided score for the BILL setting, whereas the bottom panel plots the aided score for the linear setting against the unaided score. Triangles represent scores obtained in babble, whereas squares represent those obtained in cafeteria noise. There are a total of 880 datapoints in each panel (440 for babble and 440 for cafeteria noise) for the 110 participants in this study. Solid lines in each panel represent diagonal and 95% critical-difference bound-aries above and below the diagonal (Thornton & Raffin, 1978).

Figure 11. Scatterplot of mean NU-6 scores obtained in babble for each hearing-loss subgroup plotted as a function of the Speech Intelligibility Index (SII). All unfilled symbols represent unaided listening conditions, and filled symbols represent aided listening. The solid line represents the best-fitting 3rd-order polynomial, whereas the dotted lines represent 95% prediction intervals, or population confidence intervals, above and below the best-fitting polynomial. The best-fitting polynomial accounts for 85.2% of the variance.

linear amplification, including significant improvements in speech-recognition performance in noise. Moreover, equivalent benefit was measured for the linear and BILL-processing circuits. Although BILL BILL-processing did not re-sult in aided performance exceeding that achieved with well-fit linear amplification, both circuits provided signifi-cant, measurable, and equivalent benefit across a wide range of measures and conditions.

It is important to recognize the importance of the term well-fit in the preceding sentence. The real-ear gain measured in these participants was in close agreement with that prescribed by the NAL-R procedure out to a frequency of 4000 Hz, thereby making a wide portion of the speech spectrum audible. The importance of the re-gion from 2000 to 4000 Hz to the understanding of speech in noise by hearing-impaired listeners has been empha-sized previously (e.g., Lee & Humes, 1993). Moreover, output limiting was adjusted so that SSPL90 was main-tained below an uncomfortable level for each patient. Although this may not have had a direct bearing on the measures of objective benefit, it most likely had a sig-nificant impact on the amount of time the hearing aids were worn by the participants.

A wide range of listening conditions was sampled in this study: a total of 16 measures of speech recognition in noise, resulting in 16 objective measures of benefit

(19)

per participant. When attempting to determine indi-vidual differences in the benefit provided by the hear-ing aids, the large number of measures poses some chal-lenges in defining benefit for individual participants. For example, should one consider a participant to have con-clusively demonstrated significant benefit only when showing significant benefit on all 16 speech-recognition measures or is 8 out of 16 adequate? Perhaps one should only care that the participant showed significant im-provement on any one condition of the 16 sampled. That one condition, for instance, could prove to be a highly important and frequently encountered listening situa-tion for the participant.

Our approach to determining the number of individual participants who demonstrated significant benefit was the following. First, the data were grouped by test material (NU-6, CST), hearing-loss category (mild, moderate, se-vere), and speech presentation level (60, 75 dB SPL). For each of these subgroups, a total of four listening condi-tions remained (two SNRs and two noise types). The per-centage of participants showing significant (p < .05) ben-efit on at least two of these four conditions was then determined. To evaluate significance, the individual scores were converted to rationalized arcsine units (RAUs; Studebaker, 1985), which enable the use of a constant-RAU difference score as a criterion for significance. For

the NU-6 test, a RAU difference of 18.1 was used to es-tablish significance (Studebaker, 1985), whereas Cox et al. (1988) recommend a 15.5 RAU difference-score crite-rion for the CST. Table 12 provides the results of this analy-sis. Each entry in the table represents the percentage of participants showing significant improvements for two of the four conditions included in each subset of conditions represented in the table. It is apparent from the values in this table that about 60–80% of the participants showed significant improvement in at least half of the conditions when speech is presented at normal conversational levels (60 dB). This statement applies across test materials and hearing-loss categories, although there appear to be some systematic variations in this percentage with hearing-loss category. At the raised speech level (75 dB), a lower per-centage of each hearing-loss group demonstrated signifi-cant improvement than at normal conversational levels. At the higher speech level, moreover, the effects of speech material (NU-6, CST) and hearing-loss category on the observed percentages are more apparent. It is probably not too surprising that fewer participants demonstrated significant benefit for raised speech levels than for nor-mal levels. Recall that benefit is the relative difference between unaided and aided performance. At higher speech levels, listeners—especially those having milder impair-ments—will obtain higher unaided scores and conse-quently will have less room for improvement when the hearing aid is worn. Even so, approximately 20–60% of listeners with moderate or severe hearing loss demon-strated significant benefit for speech recognition in noise in this study at the higher speech presentation level.

Acknowledgments

This work was supported, in part, by research contracts provided by Dahlberg, Inc. to the investigators at each of the clinical sites. The first author was not one of the clinical investigators for the study, but was hired by the study sponsor as a paid consultant while the study was underway to oversee the data collection, analyses, and internal reporting. The authors would like to express appreciation for their support to all those involved in this project at Dahlberg, Inc.—especially Melanie Raska and Tom Scheller (presently with another hearing-aid company); those at Bausch & Lomb Incorpo-rated—especially Eric Ankerud, Heather Bornemann, and Tom Crescuillo; and those involved at the participating clinical sites—especially Tara Thomas. In addition, two individuals at the University of Iowa deserve special mention for their significant contributions to this project. First, Don Schum, presently working for another hearing-aid manufac-turer, played a major role in the design and development of the clinical protocol used in this study. Second, Aaron Parkinson, who assumed Don’s on-site responsibilities at the University of Iowa when Don left Iowa. Finally, preparation of this work for publication was supported, in part, by a grant from the National Institute on Aging to the first author and, in part, by the Retirement Research Foundation.

Figure 12. Scatterplot of mean CST scores obtained in babble for each hearing-loss subgroup plotted as a function of the Speech Intelligibility Index (SII). All unfilled symbols represent unaided listening conditions, and filled symbols represent aided listening. The solid line represents the best-fitting 2nd-order polynomial, whereas the dotted lines represent 95% prediction intervals, or population confidence intervals, above and below the best-fitting polynomial. The best-fitting polynomial accounts for 80.6% of the variance.