Association of quality of speech perception with quality of life in cochlear implant users with single sided deafness

Introduction

Cochlear implants (CIs) have become powerful tools for addressing the growing burden of hearing loss (HL)—a highly prevalent condition that correlates with reduced cognitive function, increased feelings of isolation, and reduced quality of life (QOL; Yu et al., 2024; Lin, 2024; Zahnert, 2011). CIs help mitigate the sequelae of HL by restoring auditory input. Studies consistently demonstrate that cochlear implantation is associated with improvements in speech clarity and QOL, and reduces listening effort and social isolation (Boisvert et al., 2020; McRackan et al., 2018; Wick et al., 2020; Carlson, 2020; McRackan et al., 2017).

A key outcome following implantation is improvement in the implantee's speech perception (SP), which is the ability to recognize and understand speech, usually measured clinically via word and sentence recognition testing. Current guidelines recommend that, following cochlear implantation, recipients undergo individualized programming sessions focused on optimizing loudness and electrical threshold settings to enhance SP outcomes (Dillon et al., 2022; Buchman et al., 2020a). CIs have achieved success in restoring speech understanding, and have consequently become the standard of care for individuals achieving inadequate rehabilitation with hearing aids (HAs; Dornhoffer et al., 2021; Buchman et al., 2020b). However, SP has been shown to have weak associations with QOL in CI users, and is only able to explain a small portion of overall QOL changes (McRackan et al., 2018; Buchman et al., 2020b; McRackan et al., 2019a; Dorismond et al., 2023).

A related but distinct outcome is that of speech quality perception (SQP), which is the impression of how speech sounds to a CI user. SQP colors the auditory listening experience and encompasses paralanguage such as tone, clarity, and naturalness that are important to navigate social interactions (Taitelbaum-Swead et al., 2022; Kelly et al., 2025; Marx et al., 2015). CI users have been shown to have lower confidence and self-efficacy in social interactions regardless of speech understanding; this phenomenon has been attributed to complex factors that may include SQP (Carlson, 2020; Reddy et al., 2024). Many CI recipients report poor SQP, often describing their experience of speech as overly mechanical or cartoonish. Some also struggle to identify the speaker's gender, lower frequencies, or emotional cues in speech (Luo et al., 2007; Fu et al., 2005). However, the possible relationship between SQP and QOL remains unexplored. Individuals with single sided deafness (SSD), have one ear with normal hearing ear and one ear with HL. SSD CI users remain an important population of study, due to the ability to assess distinct contributions of both a CI-aided and a NH ear.

Several tools have been developed to assess both SQP and QOL in CI users. The Hearing Implant Sound Quality Index (HISQUI19) tests, the Speech Spatial and Qualities of Hearing Scale (SSQ), and the Columbia Speech Quality Instrument (CSQI) are among them (Kelly et al., 2025; Fu et al., 2005; Chen et al., 2018). CSQI is a validated computerized tool that tests 14 speech attributes using 9 audio clips in both normal-hearing and CI populations (Chen et al., 2018). CSQI specifically targets speech aspects that CI users often find unpleasant (Lalwani et al., 2024). A recent study found that for individuals with single-sided deafness (SSD)—unilateral implantees with normal hearing (NH) in the contralateral ear—SQP was worse in the CI ear than in the NH ear, but when listening binaurally, SQP was comparable to NH controls (Kelly et al., 2025). For QOL, the Cochlear Implant Quality of Life instruments—CIQOL-10 and CIQOL-35—are currently the most comprehensive and well-validated tools available (McRackan et al., 2019a). The CIQOL-35 provides an overall global QOL score, which correlates with an abbreviated 10-item version of the instrument (known as the as the CIQOL-10). CIQOL-35 also outputs domain-specific assessments including communication (ease and clarity of communication), emotion (impact of HL on emotional wellbeing), entertainment (enjoyment and clarity of media), environment (distinguish and localizing environmental sounds), listening effort (effort and fatigue with listening), and social (ability to socialize in groups).

Given the documented changes in SQP when listening through a CI, and the lack of correlation between speech understanding and QOL among CI users, we sought to use validated measures of SQP and QOL to investigate whether these variables may be related among a population of CI users with SSD. Acoustic and prosodic elements of speech have been shown to have a role in impression formation and emotional response; thus we expect that SQP would most strongly influence quality-of-life domains related to interpersonal engagement, media participation, and emotional wellbeing (Walter-Terrill et al., 2025; Coutinho and Dibben, 2013; Weninger et al., 2013; Schewski et al., 2025). Therefore, we hypothesized that higher SQP scores would be positively associated with global CI-related QOL as well as the communication, entertainment and emotion domains.

Materials and methods

Participants

This research was approved by the Columbia University Irving Medical Center Institutional Review Board (AAAU6865). Participants were recruited from the audiology practices at a tertiary medical center. All participants met the following eligibility criteria: (1) Adults (18 + years old), (2) English-speaking, (3) unilateral CI use with >6 months since activation, and (4) audiometrically confirmed NH (defined as thresholds of 25 dB or less in all measured frequencies) in the contralateral ear. The cause of patients' HL was obtained from participant's audiologists and electronic medical records. Most recent audiometric data including CI-aided pure tone average (PTA; average of 0.5, 1, 2, and 4 kHz thresholds) and post-operative full word Consonant-Nucleus-Consonant (CNC) of the implanted ear after implantation was also obtained from participants' electronic medical records.

Study design

An online cross-sectional survey was administered through Qualtrics, a web-based survey tool (Qualtrics, Provo, UT). Participants completed the survey and listening tasks online at home at their own pace. Participants completed a brief questionnaire covering demographic history (including age, gender, ethnicity, race, and education) and otologic history. Participants also subjectively rated their overall understanding of speech, and their overall enjoyment of hearing speech, on 10-point Likert scales (0–1 = “Poor,” 8–10 = “Excellent”; see Supplementary Figure 1). Ratings were designed to reflect their current, general listening experience in everyday life; thus, Likert items were administered prior to any listening task, and were not performed under any listening condition (CI ear only, NH ear only, etc.).

Next, participants completed the CIQOL35 questionnaire, which consists of 35 statements to which participants respond “Never,” “Rarely,” “Sometimes,” “Often,” or “Always,” based on how each statement applies to their experience with CIs. The full questionnaire is available in Supplementary Table 1.

After completing the quality-of-life assessment, participants were instructed to use external speakers in a quiet environment to complete the CSQI (Kelly et al., 2025; Chen et al., 2018; Lalwani et al., 2024). The CSQI alters one SQP attribute per clip and evaluates listener's ability to identify different dimensions of SQP across clips. The CSQI consists of nine audio clips where a male or female speaker reads the Rainbow Passage. Of these nine clips, two audio clips contain the original male audio and the original female audio. The other 7 clips manipulate one specific quality of speech: bassy, cartoonish, not-speech, far, garbled, mechanical, or rough. Participants listen to each clip and then rate 14 qualities of speech using a Visual Analog Scale (see Table 1). Participants completed the CSQI in three conditions: with the CI ear only (directly streamed to CI), with the NH ear only (external speaker with CI turned off) and binaurally with CI + NH ears (external speaker with the CI on). Listening condition order was randomized for each participant to reduce potential order effects. A volume adjustment occurred prior to each condition, during which participants were instructed to set the volume to a loud but comfortable level using a demo audio recording, and then instructed to not adjust volume again until the completion of that section. The survey was structured to guide participants sequentially through each section and incorporated recurring questions to confirm that they were completing the survey in a quiet listening environment.

Table 1. Columbia speech quality instrument (CSQI): 14 characteristics assessed per audio sample.

Participants listen to each of the nine clips in the CSQI and rate the following 14 attributes of speech using a Visual Analog Scale.

Statistical analysis

Statistical analysis was conducted using Microsoft Excel and R (version 4.4.1). CIQOL-35 scores were calculated in accordance with the Cochlear Implant Quality of Life (CIQOL) User Manual (McRackan et al., 2019a). Each survey item was designated as either forward- or reverse-scored, with responses (“Never,” “Rarely,” “Sometimes,” “Often,” and “Always”) assigned corresponding numerical values from 1 to 5 based on that designation. Each QOL domain (communication, emotion, entertainment, environment, listening effort, and social) was composed of a specific set of questions. The scores for each domain were summed and then converted to a standardized outcome measure using the domain-specific score conversion table provided in the manual. A Global CIQOL-10 score was similarly computed by summing the responses to the 10 designated questions across all six domains and applying the corresponding conversion. The maximum possible outcome score for each domain, as well as for the global CIQOL-10 score, was 100.

To calculate SQP scores, CSQI ratings were standardized in line with prior publications, such that a score of 0 uniformly reflected a greater perception of negative qualities or fewer positive qualities (Kelly et al., 2025; Chen et al., 2018; Lalwani et al., 2024). To normalize, scores for each negative quality were subtracted from 10. The following were defined as negative qualities: Cartoonish, Mechanical, Breathy, Hoarse, Echo-y, and Bassy. The quality Male was treated as negative only when rating the original female audio. The final CSQI score was then calculated by summing the normalized scores for all rated qualities across the 9 audio clips and dividing by 10. The maximum possible CSQI score is 126 (based on 10 points per 14 qualities across 9 clips). CSQI scores were computed separately for the CI ear, NH ear, and combined CI+NH listening conditions. The maximum CSQI score for each condition was out of 126. CSQI scores were interpreted relative to the published normal hearing participants' performance (mean [SD] = 68.7 [7.0]), with lower scores indicating poorer speech-quality (Chen et al., 2018).

Paired T-tests were performed to compare CSQI scores between the CI and the NH conditions, the CI and the CI+NH condition, and the NH and CI+NH conditions. Multiple linear regression analysis was performed to evaluate possible associations between CSQI scores (exposure) and each QOL domain, as well as the global QOL score (outcomes).

Multiple multivariable linear regressions were used to assess the possible association between participants' speech scores and both domain-specific and global QOL scores. In these analyses, the primary predictors were subjective ratings of speech enjoyment, subjective ratings of speech understanding and CSQI scores in the NH ear, the CI ear and the binaural (CI+NH) condition. The outcomes assessed were communication, entertainment, emotion and global QOL scores. All models adjusted for duration of HL and duration of CI use. No initial exploratory uncontrolled models were run. Given the number of models tested, Bonferroni correction was applied to account for multiple comparisons (adjusted significance threshold α = 0.05/20 = 0.0025).

In order to determine if observed associations were independent speech recognition ability, sensitivity analyses adjusting for CNC word recognition that tested the same predictors and outcomes was performed. Pearson cross-correlational analyses were also performed to assess potential collinearity among predictors.

Results

Demographics

A total of 16 CI users with SSD completed the study. Demographic data can be found in Table 2. The most common HL diagnosis was idiopathic sudden sensorineural hearing loss (68.8%). The mean [standard deviation (SD)] age of the participants was 54 (11.4) years, and 50% were female. The mean (SD) age of HL diagnosis was 43.2 (10.8) years, and mean (SD) years of CI use was 3.82 (2.0) years. In terms of audiometric function tests in the implanted ear, post-operative PTA was available for 14 of the participants and post-operative CNC was available for 13 of the participants. The mean (SD) post-implantation CNC score was 44.4 (15.2)% and post-implantation PTA was 29.8 (6.72) dB HL.

Table 2. Demographics of participants.

Yr, years; SD, standard deviation; HL, hearing loss; SSD, single sided deafness; CI, cochlear implant.

Speech quality perception scores

Participants self-reported moderately positive experiences with speech, with mean (SD) ratings of 7.6 (2.3) for overall enjoyment of hearing speech and rating of 6.9 (2.0) for overall understanding of speech (see Table 3).

Table 3. Self-reported speech enjoyment and understanding.

SD, standard deviation.

CSQI results showed of all the listening conditions, the CI ear only condition had lowest SQP score with a mean (SD) speech quality perception score of 49.8 (10.5). NH ear and combined CI +NH conditions had similarly higher mean (SD) scores of 69.1 (8.4) and 69.1 (7.9), respectively, with no significant difference between the two groups via t-test (p = 0.934). However, there was a highly significant difference between the CI only and CI + NH condition (p < 0.001) and between the CI only and NH only condition (p < 0.001). These results are summarized in Table 4.

Table 4. Average speech quality perception (SQP).

Columbia Speech Quality Instrument (CSQI) tool calculated SQP scores for three conditions, the Cochlear Implant (CI) ear only, o (NH) ear only, and combined CI +NH ears. SD, standard deviation; CSQI, Columbia speech quality instrument; CI, cochlear implant; NH, normal hearing. Minimum score = 0, maximum score = 126.

Quality of life scores

Participants had highest QOL score in the social domain of the CIQOL-35 with mean (SD) of 72.2 (16.8) and the entertainment domain with mean (SD) of 71.0 (22.7). The emotion and environment domains had mean (SD) ratings of 63.4 (9.7) and 60.3 (14.4) respectively. Mean (SD) communication-related QOL was 51.8 (10.6) and global QOL score was 53.0 (8.0). Listening effort had a mean (SD) of 38.3 (9.3). Full CIQOL-35 score results can be found in Table 5.

Table 5. Quality of life scores.

CIQOL-35, cochlear implant quality of life. Minimum score = 0, maximum score = 100.

Self-reported speech ratings as predictors of SQP and QOL

After Bonferroni correction (α = 0.0025), no significant associations were observed between speech enjoyment or understanding ratings and global or domain-specific CIQOL scores in models adjusting for HL duration and CI use. These results are summarized in Table 6.

Table 6. Self-reported speech ratings predicting quality-of-life, controlling for duration of hearing loss and duration of cochlear implant (CI) use.

CIQOL-35, cochlear implant quality of life; SE, standard error; CSQI, Columbia speech quality condition; NH, normal hearing.

CSQI as predictor of QOL

After Bonferroni correction (α = 0.0025), no significant associations were observed between CSQI scores (CI-only, NH-only, or binaural) and CIQOL domains. These results are summarized in Table 7.

Table 7. Speech quality perception scores as predictor of quality-of-life, controlling for duration of hearing loss, and duration of cochlear implant (CI) use.

CIQOL-35, cochlear implant quality of life; SE, standard error; CSQI, Columbia speech quality condition; NH, normal hearing.

Secondary CNC-adjusted models

Three participants were missing CNC data and were excluded from the secondary analysis adjusting for postoperative CNC scores; despite this, p-values of the CNC adjusted scores showed similar significance to primary models. These results are summarized in Supplementary Table 2.

Cross correlational analysis

Pearson correlation analyses demonstrated no statistically significant associations among primary predictors or covariates (all p > 0.05), suggesting minimal collinearity within regression models. The pertinent correlation results are presented in Table 8, and the full correlation matrix can be found in Supplementary Table 3.

Table 8. Pearson correlation coefficients (r) among primary predictors and covariates used in multivariable regression analyses.

All values in format of r (p-value).

Discussion

In this study, we investigated QOL outcomes in CI users with SSD and if SQP is associated with quality-of-life outcomes. We found that CI recipients with SSD have CIQOL scores in the global, communication, and listening-effort domains below the scale midpoint, reflecting diminished QOL in these areas. However, subjective ratings of speech enjoyment, subjective ratings of speech understanding, and SQP scores (CI only, NH only, binaural CI+NH) were not significantly associated with global or domain-specific QOL. Comparable results were observed in the secondary sensitivity analyses controlling for CNC.

These results expand on prior research exploring the effects of SQP and SP among CI users with SSD. A meta-analysis by McRacken et al. in 2018 found that although SP is widely considered the gold standard for CI assessment because of its objective measurability, it is only weakly correlated with QOL improvements (McRackan et al., 2018). Multiple studies in the meta-analysis showed weak correlations between speech recognition (in quiet or noise) and QOL (McRackan et al., 2018). This is likely due to the limitations of traditional SP tests, which typically evaluate isolated words or sentences and fail to capture the complex listening demands of daily life (McRackan et al., 2018, 2019b; Dorismond et al., 2023; Capretta and Moberly, 2016). Consequently, many CI users may perform well on SP tests but still experience substantial QOL challenges. In line with these studies, our findings show that subjective speech understanding is not associated with most CI-related QOL among CI users with SSD.

Moreover, while focus group findings have emphasized that CI users who were happier with their listening experience after implantation reported greater engagement with people, media and the environment, increased self-confidence and reduced stress, our results suggest these improvements are likely not related to speech quality (McRackan et al., 2017). Specifically, these findings from prior focus groups describe improvements in the emotion, entertainment, and engagement domains of QOL. However, in SSD CI users in our cohort, none of these domains were significantly associated with subjective appraisal or standardized SQP scores.

This lack of association raises broader questions about whether speech-related mechanisms (e.g., SQP and SP) improve distinctly from improvements in QOL. Although SQP remains relatively understudied, prior work examining SP in non-SSD CI users has similarly suggested that speech performance and QOL may improve independently. One study demonstrated that while changes in SP did not strongly correlate with changes in QOL, baseline QOL predicted post-implant QOL, and baseline SP predicted post-implant SP (Chau et al., 2026). This study thus implied a parallel but potentially distinct trajectories of auditory performance and QOL improvements post-implantation. Furthermore, other work has shown that SP performance in CI users is most strongly predicted by sensory skills, whereas QOL outcomes were influenced by a more complex interplay of cognitive, sensory, and demographic factors (Skidmore et al., 2020). Thus, if SQP, similar to SP, is more closely grounded in sensory processing and follows a parallel improvement pattern based on baseline performance, this framework may help explain why SQP does not predict variability in QOL outcomes. Although these findings were reported in non-SSD populations, and our SSD population likely has a different contribution of HL to binaural processing, they offer a useful conceptual framework for interpreting our results in individuals with SSD.

In addition, previous studies in individuals with SSD have shown that SQP and emotional responses to music in the CI+NH condition is equivalent to that of a NH ear (Kelly et al., 2025; Alter et al., 2026), suggesting that in SSD CI users there is primacy of the NH ear centrally. This pattern partially aligned with our current findings: SQP in the NH ear was equivalent to binaural listening, but neither the NH-ear, CI-ear or binaural SQP scores predicted QOL. This suggests that while the NH-ear is the primary driver of SQP, this compensation is likely not a factor in driving the QOL in SSD CI users.

These findings emphasize that QOL in SSD CI users is complex and cannot be explained holistically by one isolated factor influencing speech. Clinically, these results are important as they challenge that speech-related metrics necessarily translate into meaningful gains in patient-reported QOL in SSD populations. Thus, these insights suggest that while there is an expectation based on clinical data to expect most significant improvements in speech metrics by 3–6 months, the QOL benefits may not follow this pattern as they are shaped by a broader interaction of psychosocial, cognitive, and demographic factors beyond what is tested with speech tools (Carlson, 2020; Ma et al., 2023; Neal et al., 2022). Recognizing this dissociation should be used to counsel patients accordingly, and supports more individualized rehabilitation that addresses multidimensional outcome improvement.

Despite its novel analyses, this study has several limitations. Firstly, due to the relatively small sample size, our analysis may not have captured all potential associations. Our study examined cross-sectional speech quality and quality-of-life scores and is therefore limited in its ability to assess longitudinal changes in SQP and CIQOL from pre- to post-implantation. Second, our study focuses on the SSD population, limiting extrapolation of its results to bilateral or bimodal CI users, Moreover, data logging about consistency of CI use was not assessed in our study, which has the potential to influence speech and quality-of-life outcomes. Our cohort also had limited diversity in race and education, making it difficult to generalize to a more diverse population. In addition, while linear regression was utilized to account for potential covariates, it is possible that other confounding variables not controlled for in our analysis—such as etiology of HL—may contribute to our outcomes of interest.

Nonetheless, our study poses several advantages. SQP and QOL were assessed with extensively validated tools specific to CI users. The focus on individuals with SSD allowed for a specific analysis of both implanted and NH ears and their potential contributions to QOL. Another strength of this study is the use of multivariable models adjusting for demographic and hearing-related factors, thereby strengthening the rigor of the statistical analysis. Additionally, sensitivity analyses incorporating CNC scores corroborated our findings, and cross-correlation testing demonstrated low collinearity among predictors, further supporting the robustness of our results.

In this study, we found that subjective ratings of speech, and measured SQP scores were not significantly associated with the entertainment, emotion, engagement or global QOL or entertainment, emotion, or engagement QOL sub-domains. These findings suggest that CI treatment strategies should not focus solely on speech-related outcomes in isolation, as improvements in QOL do not necessarily follow improvements in speech.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fauot.2026.1740450/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Hans Thomeer, Utrecht University, Netherlands

Reviewed by: Richard Charles Dowell, The University of Melbourne, Australia

Colleen Psarros, Macquarie University, Australia