REDUCED DIGIT SPANS AND EAR DOMINANCE USING DICHOTIC DIGITS IN BIMODAL COCHLEAR-IMPLANT USERS

Abstract

Bimodal stimulation, a cochlear implant (CI) in one ear and a hearing aid (HA) in the other, provides highly asymmetrical inputs. To understand how asymmetry affects perception and memory, forward and backward digit spans were measured in nine bimodal listeners. Spans were unchanged from monotic to diotic presentation; there was an average two-digit decrease for dichotic presentation with some extreme cases of decreases to zero spans. Interaurally asymmetrical decreases were not predicted based on the device or better-functioning ear. Therefore, bimodal listeners can demonstrate a strong ear dominance, diminishing memory recall dichotically even when perception was intact monaurally.

I. INTRODUCTION

Hearing with two ears is usually superior to one ear, which motivates bilateral hearing assistive devices for treatment of sensorineural hearing loss. Bilateral assistive devices can be superior to single devices (Byrne, 1981; Litovsky et al., 2009) because the individual experiences improved speech recognition, sound quality, and spatial hearing (e.g., Warren and Dunbar, 2018; Yawn et al., 2018). Individuals often receive two hearing aids (HAs) or two cochlear implants (CIs) with the hopes of experiencing relatively symmetrical auditory inputs, where binaural advantages are best realized (Rothpletz et al., 2004; Bernstein et al., 2016).

In contrast, hearing through asymmetrical inputs can lead to detriments (Rothpletz et al., 2004; Goupell et al., 2018; Bernstein et al., 2020). “Bimodal stimulation” is one example of providing a highly asymmetrical input across the ears. One ear is provided access to lower-frequency sounds acoustically with a HA (often up to about 1 kHz, providing access to a speaker’s fundamental frequency, prosodic speech pitch cues, and musical pitch) and the other ear is provided sound electrically with a CI (typically 200–8000 Hz, which is highly important for speech recognition; however, it is highly degraded spectrally and omits information <200 Hz critical for pitch perception). Therefore, bimodal CI users receive electric and acoustic signals that interaurally differ in frequency, timing, and intensity (Pieper et al., 2022). While it is possible to program the CI and HA to have the frequency information across ears be complementary or be better interaurally matched over the middle frequencies around 1 kHz, there remain other interaural matching issues that could impact outcomes, including a lack of CI and HA processing delay synchronization and how interaural loudness should be set for the CI and HA (Francart et al., 2011; Holder et al., 2022; Sharma et al., 2023).

Bimodal CI users can exhibit improved speech recognition utilizing bimodal stimulation compared to just one CI (e.g., Yawn et al., 2018), which appears dependent on the effective interaural integration of sound (e.g., Yoon et al., 2015; Sheffield et al., 2016). Other bimodal CI users may receive smaller benefits or even detriments (i.e., interference), likely because of the asymmetry caused by interaurally mismatched inputs (e.g., Ching et al., 2007).

Interference can occur with bilateral HA users (Henkin et al., 2007; Jerger et al., 2017) and other groups of CI users (Goupell et al., 2018; Bernstein et al., 2020), particularly in some spatial-hearing tasks. In 9 bilateral CI listeners, Goupell et al. (2018) found an average of 9 dB of contralateral speech interference from the binaural configuration in a contralateral unmasking task. For this task, a binaural speech benefit or interference was measured by comparing speech recognition for conditions with a target and masker in a target ear (monaural) compared to when there is a copy of the masker added to the contralateral ear (dichotic). For the dichotic condition, the listener’s binaural processing should produce a change in perceived spatial location of the masker compared to the target (resulting in a binaural speech recognition benefit) or the listener should direct attention to only the target ear and ignore the other ear (resulting in a relatively small decrease in speech recognition). The bilateral CI listeners of that study had a wide range of overall performance, interaural differences in performance (three listeners were specifically recruited because of their asymmetry), better ears, and ages. Contrary to expectation of improvement or at least no change, some of those listeners experienced interference when the target was in the relatively poorer ear and some listeners experienced interference in both ears. Better speech recognition in the target ear reduced interference, but better speech recognition in the contralateral ear or a longer duration of deafness in the target ear increased it. In a separate study, Bernstein et al. (2020) found a wide range of contralateral speech interference for targets in the CI ear of 13 single-sided-deafness CI users (CI in one ear, normal to near-normal acoustic hearing in the other ear). Interference was experienced when the target was placed in the ear with the CI, but not when the target was placed in the ear with acoustic hearing. In addition, there was large inter-individual variability and the magnitude of interference increased with increasing age. In summary, some bimodal, bilateral, and single-sided-deafness CI listeners experience benefits from bilateral inputs (Bernstein et al., 2016; Yawn et al., 2018). Some, however, experience interference for speech recognition in some tasks and it is unknown how to predict such an occurrence. Interference could occur primarily for poorer performing ears or only in ears with CIs.

Therefore, to better understand how hearing asymmetry affects the individual variability in bimodal CI use, digit recall (a test of short-term or working memory) was measured under conditions of monotic, diotic, and dichotic signal presentation. Bimodal CI users can be considered even more asymmetrical than bilateral CI and single-sided-deafness CI users because of the mostly non-overlapping nature of their low-frequency acoustic and electrical inputs across ears. They present an interesting case to compare to bilateral and single-sided-deafness CI users because the relative weighting of their electric vs acoustic input appears to rely on their residual hearing and other demographic factors. The task was digit recall because it may reveal more subtle changes in abilities related to memory, cognition, and listening effort (Sladen et al., 2018; Macpherson et al., 2019; DeRoy Milvae et al., 2021) as compared to speech recognition. We hypothesized that bimodal users’ digit spans would decrease for dichotic compared to monotic presentation, and that there would be an ear dominance (i.e., larger digit spans) for the functionally better ear (DeRoy Milvae et al., 2021; Goupell et al., 2021).

II. METHODS

A. Listeners

Nine middle-aged to older bimodal CI listeners (55–85 yrs, six female, demographic information in Table I) with sensorineural hearing loss participated in this study. They reported regular use of a digital hearing aid, either a receiver-in-the-canal or behind-the-ear style. The listeners’ thresholds in the non-implanted ear were within the mild-to-profound range at 250–8000 Hz [Fig. 1(A–I), individual inset panels]. They had monaural-aided single digit identification scores ≥50% in the HA ear. Two other listeners started the experiment but did not meet the digit identification score requirement. The listeners also passed an age-related cognitive decline screening (standard non-hearing-impaired Montreal Cognitive Assessment; Nasreddine et al., 2005) with a score of >22/30 (e.g., Goupell et al., 2018).

Table I:

Listener demographic and device information, including the self-reported Better Ear (BE).

Listener	Age (yrs)	Sex	CI Ear	BE	Implant Age (yrs)	Onset of Deafness (yrs)	CI Manufacturer
S1	74	F	R	CI	66	40	Cochlear Ltd.
S2	73	F	R	HA	65	5	Cochlear Ltd.
S3	73	M	L	CI	68	43	Cochlear Ltd.
S4	67	F	R	CI	47	0	Cochlear Ltd.
S5	85	M	R	CI	79	77	Cochlear Ltd.
S6	78	M	R	HA	62	55	Cochlear Ltd.
S7	55	F	L	HA	50	0	Cochlear Ltd.
S8	71	F	R	CI	61	61	Advanced Bionics
S9	66	F	R	HA	61	54	Cochlear Ltd.

Fig. 1.

Forward and backward digit spans for the individual listeners (panels A-I), with the hearing thresholds (HT) reported in dB HL in the inset in each panel. Forward and backward digit spans are reported on average for the better ear (BE) and poorer ear (PE) (panel J) and on average for CI and HA ears (panel K). Note that both forms of averaging reveal similar patterns of performance with minor only differences. The error bars in panels (J) and (K) represent ±1 standard error.

B. Procedure

Listeners were tested in a double-walled sound-attenuating booth (IAC, NY). They used their own CI sound processors in their everyday program. Before testing, an experimenter received a copy of the CI programs and informally verified they were programmed within standard practices. The audiometric hearing thresholds and digit recognition were measured in the acoustic-hearing ear. To avoid variability from different HAs, the listeners were provided a research-dedicated behind-the-ear HA (Naida V90-UP, Phonak/Sonova; Stäfa, Switzerland) that was programmed according to their hearing loss using NAL-NL2 patient fitting algorithm and real-ear verification (Verifit2, Audioscan; Dorchester, Ontario). Since NAL-NL2 is one of the more commonly used amplification fitting methods and informal inspection of the listener’s HA fitting seemed appropriate for a listener’s hearing loss, it was likely that minimal acclimatization to the experimental HA program would be required. The HA was set to the direct-audio-input (DAI) program with microphones disabled to eliminate ambient noise. Stimuli were presented via DAI to each device (CI and HA) using a desktop computer and custom experimental software (MATLAB, Mathworks; Natick, MA). Listener S8 was an exception, who listened to the stimuli under open-back circumaural headphones (HD650, Sennheiser; Hannover, Germany) because there was no DAI on the Advanced Bionics CI sound processor. This listener experienced no problems with feedback of the open-back headphone over the HA.

After completing the audiogram, using it to fit each listener’s HA, and verifying that the HAs met prescriptive targets, single digits were presented via live voice to the acoustic ear at the listener’s most comfortable listening level and without visual cues to determine if they could accurately identify ≥50% of the digits with the HA alone.

The perceived loudness of the acoustic and electric stimuli was balanced between ears. Listeners were presented single digits and were asked to identify which ear sounded subjectively softer in presentation level. The experimenter manually adjusted the stimulus loudness in the CI ear using the loudness of the HA ear as an anchor.

Then single digits were presented to each ear individually to verify that the listeners could hear and identify the digits at a sufficient level. Randomly ordered digits, presented one digit at a time, were tested monotically. There were ten trials per digit. If the listener had <80% correct identification for an individual digit, that digit was excluded from the main experiment. This was to attempt to minimize identification errors as a confounding factor in measuring memory spans (Cleary et al., 2018), while also attempting to not remove too many digits. For example, if the number five had <80% identification in the CI ear but not the HA ear, then it was removed from the conditions where the CI would encounter that number (specifically for this case: Monotic CI, Diotic, and Dichotic CI, see description of condition below). Two listeners had one digit removed, one listener had two digits removed, and one listener had four digits removed.

The main experiment assessed the listeners’ short-term and working memory capacity using standard forward and backward digit span measurements (Wechsler, 1997). The task consisted of presenting recorded strings of randomized numbers 1–10 (but not 7, because it is multi-syllabic). For simplicity, the number ten is referred to as a digit. Stimuli were presented at one digit per second in a randomized order. List lengths were randomized between two, four, six, and eight digits. Listeners used a mouse or touch screen to report the digits in the correct order (forward or backward). The graphical user interface was labeled with numbers in the form of an analog clock (Cleary et al., 2018).

The conditions were tested in blocks that were randomized across listeners. They were instructed if they would be presented sound to their left ear, right ear, or both ears before each block. The conditions were: (1) monotic to the CI ear, (2) monotic to the HA ear, (3) diotic (the same digit string to both ears), (4) dichotic attending to the CI ear, and (5) dichotic attending to the HA ear. In the dichotic cases, the two digit strings had digits that were time synchronized but individual digit pairs were required to differ across the ears; single digits could otherwise be repeated within or across ears.

The reported digit span was the 50% correct point taken from a sigmoidal fit to the psychometric function (psignifit; Wichmann and Hill, 2001). Ten trials per condition were tested. This created 4 list lengths×5 conditions×2 directions×10 trials=400 total test trials. The testing took approximately 8 hours, typically two sessions of 4 hours each. The listeners were given breaks between blocks as needed to avoid fatigue.

III. RESULTS

The individual data are shown in Fig. 1(A–I), which were notably variable. There were cases where the digit spans decreased minimally between monotic and dichotic conditions (e.g., S1 for the forward CI condition, S3 for the forward HA condition, and S6 in several conditions that could be the result of a ceiling effect). There were relatively few cases of symmetrical decreases between monotic and dichotic conditions (e.g., S2, S3, and S8 for the backward conditions). More common were cases with asymmetrical decreases between monotic and dichotic conditions. Sometimes the HA ear decreased more under dichotic presentation (e.g., S1) and sometimes the CI ear decreased more (e.g., S2). Interestingly, there were even extreme cases where the dichotic digit span dropped to zero (from spans of about 4 in the monotic HA conditions for S4, from a span of about 5 in the forward monotic HA condition for S7). In addition, S4 and S7 had some of the lowest digit spans, had some of the most high-frequency hearing loss, had very early (prelingual) onsets of deafness, and had the longest durations of severe-to-profound hearing loss (see Table I).

We attempted to predict the cases of relatively larger decreases between dichotic and monotic conditions (i.e., the asymmetry in performance) by first sorting the ears into a relatively better ear (BE) and poorer ear (PE) based on listener self-report. Five listeners reported their CI as their BE (Table I). For seven of nine listeners, the self-reported BE was consistent with the BE based on the monotic digit string performance. In other words, self-report was mostly but not entirely consistent with an objective measurement of BE and PE performance.

Average digit spans for the BE and PE are shown in Fig. 1(J). For the forward digit spans, the average difference between BE monotic and diotic digit spans was −0.11±0.19 (average±standard error), and between PE monotic and diotic digit spans was 0.04±0.20. For the backward digit spans, the average difference between BE monotic and diotic digit spans was −0.38±0.17, and between PE monotic and diotic digit spans was 0.32±0.25. These observations for the digit spans were supported by a two-way repeated-measures analysis of variance (RM ANOVA) with factors condition (three levels: diotic, monotic BE, monotic PE) and direction (two levels: forward, backward). Forward spans were significantly larger than backward spans [F(1,8)=19.9, p=0.002, ${\eta }_p^2=0.71$] by about 0.4 digits. The effect of condition was not significant [F(2,16)=2.71, p=0.097, ${\eta }_p^2=0.35$]. The direction×condition interaction was not significant [F(2,16)=1.05, p=0.372, ${\eta }_p^2=0.20$]. In summary, the monotic control and diotic conditions produced no appreciable difference in digit spans.

Average digit spans decreased between the monotic/diotic and dichotic conditions. To better highlight the changes in the digit spans and account for the different baseline monotic spans across listeners, we calculated the change in digit span for the monotic minus the dichotic span (not shown). For the forward digit spans, the average difference between BE monotic and BE dichotic digit spans was 2.22±0.47, and PE monotic and PE dichotic digit spans was 2.38±0.55. For the backward digit spans, the average difference between BE monotic and BE dichotic digit spans was 0.87±0.34, and PE monotic and PE dichotic digit spans was 1.96±0.50.

Next, we attempted to predict asymmetrical digit spans decreases between the monotic/diotic and dichotic conditions by considering the BE and PE. A three-way RM ANOVA with factors condition (two levels: monotic, dichotic), ear (two levels: BE, PE), and direction (two levels: forward, backward) was performed on the raw digit spans (not the difference) shown in Fig. 1(J). Forward spans were significantly larger than backward spans [F(1,8)=38.922, p=0.0002, ${\eta }_p^2=0.83$] by about 0.4 digits. Monotic conditions had significantly larger spans than dichotic conditions [F(1,8)=8.63, p=0.019, ${\eta }_p^2=0.52$]. The BE was not different than PE [F(1,8)=1.13, p=0.320, ${\eta }_p^2=0.12$]. The condition×direction interaction was significant [F(1,8)=16.1, p=0.004, ${\eta }_p^2=0.67$]. Subsequent two-tailed paired t-tests Bonferroni corrected for six comparisons (α=0.05/6=0.00833) revealed that the backward monotic and backward dichotic conditions were not different (p>0.00833) and the remaining five conditions were significantly different (p<0.00833 for each). The condition×ear, direction×ear, and the direction×condition×ear interactions were not significant (p>0.00833 for all three). In summary, because the main effect of ear and interactions with ear were not significant, the BE and PE distinction was not predictive of asymmetrical decreases between the monotic/diotic and dichotic conditions.

Next, we attempted to predict asymmetrical digit span decreases between monotic/diotic and dichotic conditions by considering the device type. Average digit spans sorted by CI and HA are shown in Fig. 1(K). The average digit spans changed minimally between Fig. 1(J) and 1(K). A three-way RM ANOVA using the factor of device (two levels: CI, HA) rather than ear (i.e., the ears are not sorted into BE and PE) revealed that the main effect of direction, main effect of condition, and the direction×condition interaction remained significant (p<0.05 for all). The main effect of device and the other three interactions were not significant (p>0.05 for all). Therefore, the data and analyses were consistent regardless of how the ears were sorted, either by BE/PE or by device type. In summary, the ear with the CI or HA was also not predictive of asymmetrical decreases for the dichotic conditions compared to the monotic/diotic conditions.

In summary, the magnitudes of the asymmetrical digit span decreases between monotic/diotic and dichotic were idiosyncratic, not consistently aligning with BE/PE [Fig. 1(J)], device type (CI/HA) [Fig. 1(K)], or residual acoustic hearing (not shown).

IV. DISCUSSION

Bimodal CI users receive highly asymmetrical inputs, low-frequency acoustic hearing in one ear and higher-frequency electric hearing in the other. The large individual variability of bimodal CI performance occurs across a myriad of tasks (e.g., Ching et al., 2007; Yawn et al., 2018; Holder et al., 2022). Some spatial-hearing and dichotic-listening tasks reveal how well the ears function together or interfere with each other (Bernstein et al., 2016; Goupell et al., 2018; Bernstein et al., 2020). We hypothesized that interaural asymmetry may affect memory of digit strings. Such effects could be related to effortful listening (Sladen et al., 2018; Macpherson et al., 2019), and become pronounced in cases of dichotic presentation when there can be a relatively better (more salient) ear and a poorer (less salient) ear (DeRoy Milvae et al., 2021; Goupell et al., 2021).

We found that digit spans did not significantly change between monotic and diotic conditions when there was a single digit string [Fig. 1(K)]. This suggests that bimodal CI listeners’ memory is unaffected when a single sound source is presented to one or both ears, similar to speech recognition of a single talker in quiet. However, digit spans significantly decreased by almost two digits for dichotic presentation, when the information across ears was conflicting. Such a situation could be considered similar to listening to a target talker with interfering sound sources present (i.e., cocktail party listening).

The addition of an extra different digit string in the non-target ear resulted in the decrease in digit spans. One interpretation of this decrease is that there was increased listening effort under dichotic presentation (DeRoy Milvae et al., 2021; Goupell et al., 2021). Forward digit spans (short-term recall) were longer than backward digit spans (working memory) [0.4 digits on average; Fig. 1(K)], also consistent with the idea that the dichotic condition was a more demanding and effortful task. Such an interpretation would be also consistent with the idea that there are some situations where hearing with two ears may not be better than one (Ching et al., 2007; Henkin et al., 2007; Jerger et al., 2017; Goupell et al., 2018; Bernstein et al., 2020). Future studies on dichotic listening in bimodal CI users could include other measurements of listening effort, such as pupillometry.

Another interpretation of these data is related to the cases where digit spans dropped to zero for dichotic presentation [Fig. 1(D), 1(G)]. In a separate similar study, two of 11 bilateral CI listeners showed a similar dramatic decrease in performance in dichotic listening (Goupell et al., 2016). Goupell et al. (2018) discuss these extreme drops in performance as a possible “perceptual extinction,” or the notion that the PE perception is so relatively weak that it completely vanishes in the presence of the BE perception. An abnormally strong pathway for the bimodal CI listeners’ dominant ear could occur for listeners if they were implanted later in life, had a notably longer duration of deafness in just one ear, or experienced a lack of auditory stimulation during critical development periods. Future studies could verify perceptual extinction in CI users, including bimodal CI users, by performing dichotic listening experiments that include a secondary task that explicitly asks the listener to report the perceived number of sources (i.e., was there sound in one or both ears?).

There were cases of better digit spans in the CI ear compared to the HA ear [e.g., Fig. 1(A), 1(D), 1(E)] and vice versa. If dichotic listening is related to binaural benefits and contralateral speech unmasking/interference, this would suggest that large interference effects in the CI ear of single-sided-deafness CI listeners are a result of the relatively poor signal quality (Bernstein et al., 2020; Goupell et al., 2021). Likely because of the small sample size, it could not be determined whether the relatively worse dichotic digit spans occurred for the PE in the bimodal CI users of the current study. Future studies would benefit from using a larger sample size. In addition, collecting just the hearing thresholds in the acoustic hearing was not sufficiently enlightening about the BE and PE functionality. It would also be beneficial to define the BE and PE based on an independent and objective measurement of functional performance, such as word or sentence recognition (Goupell et al., 2018) or a non-speech task like spectro-temporal ripple discrimination (Archer-Boyd et al., 2018).

The sample size of nine bimodal CI listeners was also too small to seriously investigate potentially interesting relationships with hearing thresholds and demographic factors such as age, onset severe-to-profound hearing loss, and duration of severe-to-profound hearing loss. Future studies with a larger sample size and a wider range of hearing abilities in the CI vs the HA ears would be particularly informative. For example, other similar studies show changes in dichotic listening and observe larger interference with increasing age (Walden and Walden, 2005; Henkin et al., 2007; Bernstein et al., 2020).

Other methodological choices can be explored in future studies. For example, it is unclear how loudness should be set across the ears (Holder et al., 2022). Is testing using loudness balanced inputs appropriate for all bimodal CI listeners or should they listen to stimuli at the loudness levels they experience typically? There could also be effects of stimulus acclimatization and practice. Macpherson et al. (2019) showed that digit spans for 15 bimodal CI users significantly improved a year after surgery for bilateral CI use. While that study lacked a control group that could rule out practice effects on the tests, their results could suggest that memory for bimodal CI users may be impaired by the input asymmetry and possibly effortful listening.

Ultimately, controlled laboratory tests like dichotic listening might reveal how CI users functionally perform in real life, particularly in terms of binaural benefits. For example, Walden and Walden (2005) found speech recognition in noise was better for a unilateral HA compared to bilateral HAs (i.e., suggesting their listeners experienced interference). Some of this was explained by ear asymmetries, although differences were small, and some analyses did not reach statistical significance. Furthermore, dichotic listening tests might be useful in device fitting to maximize outcomes. Ideally, hearing with two ears should improve speech recognition, sound quality, spatial hearing, and quality of life (e.g., Warren and Dunbar, 2018; Yawn et al., 2018), while also reducing listening effort (Sladen et al., 2018; DeRoy Milvae et al., 2021) and the fatigue that accompanies continual effortful listening.

DATA AVAILABILITY

The data that support the findings of this study are available from the corresponding author upon reasonable request.