This study suggests that SSD CI listeners do not perform as well in speech perception with their CI alone as do age-matched BCI listeners listening with a single CI. This group difference persisted despite the inclusion of factors such as age, length of device use, etiology and device manufacturer in the models. Similar results have been seen for SSD CI listeners in prior work [39, 40, 47]. However, it is important to note that these studies did not directly compare SSD speech perception with that of BCI listeners. The control group in these studies consisted of CI users who were bilaterally deafened but who used only one CI. In this case, the fact that the SSD group — but not the control group — was listening with an atypical configuration (only one ear) could have heightened any differences between the groups. In contrast, our comparison controls for this and still shows a robust difference.

The number of hours recorded using the CI appeared to mediate group differences. Specifically, when this factor was added to the model, the group differences disappeared. However, adding device use to the model did not account for significant new variance over and above that of just group alone. Thus, this suggests that these effects share variance — SSD users do use their devices less and perform worse at speech perception. But the causal direction is not clear.

Device manufacturer significantly improved model fit across both matching conditions. However, this improvement was only driven by a significant main effect: Advanced Bionics users had overall worse CNC performance in the best matched condition. However, there were no other main effects or moderations of group across models. Additionally, there were no differences among manufacturers in the SSD group alone. Importantly, the group effect remained significant with the addition of this term in the models. These results suggests that the relationship between device manufacturer and CNC performance may be fairly inconsistent, mirroring prior research reporting mixed results with manufacturer comparisons [48]. Thus, we do not make strong claims about differences among CI manufacturers. Further research is needed to more comprehensively investigate the influence of manufacturer differences on CNC performance.

The underperformance of SSD CI listeners relative to BCI users suggests that SSD listeners may experience a blocking effect when adapting to their CI. That is, their normal hearing ear may be sufficient for speech perception in many daily tasks, creating less need to adapt the ear that perceives a degraded signal. Nonetheless, it should be noted that all these participants were tested at least 18 months post implantation, and many had quite a bit more (the average was over 6 years). Therefore, the strategies that SSD listeners are employing are not likely to be due to lack of experience with the implant. Supporting this, in this population of highly experienced users there were no significant overall effects of length of device use on CNC performance, suggesting that no further adaptation is likely to be seen.

This blocking effect could arise from several (not mutually exclusive) causes. It may be as simple as the fact that the NH ear perceives speech “better”, and the CI user simply turns off the CI in many situations. This is supported by their significantly lower daily usage rates. Alternatively, even if the CI is on, the NH ear may simply get more attention, again attenuating the experience. However, in addition to this, the introduction of a second “ear” with a highly distinct (and spectrally degraded signal), with a potentially misaligned place/pitch match [27, 49]. Additionally, the second ear could pose challenges to binaural integration [37], further leading listeners to ignore the CI ear. The CI ear could also create additional effort or fatigue. Such effects could lead a listener to use their CI less (further exacerbating the problem). That is, under either of this account, blocking is not inherent to the mechanisms of learning or adaptation, but arises from differential use (either real usage, or attentional differences). However, if this effect was only due to a matter of experience with the CI, 6 years at even a quarter of the daily usage (or attention levels) of a BCI listener would provide two years of functional experience—more than is needed for the adaptation period of a typical CI. Therefore, this concern should not significantly affect our results. Nonetheless, we tested this by asking if BCI users with more asymmetrical hearing show a profile more similar to SSD listeners. This prediction was not confirmed: CNC performance for BCI listeners with more asymmetry was still better than for SSD listeners. This supports our claim that any potential disparity between BCI listeners was not enough to account for the difference between groups.

Alternatively, blocking may be a consequence of error-driven learning [38]. In most learning and adaptation systems, learning is a function of the difference between the desired output of the system and its actual output. When this difference is low, little learning occurs. This is shown in classical conditioning experiments in which a learner first learns to pair one stimulus (e.g., a light) with a reward. When a second one is introduced (a tone is added to the light) they do not learn—even though it consistently predicts the reward. That is, when they later experience the second stimulus (the tone) alone, they show no evidence of learning. This was because the reward was already perfectly predictable from the first stimulus—there was no error signal present to drive learning. This form of error-driven learning—including the blocking effect—has been shown in learning novel speech categories [50], raising the possibility that may also apply here. SSD listeners offer a clear analog to this. In an SSD listener, the NH ear offers a clear and useful signal in most circumstances. As a result, even if the CI offers suboptimal input (that would induce high degrees of error in its own), the actual error signal is quite low, and little learning occurs.

Blocking may also be relevant in listeners with other forms of binaural integration. For example, in sequential BCI listeners, the first implanted CI is likely to have undergone some adaptation before the second CI is implanted. This adaptation is likely to be strong because there is no NH ear to fall back on: the error signal is high, and the adaptation has high stakes. However, when the second CI is added, performance is already good using the first CI, so there is less learning with the second CI. This is exactly what we observed in our exploratory analysis of the sequential BCI users in this study (Fig. 2). In contrast, simultaneously implanted BCI listeners must quickly adapt to the challenging auditory input provided by both CIs immediately following implantation; the error signal is high for both; they have no NH ear to fall back on. This may lead to a more robust adaptation. This hypothesis was supported by the fact that speech performance with the second implanted ear for sequential BCI listeners was significantly worse than that with the first implanted ear (Fig. 2), suggesting that BCI listeners could be relying on their better ear to perceive speech. However, it should be noted that this group of sequential listeners is too small to make strong claims.

These results have implications on audiological and clinical outcomes for SSD listeners who plan to receive a CI. Specifically, these results imply that SSD listeners may have unique needs for adaptation if they want to get the most out of their CI. Specifically, SSD listeners need more experience with their CI only to achieve equivalent levels of performance, and it may be important to control other strategies that they might use to minimize the error signal in these cases (e.g., subtitles) to promote better adaptation. Additionally, our results within sequentially implanted BCI listeners (that the first implanted ear performed better than the second ear), suggests there may be benefits of simultaneous implantation in this group (or at least simultaneous activation).

This study has several limitations worth noting. To start, analyses based on biological sex were not performed in this study since the comparison groups of interest (SSD vs. BCI) were generally matched on biological sex distribution. Due to this, there was little evidence for justification of biological sex influencing CI group differences in CNC scores.

Most importantly, we do not know what kind of auditory rehabilitation the listeners may have undergone. As part of the standard of care at our center, participants were recommended by their clinicians to do auditory training for 30 min a day for the first three months using the manufacturers rehabilitation programs. However, we did not have specific records of aural rehabilitation or auditory training in either group. Empirical and clinical work suggests a strong association between the persistent presence of auditory training and better speech perception outcomes [51,52,53,54].

Some experimental work suggests that adaptation can occur within a short period of time [55]. Since BCI listeners require more rehabilitation for both ears, it is quite possible that the BCI group had overall more access to training and rehabilitation, leading to the difference in speech perception between groups. The data logging analysis indexing CI usage in this paper helped to address part of this concern. This sample of data logging was small yet showed very large differences between groups. This makes it difficult to determine whether the effect of SSD listening was purely due to reduced experience/usage or derives from other causes. Additionally, without this specific information, it makes it difficult to determine the appropriate length of time (short or long) that SSD listeners would need to adapt to their implant. It is also possible that the differences observed in BCI participants with a large interaural difference may be related to differential device use between their two CIs. However, due to the limited availability of data logging, it is difficult to determine the extent to which device use may have contributed to these findings.

Additionally, all of the CI users had undergone CI processer tuning prior to speech perception testing, ensuring that they were in a similar state prior to testing. However, as we did not have access to aided audiometry for most participants, we cannot fully rule out the possibility that either the BCI and SSD groups were not using their processors at the right or consistent levels at test. Differences in processor volume or strategy between groups could have also contributed variance to the difference in speech perception.

Two factors prevented us from accounting for all etiological differences. First, for many patients in this sample, the etiologies were unknown. Second, there were many etiologies with only one or two patients — too few to make any definitive conclusions. However, we suspect that this is not a major factor, as in this specific study, the sample showed similar etiologies across the two groups despite expectations that there would be more single ear causes of deafness such as trauma or schwannoma. It may be important to evaluate CI only performance in listeners with a broader range of etiologies to determine if they play a role in adaptation. Moreover, with a highly diverse array of etiologies (and only a moderate sample), led to an analysis of etiology only by broad groups — a larger sample could potentially identify more specific factors that are relevant.

The focus of this study was entirely on single CI performance. However, the most important benefit of the CI is for binaural and spatial hearing. Thus, even as CI performance is lower than it could be, this may be outweighed by benefits of spatial hearing (or even of head shadow and squelch). Therefore, the results do not speak to whether there is overall benefit of SSD CI use. Rather the results are more focused on whether SSD CI users are achieving the same level of CI performance as other CI users.

Finally, the SSD listeners in this study achieved an average score of 34.9 on word recognition. It is important to note that this score is not 0, implying some degree of speech perception. These findings suggest that SSD listeners are indeed adapting to their CI, albeit through a potentially less efficient process when contrasted with their bilateral CI counterparts. Consequently, these results underscore the potential for growth and enhancement through intervention and training. However, a comprehensive understanding of the most effective auditory rehabilitation methods necessitates a deeper exploration of the fundamental mechanisms underlying binaural perception and its influence on speech perception in SSD CI listeners, and with binaural fusion.