Spatial Benefit of Bilateral Hearing Aids

Participants

Twenty-one^* older adults (mean age, 75.3 yr; range, 69–83 yr) with adult-onset sensorineural hearing loss and normal immittance measures participated in the study. Mean pure-tone thresholds in quiet (±1 SD) for each ear measured under headphones (ANSI 2004) are shown in Figure 1. For all subjects, interaural differences, averaged across frequency from 0.25 to 6.0 kHz, were ≤11 dB; for 13 subjects, differences were ≤5 dB. Seven of the 21 subjects were experienced hearing aid users, that is, they used hearing aids with various types of compression circuits for at least 1 yr. In addition to being compensated on an hourly basis for their participation, subjects were offered the opportunity to keep their hearing aids (provided by Starkey Laboratories at no cost) at the end of the study.

Subjects had no previous experience with the speech materials used in the experiment or with the listening tasks. During training, the examiner provided extensive feedback to the subject until the subject understood the task and provided consistent results. For the simple task of measuring detection thresholds with narrowband noise, this was accomplished after four to five measurements for most participants. Practice sentences provided with the speech materials were used to train participants. The narrowest and widest of three low-pass-filter cutoffs were included during this practice period in both unaided and aided conditions. After approximately 1 hr of practice with the various tasks, data collection was completed in two or three 2 hr sessions (see Table 1).

Apparatus and Stimuli

Procedures

Data Analyses

Outcome measures included the following values: (1) observed thresholds for HINT sentences, hearing aid benefit (defined as unaided threshold minus aided threshold), and spatial benefit (defined as threshold with babble at 0° minus threshold with babble at 90°); (2) AAI-predicted thresholds for HINT sentences, hearing aid benefit, and spatial benefit; (3) observed ANL, hearing aid benefit for ANL (defined as unaided ANL minus aided ANL), and spatial benefit for ANL (defined as ANL with babble at 0° minus ANL with babble at 90°); and (4) questionnaire responses and cognitive measures. Differences in threshold and benefit measures were assessed by repeated-measures analysis of variance. Associations among subjective and cognitive measures and objective speech-recognition measures were assessed by Pearson product moment correlations. For all analyses, effects were significant with p < 0.05.

Thresholds for Speech in Babble

Acceptable Noise Level

Figure 6 shows mean ANL values; lower ANLs indicate that subjects were willing to tolerate more noise (babble). ANL decreased significantly [F(1,20) = 9.70; p = 0.006] with hearing aids (i.e., subjects tolerated less favorable SNRs), as seen by comparing solid and striped bars for both the 0° and 90° conditions. In addition, ANL decreased significantly when speech and babble were spatially separated [F(1,20) = 8.79; p = 0.008], as seen by comparing the 0° and 90° bars for both the unaided and aided conditions. Thus, in this group of listeners, ANL was not a characteristic of the individual listener but varied with amplification and noise-source location. No significant interaction was found between ANL and hearing aid condition or noise-source location, although there was a trend for more benefit for ANL in the aided, spatially separated condition.

Significant positive correlations were found between unaided thresholds for HINT sentences and unaided ANL with babble at 0°, as shown in Figure 7. To simplify comparisons, thresholds for HINT sentences were converted to SNR values (HINT threshold minus the 65 dB babble level). As with ANL, lower SNRs signify better performance. Thus, listeners who tolerated less favorable SNRs at threshold for unaided speech also tolerated less favorable SNRs for the unaided ANL. In addition, although the two measures were significantly correlated, the range of SNRs for thresholds for HINT sentences was narrower (−0.3 to 6.0 dB) than for ANL (−1.0 to 14.0 dB), indicating larger individual differences for ANL than for thresholds for speech. Large individual differences in the ANL were reported by Rogers et al. (2003), although speech recognition was not measured in that study.

Significant positive correlations were also found between aided spatial benefit for HINT sentences and aided spatial benefit for ANL, as shown in Figure 8. That is, listeners who achieved more spatial benefit with their hearing aids for speech recognition tolerated less favorable SNRs with their hearing aids when speech and babble were spatially separated. Significant positive correlations between ANL and speech recognition supports an association between the two measures, which is not consistent with previous results (Nabelek et al. 2004, 2006).

Subjective and Cognitive Measures

We hypothesized that self-report and cognitive measures would relate to the various objective measures of speech recognition. For example, because the SSQ was designed to assess binaural listening situations, subjects with more spatial benefit measured objectively were expected to have higher SSQ scores, reflecting a better outcome. In addition, subjects with more hearing aid benefit were expected to have higher satisfaction ratings on both the GHABP and the SADL. However, few correlations between objective and subjective measures reached statistical significance. One example of a significant correlation is shown in Figure 9. A significant positive correlation was found between aided benefit for ANL and hearing aid use. Here, hearing aid use was determined from the GHABP, where listeners reported the proportion of time they wore their hearing aids in several situations. Those individuals who had more hearing aid benefit, as determined by the ANL (i.e., tolerated more babble aided than unaided), also reported more hearing aid use. Interestingly, there was also a significant positive correlation (r = 0.456, not shown) between the unaided ANL and hearing aid use; that is, listeners with higher ANL values without hearing aids (i.e., tolerated less babble) also reported more hearing aid use. This is inconsistent with earlier findings of Nabelek et al. (2004, 2006) who reported that listeners who tolerated more noise without hearing aids reported more hearing aid use.

Significant positive correlations were also found between the aided Speech-in-Noise factor (a subscale of the speech section) from the SSQ and spatial benefit of hearing aids for speech at 1.8 kHz (r = 0.433) and 3.6 kHz (r = 0.482). Two questions that are components of this factor are, “You are in a group of about five people in a busy restaurant. You cannot see everyone else in the group. Can you follow the conversation?” and “You are talking with one other person and there is a TV on in the same room. Without turning the TV down, can you follow what the person you’re talking to says?” Significant positive correlations revealed that as aided SSQ score increased, spatial benefit of hearing aids (i.e., aided spatial benefit minus unaided spatial benefit) increased, reflecting a greater ability when listening in noise while wearing hearing aids. This finding is consistent with that reported by Gatehouse and Akeroyd (2006), who observed a significant correlation between the Speech-in-Noise subscale and a dynamic psychoacoustic measure of masking level difference in which the interaural correlation of the masking noise varied as a function of time.

Hearing Aid Benefit

The current study examined the benefit attributable to bilateral hearing aids with and without spatial separation as a function of low-pass cutoff frequency of speech and babble. In addition, thresholds for HINT sentences, hearing aid benefit, and spatial benefit predicted by the AAI provided a means to determine the extent to which hearing aids improved speech audibility and listeners benefited from improved speech audibility. Although mean speech thresholds were significantly lower aided than unaided, hearing aid benefit was relatively small. As discussed earlier, this may be attributed to (1) less-than-optimal improvements in speech audibility provided by amplification or (2) subjects’ inability to benefit from improved speech audibility. Three approaches were taken to differentiate these two explanations for individual subjects (analyses of aided spectra, predicted speech thresholds, and observed–predicted differences in speech thresholds).

First, calculations made using ear-canal recordings of unaided and aided speech and babble spectra showed that, for many subjects, quiet thresholds in the higher frequencies rendered portions of the babble spectrum inaudible, even with amplification. Specifically, the level of the aided babble spectrum in the 0° condition at the highest cutoff frequency (5.6 kHz) was lower than quiet thresholds for 9 of 21 subjects at 3.0 kHz and for 15 of 21 subjects at 4.0 kHz. As a consequence, speech audibility for these subjects was not restored with hearing aids across the full bandwidth of speech (i.e., amplification did not fully compensate for their increased thresholds). Limited high-frequency speech audibility likely resulted in higher thresholds for speech (more favorable SNRs) compared with normal subjects (Lee & Humes 1993; Van Tasell & Yanz 1987). This finding is consistent with the significant positive correlation found between observed aided HINT thresholds and high-frequency hearing loss (Fig. 4, top panel) and is also in agreement with many studies (described earlier) that report reduced aided speech recognition in noise for individuals with high-frequency hearing loss. Thus, these results support the conclusion that less-than-optimal speech audibility provided by amplification contributed to higher aided thresholds for speech and limited hearing aid benefit. This lack of audible speech across bandwidth may have been a consequence of the relationship between NAL-NL1–prescribed hearing aid gain and magnitude of high-frequency hearing loss. That is, as hearing loss increased, the amount of prescribed gain did not increase accordingly. Although improving high-frequency audibility is a main goal of amplification, there is generally a tradeoff between optimization of speech audibility and subjective quality judgments of the hearing aid user (Byrne 1986; Byrne & Tonisson 1976).

Second, aided thresholds for speech predicted from AAI values can also help determine if individual subjects were provided with adequate speech audibility by their hearing aids and were unable to use the newly provided information or if amplification was inadequate. Mean predicted thresholds in Figure 3 (top panel) showing significantly lower speech thresholds when aided than unaided demonstrate that, on average, hearing aids provided at least some increase in audible speech. For individual subjects, if hearing aids provided audibility across the full bandwidth of speech, no significant correlation between predicted HINT thresholds and high-frequency thresholds would be observed. However, a significant positive correlation was found between predicted aided HINT thresholds and high-frequency hearing loss (Fig. 4, middle panel), consistent with the finding (described above) that amplification did not fully compensate for increased thresholds for all subjects.

Third, comparing observed and predicted aided thresholds for speech reflects the extent to which aided HINT thresholds were a result of limited speech audibility. That is, if poor speech thresholds were entirely attributed to reduced audibility resulting from increased thresholds and less-than-optimal amplification, observed speech thresholds would be equivalent to predicted thresholds. However, mean results in Figure 3 showing aided speech thresholds that were poorer than predicted (top panel) and smaller-than-predicted hearing aid benefit (bottom panel) suggest that, on average, these subjects with hearing loss did not fully benefit from the increased audibility provided by their hearing aids. These results are consistent with the positive correlations found between aided observed–predicted differences and high-frequency thresholds (Fig. 4, bottom panel). That is, subjects with more high-frequency hearing loss required more favorable SNRs than needed to compensate for their increased thresholds. Taken together, these results suggest that: (1) hearing aids did not provide optimal audibility across the full speech bandwidth for all subjects, (2) factors beyond inadequate aided speech audibility resulted in higher speech thresholds, and (3) improvement in speech information provided by amplification declined with increasing high-frequency hearing loss.

Several factors may have contributed to smaller-than-predicted hearing aid benefit. First, listeners with hearing loss may have higher masked thresholds in background noise compared with listeners with normal hearing (Dubno & Ahlstrom 1995a,b; Hornsby & Ricketts 2003). Indeed, in these previous studies, AI predictions were more accurate when actual masked thresholds were used to determine speech audibility, rather than relying solely on expected masked thresholds. Higher-than-normal masked thresholds of older listeners with hearing loss may reduce the amount of audible speech and account for poorer-than-normal speech thresholds. Here, because of the nature of the experimental design, masked thresholds were not measured and, therefore, AAI predictions were based on quiet thresholds. From data obtained in a recent study conducted in our laboratory (Horwitz et al. 2008), excess masking was estimated for 10 of the subjects in the current study who listened to speech and noise that had been shaped with a gain-frequency response similar to NAL-NL1. Excess masking was defined as the difference between thresholds measured in the NAL-NL1 speech-shaped noise and thresholds predicted in that same noise using normal critical ratios. A significant positive correlation (r = 0.644) was found in that listeners with higher-than-predicted masked thresholds had poorer-than-predicted aided HINT thresholds. Furthermore, as part of the protocol for an ongoing study of age-related hearing loss conducted at the Medical University of South Carolina, masked thresholds were measured in the presence of a high-level noise low-pass filtered at 0.5 kHz. For the listeners in the current study, masked thresholds at 1.5 kHz were significantly (r = 0.565) correlated with aided HINT thresholds for speech and babble filtered at 5.6 kHz. That is, subjects with more upward spread of masking tended to have poorer aided speech thresholds in babble. A partial correlation between quiet thresholds at 1.5 kHz and HINT threshold was not statistically significant, suggesting that quiet thresholds did not affect this result. Because no correction was made for spread of masking in the AAI calculations, aided audibility may have been overestimated, resulting in lower predicted speech thresholds and poorer-than-predicted speech thresholds and aided benefit.

Second, signal distortion in the hearing aids, in the form of circuit noise and nonlinear distortion, could reduce the positive effects of amplification and reduce speech recognition (Kates & Arehart 2005). Coherence measurements are becoming important in evaluating hearing aids and measurement procedures have been standardized (ANSI 1992). Several recent studies have attempted to quantify effects of noise and various types of nonlinear hearing aid distortions on speech recognition and sound quality (Arehart et al. 2007; Kates & Arehart 2005). Modifications to the Speech Intelligibility Index (ANSI 1997) to include a measure of the signal-to-distortion ratio calculated from coherence measures may provide a useful method for estimating the effects of distortion and noise on speech recognition (Kates & Arehart 2005). Nevertheless, additional measures suggested that the hearing aids in the current study did not add an appreciable amount of distortion, so that this factor may have had a negligible effect.

Finally, those subjects whose performance was poorer than predicted by the AAI may have additional peripheral and central impairments that may not be corrected with amplification and may result in more limited benefit of hearing aids. For instance, the consequences of high-frequency hearing loss may not be limited to a loss of high-frequency speech audibility, given the importance of the base of the cochlea to the encoding of lower frequency information (Horwitz et al. 2002; Joris et al. 1994; Kiang & Moxon 1974). Accordingly, providing amplified high-frequency speech would not necessarily restore the contribution of the base of the cochlea for lower-frequency cues. Although no significant correlations were observed between a cognitive measure and speech thresholds or hearing aid benefit, several studies of speech recognition in older adults in which background noise or reverberation were used have shown age-related declines in auditory behavior that may result from cognitive deficits. For example, Humes (2007) reported that differences in hearing aid benefit among older adults may result from aging, cognitive, or central factors once audibility is restored through amplification.

Although improvements in speech recognition with hearing aids were small, it should be noted that scores for HINT sentences increase rapidly with level, with slopes of psychometric functions ranging from 8.9%/dB (Nilsson et al. 1997) to 11.8%/dB (Eisenberg et al. 1998). Taking the average of these two slopes as an estimate, even a seemingly small increase in hearing aid benefit of 1.0 dB corresponds to ~10% change in sentence recognition score.

Spatial Benefit

There is an extensive body of research demonstrating the benefit to speech recognition of spatially separating speech and noise sources (Bronkhorst & Plomp 1988; Levitt & Rabiner 1967; Plomp 1976; Zurek 1993a). Although the issue remains unresolved, several studies have shown objective and subjective benefit of providing bilateral amplification (Byrne 1981; Holmes 2003; Noble & Gatehouse 2006; Silman et al. 1984; Simon 2005). Recent research has begun to focus on quantifying and predicting an individual listener’s ability to make use of spatial cues provided by bilateral hearing aids, in part because of the increased use of directional microphones (Dhar et al. 2004; Hornsby & Ricketts 2007; Ricketts et al. 2005). In the current experiment, unaided spatial benefit predicted by the AAI was expected to be very small, assuming that subjects’ high-frequency hearing loss would reduce the potential benefit of an improved SNR at higher frequencies because of head shadow. That is, simple estimates of audibility (AAI) would not take into account interaural difference cues that can provide a sizable spatial benefit, especially at low frequencies.

To the extent that amplification increased the audibility of high-frequency speech and babble, observed and predicted aided spatial benefit was expected to be larger than unaided spatial benefit. That is, it was anticipated that the increase in speech and babble level because of amplification and the reduction in the level of the babble in the far ear would result in an overall improvement in audibility of high-frequency speech. This was not the case. As noted in the discussion of hearing aid benefit, high-frequency amplification was limited for some subjects. In those cases, the measured improvement in audibility because of head shadow would be negligible, resulting in similar unaided and aided mean predicted spatial benefit.

Observed spatial benefit was significantly greater than predicted, suggesting that listeners were able to take advantage of interaural level and time difference cues to improve speech recognition in babble. This improvement increased when amplification was provided, especially for the two higher cutoff frequencies. Although interaural time cues are carried more efficiently by low frequencies, these cues are present in the envelope of speech and therefore may be conveyed across the entire frequency range (Henning 1974). Given the absence of predicted spatial benefit, the observed improvement in thresholds with spatial separation was likely related to the use of binaural difference cues, rather than simply an improvement in SNR because of head shadow.

The amount of unaided spatial benefit observed in this study was similar to that reported by Dubno et al. (2002) for a group of older listeners with hearing loss, although background noise and cutoff frequencies were somewhat different between the studies. Relative to younger subjects, less spatial benefit was observed for older subjects with hearing loss. For unfiltered speech and noise, unaided spatial benefit obtained by subjects with hearing loss in the current study (3.8 dB) was less than obtained by subjects with normal hearing in the previous study (5.3 dB). With amplification, spatial benefit increased to 4.7 dB, more similar to results for listeners with normal hearing.

Interestingly, for the lowest cutoff frequency (1.8 kHz), more spatial benefit was seen unaided than aided. This resulted from better unaided than aided HINT thresholds at 1.8 kHz in the spatially separated condition. This finding is consistent with previous results (Dubno et al. 2002), which demonstrated that listeners with hearing loss rely more on low-frequency timing cues than high-frequency level cues, most likely related to the configuration of their hearing loss. Some studies have shown that immediately after hearing aid fitting, bilateral hearing aids have a negative effect on the use of binaural cues (Drennan et al. 2005; Van den Bogaert et al. 2006). Furthermore, Noble and Byrne (1990, 1991) reported that listeners performed better on a localization task with their own hearing aids compared with novel hearing aids. Therefore, more unaided than aided spatial benefit for low-pass filtered speech and babble may suggest that subjects are more efficient at using low-frequency binaural cues to which they have become accustomed in frequency regions for which their hearing is best.

Acceptable Noise Level

It has been reported that the ANL is a feature inherent to an individual listener and is a predictor of hearing aid use (Nabelek et al. 2004, 2006). In previous studies using SPIN sentences, no correlation was found between speech-recognition scores and ANL (Crowley & Nabelek 1996; Nabelek et al. 2004, 2006). Additionally, ANL did not change significantly with amplification, although SPIN scores were better aided than unaided. These findings led to the conclusion that acceptance of background noise and speech-recognition scores obtained in background noise were different measures of hearing aid improvement and outcome (Nabelek et al. 2006). In contrast to those results, significant correlations between ANL and speech recognition in babble were observed in the current study, which suggests an association between the two measures.

In contrast to previous studies, ANL improved significantly with hearing aids, similar to the aided improvement in ANL reported by Mueller et al. (2006) for hearing aids with digital noise reduction. In the current study, ANL improved significantly when aided speech and babble were spatially separated. That is, more babble was tolerated with the babble at the side than in front of the listener. Similarly, in a study designed to assess the benefit of directional hearing aids, Freyaldenhoven et al. (2005) reported that ANL spatial benefit (defined as omni-directional ANL minus directional ANL at 180° azimuth) was observed and was of similar magnitude to more traditional measures of hearing aid directional benefit, such as masked speech-recognition thresholds and front-to-back ratios. These authors concluded that ANL was a valid measure for assessing directional benefit in hearing aids. Although differences in speech materials and hearing aid parameters may account for some differences among the studies, additional research is required to determine the role of noise tolerance in predicting hearing aid outcomes.

Subjective Measures

Few significant correlations were observed between self-report measures and objective measures of speech recognition. This was not entirely unexpected given the complexities associated with hearing aid outcomes and the multiple factors that influence hearing aid benefit (Saunders & Cienkowski 2002). Several studies have examined self-report outcome measures for individuals using one hearing aid compared with two (see Noble 2006 for a review). In many of these studies, the self-report measures assessed only a limited range of listening conditions, which may have underestimated the benefit of bilateral amplification (Gatehouse & Akeroyd 2006; Noble & Gatehouse 2006). It has been suggested that the lack of significant correlations between objective psychoacoustic measures and subjective self-report measures reflects the inherent differences between traditional “static” laboratory measures in which reverberation and head movement are controlled and more dynamic real-world listening environments (Gatehouse & Akeroyd 2006; Mencher & Davis 2006). Here, the significant positive correlations found between aided spatial benefit and a subscale of the SSQ supported an association between objective and subjective binaural performance. This is consistent with positive correlations observed by Mendel (2007) between aided HINT thresholds in spatially separated conditions and several categories of the Hearing Aid Performance Inventory (Walden et al. 1984).

Furthermore, the lack of association between cognitive ability and objective speech measures may also relate to the laboratory environment, as well as the hearing aid time constants used in this study. Gatehouse et al. (2003) observed an interaction between cognitive ability, temporal characteristics of background noise, and hearing aid time constants. Those listeners with hearing loss with greater cognitive ability derived more benefit in a temporally modulated background noise when fast time constants were used. In the current study, the background noise was a 12-talker babble and the hearing aids had very slow time constants. Additionally, using a working-memory test and a test of verbal information-processing speed to categorize cognitive function, Lunner (2003) observed a correlation between cognitive function and a speech-recognition-in-noise measure with and without hearing aids. Similar to the present findings, however, no significant correlation was found between the cognitive variables and hearing aid benefit.

With the role of subjective outcome measures of hearing aid benefit receiving increased attention in recent years, especially with regard to user benefit claims (Humes et al. 2001; Humes 2002; Saunders et al. 2005), further research is warranted. More complex laboratory tasks that involve dynamic listening environments, and high stimulus uncertainty, and demand focused attention may better correlate with self-reported benefit of amplified speech in adverse listening conditions often found in everyday communication.

Study Limitations

In this experiment, observed thresholds for HINT sentences, hearing aid benefit, and spatial benefit were compared with predicted measures using the AAI. Although the AAI has been used in other studies with hearing aids and with laboratory simulations of amplification with headphones (Souza & Turner 1999; Stelmachowicz et al. 1994; Stelmachowicz et al. 2002), certain assumptions and measurement techniques were unique to the current study, which may have influenced the predictions.

For example, although the hearing aids provided to our subjects were WDRC instruments, it was assumed that the very slow time constants allowed the hearing aid to perform like a linear hearing aid during speech-in-babble testing and probe microphone recordings. In an attempt to verify this assumption, several subjects returned to the laboratory after the completion of the study. Their hearing aids were reprogrammed to provide the same amount of gain regardless of input level (i.e., linear gain). Thresholds for speech in babble were measured for the 0° low-pass-filter conditions in both the linear and the original WDRC gain settings. Average HINT thresholds for the two hearing aid settings were within 1 dB and did not differ significantly. This supports the assumption that hearing aid gain in the presence of fixed-level babble did not change once the compression circuit was engaged and stabilized.

Assumptions about the hearing aid compression parameters and changes in gain could also have affected AAI values for the spatially separated conditions. With the babble at 90°, the reduced level of the babble in the far ear because of head shadow may have resulted in an increase in gain in the hearing aid in the far ear. Similarly, the increased level of the babble in the near ear because of boost may have resulted in a decrease in gain in the hearing aid in the near ear. These gain changes would have been captured in the probe recordings of the babble. However, probe recordings of the speech would not have reflected a change in gain because these recordings were obtained separately. Thus, if the gain in the far-ear or near-ear hearing aid changed with noise-source location, AAI calculations of speech audibility could have been inaccurate. To explore this possibility, the amount of head shadow and boost were estimated by comparing probe-microphone measures between the far and near ear both unaided and aided (i.e., spectra near ear–spectra far ear, with babble at 90°). It was hypothesized that a smaller difference in spectra aided than unaided would suggest that the gain in the far ear had increased the gain in the near ear had decreased. AAIs were recalculated incorporating these estimated gain changes. The results did not show systematic changes in predicted HINT thresholds.

Finally, AAI calculations may have been affected by various probe microphone measurement errors. First, although the difference between a single measure and the average of many measures is small, with a standard deviation of approximately 3 dB (Dillon 2001), 12 separate recordings [(unaided and aided HINT spectra × two ears) plus (unaided and aided babble spectra × two speaker locations × two ears)] were obtained for each subject for AAI calculations. Procedures were used to minimize error, such as marking individual probe tubes to ensure consistent placement of the probe and marking hearing aid volume controls. Second, according to Dillon (2001), probe microphone measures obtained at 45° azimuth allow for the greatest amount of head movement without adversely affecting measurement accuracy. However, probe measures in the current study were obtained at 0° and 90° azimuths. To reduce error, subjects were instructed to keep their heads in a fixed position during speech testing and probe-microphone recordings. Although monitored by the experimenter, some head movement may have occurred.