Introduction

Literacy, oral language, and reading comprehension in deaf children

Literacy has long been defined as “the ability to understand the relationship between sounds and written words such that one may read, say, and understand them.” (p. 2)1. Reading is predicated upon verbal language ability and on the ability to process and “decode” the speech sounds written symbols represent. Because both the verbal language and speech processing underlying literacy development are supported by audition, it is not surprising that deaf children acquiring oral communication historically have struggled to develop language and reading skills. Oral language and reading are codependent developmental skills in all children2 and deaf children have long displayed difficulties in both domains. For example, a landmark study of reading abilities in children with severe-to-profound hearing loss conducted by Conrad3 in 1977 indicated that few reached rudimentary reading levels. Despite advances in hearing aid technology, there was substantially limited progress in reading abilities for the next two decades: Powers, Gregory, and Thoutenhoofd4 completed a comprehensive review of deaf education outcomes in 1999 and noted “There is no substantive evidence to demonstrate any overall improvement in the education of deaf pupils since Conrad’s study.” (p. 8). At a fundamental level, limited access to speech input due to severe-to-profound hearing loss continued to have significant impacts on the development of oral language and literacy. More recently, advances in cochlear implant (CI) technology have dramatically improved overall language and reading outcomes. As an example, some researchers report that, although verbal grammatical abilities and a measure of reading comprehension for pediatric CI users were significantly below a comparison group of children with normal hearing, the performance of the CI group was improved relative to previous studies of deaf children5. A recent meta-analysis examining language and reading abilities for pediatric CI users also yielded encouraging results because the gaps between children with and without hearing loss had narrowed. However, the analysis also revealed that there were important differences remaining between the groups with significant deficits and delays for children with CIs6. Estimates of aggregated effect sizes using Hedge’s g7 in this meta-analysis were -0.40 for receptive vocabulary, -1.57 for expressive vocabulary, and -1.39 for reading comprehension. This is indicative of a moderate effect size for receptive vocabulary deficits and very large negative effect sizes for expressive vocabulary and reading comprehension.

Phonological processing

In addition to oral language and literacy, the ability to process, identify in text, and mentally operate on units of speech (phonemes) has been implicated in learning to read and in reading-related learning disabilities among children with CIs8,9. There is also an extensive literature indicating that phonological processing is important for language acquisition and that weaknesses contribute to developmental language disorders in children without hearing loss10,11,12,13. Similarly, models of reading development, such as the “simple view of reading” and the “scientific view of reading”, include oral language and phonological processing as codependent contributors to reading comprehension ability14,15,16.

Notably, there is inconsistent nomenclature on phonological processing in the literature, which can make interpreting findings across studies and theoretical models difficult. Specifically, the nomenclature associated with “phonological processing” includes a variety of designations and definitions. Related terms include speech processing, phonological awareness, phonemic awareness, decoding, and in some contexts “auditory processing.” Indeed, the Test of Auditory Processing employed herein (TAPS-4)17 includes multiple measures of phonological processing and yields a phonological processing index subscale.

Herein, the broad construct “phonological processing” included measures of the following: 1) speech discrimination; 2) “decoding,” defined as “simply efficient word recognition: the ability to rapidly derive a representation from printed input that allows access to the appropriate entry in the mental lexicon, and thus, the retrieval of semantic information at the word level”18; 3) phonological awareness, defined as metalinguistic knowledge of the sound-phonetic structure of a language (in addition to phonological underpinnings of reading); 4) phonemic awareness, defined as specific knowledge that “spoken language consists of a sequence of phonemes”19 (p. 130). Phonemic awareness is evaluated by asking participants to provide information on the phonemic structure of words and syllables. Moreover, the role of phonological processing in reading has been extensively evaluated by reading scientists and neuroscientists20,21. It is generally shown to be a crucial element of reading ability in typical development and in reading-related learning disabilities.

Children with CIs are intriguing from a scientific perspective because the signal delivered to the auditory nerve, and eventually to the auditory cortex, is relatively sparse. This results in important differences in spectral resolution (i.e., the ability to resolve the frequency components of speech) capabilities22,23,24. Precise resolution of phonemic boundaries is likely problematic in CI users. Indeed, deficits in spectral resolution have previously been implicated in phonological processing deficits among children with CIs25,26. Because of the potential impact of CI signal processing on phonological skills, there is a unique opportunity to examine relation of phonological processing to language and reading comprehension among children with CIs. As noted previously, there is robust literature on this association between phonological processing, language, and reading27.

Literacy and reading comprehension

The construct “literacy” is often defined as the ability to understand written language (i.e., to understand what is read). Although some definitions also encompass writing ability and spelling1,28, which were not assessed herein, the focus of this study is nonetheless on a primary product of literacy: reading comprehension. This is directly predicated upon widely accepted models of literacy that include reading comprehension as a foundational literacy skill1,29,30. The widely used “simple view” of reading includes reading comprehension as the direct product of word recognition, including phonological processing, and language comprehension14,18. Indeed, Gough and Tunmer argued that reading disability, defined as difficulties in reading comprehension, arises from disruptions of “word recognition” as well as language comprehension. Recent updates and more nuanced revisions of the simple view of reading continue to include reading comprehension as a core literacy outcome16,31.

Gaps in the literature, research questions, and hypotheses

Currently, an extensive literature on phonological processing, oral language development, and reading indicates that deficits in phonological processing are strongly associated with impairments in language acquisition and poor reading outcomes32,33,34. Historically, this has also been reported in children with severe-to-profound hearing loss. More recently, there have been reports that some children with severe-to-profound hearing loss who use CIs have displayed improved oral language and reading outcomes. However, deficits persist, albeit of lower magnitude than reported historically for deaf children5,27,35,36,37,38,39 The extent to which these improvements are predicated on similarly improved phonological processing abilities is an intriguing gap in the current literature39. The following research questions are addressed:

  1. 1.

    Are oral language, phonological processing, and reading comprehension abilities in this sample of children with severe-to-profound hearing loss using CIs significantly below age-normative levels on standardized measures of these parameters? Based on the extant literature, one could hypothesize that all three of these domains would yield below average scores on norm-referenced tests. However, there are reports of language and reading trending into the typical range albeit largely below the average norm referenced scores8,39.

  2. 2.

    Are there significant discrepancies between oral language, phonological processing, and reading comprehension abilities in this sample of children with severe-to-profound hearing loss using CIs? In addition to hypothesizing that these domains are all below levels reported in children without hearing loss, we hypothesize that, in accord with the simple view of reading, deficits in phonological processing and language abilities will be equal in magnitude to deficits in reading comprehension.

  3. 3.

    Are there significant associations among oral language, phonological processing, and reading comprehension levels in this sample of children with severe-to-profound hearing loss using CIs? We hypothesize that there will be strong associations among these domains and that oral language and phonological processing abilities will be significant predictors of reading comprehension.

  4. 4.

    Is spectral resolution associated with phonological processing abilities in this sample of children with severe-to-profound hearing loss using CIs? In accord with theoretical models and data from groups with and without hearing loss25,26, one would predict that spectral resolution will be associated with phonological processing abilities for words and nonwords. However, given the inherent limitations in spectral resolution for CI users, one could also plausibly predict that this parameter is not associated with phonological processing abilities.

Methods

Portions of the methods descriptions herein have been included elsewhere in our publications22,40 and are adapted for this article. Some aspects of the descriptions have been directly replicated when appropriate (e.g., when applicable, participant assessments and eligibility parameters).

Participants

Participants were between the ages of 4.5 and 13.1 years old at the time of enrollment and had prelingual onset of hearing loss. All participants had bilateral sensorineural hearing loss for which they received at least one CI prior to the age of six (mean: 1.75 years; range: 0.50 to 5.58 years). Forty-three participants (91.4% of the sample) were bilateral CI users whereas four used a bimodal hearing configuration (CI plus contralateral hearing aid). Demographics can be found in Table 1. All participants utilized auditory-oral methods as the primary mode of communication and came from families that used spoken English. One participant came from a multilingual household; however, English was the primary language used in the home. In keeping with eligibility criteria for Early Intervention services, all children received birth to three monitoring and support for hearing loss, and all had special education eligibility for hearing loss. Additionally, one participant was classified as visually impaired although his near point vision was sufficient to complete the testing. One participant met eligibility for other health impaired for inattention and impulsivity but was not taking ADHD medication.

Table 1 Participant information (n = 47 unless otherwise noted).
Full size table

Data were collected between the years of 2019 and 2022 during the baseline assessments of 47 children with CIs who were enrolled in a larger longitudinal study examining the effects of image-guided cochlear implant programming (IGCIP)41. Prior to baseline assessment, participants’ standard clinical CI maps had been programmed using electrically evoked stapedial reflex thresholds (ESRTs) to set device upper stimulation levels; lower stimulation levels were verified by ensuring CI aided-detection thresholds were in the range of 20–30 dB HL from 250 through 6000 Hz in a sound-attenuated booth.

Procedures

The methods herein adhered to international guidelines for protecting the rights of vulnerable populations and were approved by the Vanderbilt University Institutional Review Board (IRB #190095). Informed consent from a parent/guardian and informed assent from participants were obtained prior to data collection in accord with ethical procedures for participation in research studies.

Speech, language, and reading comprehension assessments were administered by a board-certified speech language pathologist (CCC-SLP) in a quiet room. Of the 10 assessments in included in the analyses, portions of two, the second edition of the Comprehensive Test Of Phonological Processing (CTOPP-2)42 and the fourth edition of the Test of Auditory Processing Skills (TAPS-4)17 were administered via a standardized audio recording, while the remaining 8 were administered in live voice. For all assessments that were administered in live voice, the tester’s face was visible to participants, and for recorded tests, the volume level was selected by the participant to ensure audibility. Audiometric assessment was carried out prior to speech, language, and reading assessment to assure appropriate device programming and audibility. Testing was either carried out in a single session or in two sessions spread across two consecutive days. Each test session lasted between two to six hours; for longer sessions (> 4 h), participants and their families were provided a lunch break.

Phonological processing, language, and reading comprehension assessments

Language ability

Language ability was measured at two levels: expressive and receptive. Additionally, estimates of each domain included multiple measures including vocabulary, morphology, and syntax. Receptive language abilities were measured using the Receptive One-Word Picture Vocabulary Test-4, (ROWPVT-4)43, and the Test of Auditory Comprehension of Language-4 (TACL-4)44, which includes separate subscale scores for vocabulary, morphology, and elaborated sentences. Expressive language was measured using the Expressive One-Word Picture Vocabulary Test-4 (EOWPVT-4)45, the Structured Photographic Expressive Language Test-3 (SPELT-3)46, and the core language index on the Clinical Evaluation of Language Fundamentals-5 (CELF-5)47.

Nonverbal assessment of cognition

Nonverbal cognition was assessed using the 3rd edition of the Leiter International Performance Scale (Leiter-3)48.

Phonological processing

Phonological processing is defined as the ability to segment, discriminate, and operate on phonological units (speech sounds)32. Phonological processing was characterized using the Comprehensive Test Of Phonological Processing (CTOPP-2)42, which is a norm-referenced and widely used standardized assessment of this ability. To obtain multiple standardized estimates of phonological processing, we also administered the Test of Auditory Processing Skills (TAPS-4)17.

Reading comprehension

Reading comprehension ability was assessed using two standardized tests: the Woodcock Reading Mastery Tests (WRMT™-III)49, and the Gray Oral Reading Test-5 (GORT-5)50. Both instruments have been widely used with typically developing children and children with disabilities and, as with phonological processing, have been applied to children with hearing loss in several studies. The WRMT™-III includes phonics assessments (e.g., letter word identification) and assessment of reading comprehension (e.g., Passage comprehension). Similarly, the GORT-5 includes estimates of reading fluency, accuracy, and reading comprehension. For this analysis, only reading comprehension scores from the WRMT™-III and the GORT-5 were included.

Aggregated estimates of parameters

Five constructs were measured to examine the relationships among phonological processing, language, and reading comprehension skills in pediatric CI users: Receptive Language (RL), Expressive Language (EL), Phonological Processing for Words (PPW) Phonological Processing for Nonwords (PPN) and Reading Comprehension (RC; Fig. 1). In accord with aggregating multiple single measures of key parameters for stable estimates of these measures51,52, the following multiple subtests were utilized to estimate each construct: Receptive Language (RL) included the ROWPVT-4 and the TACL-4; Expressive Language (EL) included the EOWPVT-4, SPELT-3 and the Core Language Score from the CELF-5; Phonological Processing for Words (PPW) included the Elision, Blending Words, and Phoneme Isolation subtest scores from the CTOPP-2, and the Word Discrimination, Phonological Deletion, and Phonological Blending subtests from the TAPS-4; Phonological Processing for Nonwords (PPN) included Nonword Repetition, Nonword Blending, and Nonword Segmentation subtests from the CTOPP-2; RC included the Reading Comprehension cluster score of the WRMT™-III and the Comprehension subtest of the GORT-5.

Fig. 1
figure 1

Path diagram of construct domains and sub-scores. ROWPVT-443 = Receptive One-Word Picture Vocabulary Test (4th edition); TACL-444 = Test of Auditory Comprehension of Language (4th edition); EOWPVT-445 = Expressive One-Word Picture Vocabulary Test (4th edition); SPELT-346 = Structured Photographic Expressive Language Test (3rd edition); CELF-547 = Clinical Evaluation of Language Fundamentals (5th edition); CTOPP-242 = Comprehensive Test of Phonological Processing (2nd edition); TAPS-417 = Tests of Auditory Processing Skills (4th edition); WRMT™-III49 = Woodcock Reading Mastery Tests (3rd edition); GORT-550 = Gray Oral Reading Test (5th edition).

Full size image

Spectral resolution assessment

Spectral resolution was assessed with a spectral modulation detection (SMD) task. This task was administered via MATLAB in a sound-attenuated booth with the stimuli presented at 65 dBA from a loudspeaker one-meter from the participant at 0-degree azimuth. SMD thresholds (dB) were assessed using an adaptive three interval, two-alternative forced choice (2AFC) task in which participants were asked to distinguish between an unmodulated and a frequency-modulated noise band (frequency range: 125–5600 Hz) at a modulation rate of 0.5 cycles per octave. In each trial, participants were first presented an unmodulated noise band as a reference interval. This was then followed by two test intervals, one that contained the frequency-modulated stimulus and one that contained the same unmodulated stimulus used in the reference. Participants were instructed to identify which of the two test intervals sounded different than the reference interval by selecting a box labeled ‘2’ or ‘3’ via a graphical user interface (GUI) on a touchscreen. Modulation depth was changed in an adaptive 2-up 1-down procedure. For example, if participants correctly selected the interval with the frequency-modulated stimulus twice in a row, the modulation depth of the target interval would decrease, making the task more difficult. The adaptive step-size began at 4 dB but was reduced to 2 dB following the first two reversals. A 70.7% psychometric threshold was calculated by averaging the modulation depths at the final six reversal points of the task53. At least two runs of the task were completed per participant, with a third run administered if there was a greater than 5-dB difference in SMD threshold between the initial two runs. The SMD thresholds of the completed runs were then averaged. During this assessment, participants only used a single CI-processor, specifically that of the later-implanted or poorer performing ear in the case of bilateral implantation.

Statistical analysis

Dependent measures of language, phonological processing, and reading comprehension included quotients (M = 100, SD = 15) and scaled scores (M = 10, SD = 3). In order to put these on the same scale for integrated analyses, z-score transformation (M = 0, SD = 1) was completed. These scores were averaged together to form domain scores for the constructs outlined previously. Normality of domain scores was assessed using Shapiro–Wilk criterion54 (p > 0.05). For a full list of the analyzed measures and the distribution of test scores, see Table 2. All analyses were completed using R (v4.4.1)55 and with tidyverse (v2.0.0)56 packages for data organization. The relation between the five domains and constructs were assessed using Pearson correlation analyses with the Bonferroni Correction applied for multiple comparisons. Additionally, the relation between spectral resolution and the phonological processing domains (both Word and Nonword) was assessed using Pearson correlation analyses. Comparison of domain performance across participants was completed using a linear mixed-effects model with by-participant random intercepts. To identify which of the four language domains were the strongest predictors of reading comprehension skills, multiple linear regression was used. Missing data were handled using listwise deletion.

Table 2 Score distributions.
Full size table

Results

Score distribution

For the 18 tests and subscales included in the analysis, between 28 and 47 participants completed each assessment. Both standard scores and z-score conversions of assessment performance can be found in Table 2. As outlined in the methods, the 18 tests and subscales were divided into five domains (Table 3). All domain scores were found to be normally distributed. Outlier analysis was completed using the rstatix package (v0.7.2)57 and found no extreme outliers among domain scores. Scores could be calculated for all 47 participants for RL, EL, PPW, and PPN domains. However, it should be noted that the number of component z-scores within the aggregated domains varied as standard scores could not be obtained for participants below or above the normed age range of any given assessment. Specifically, our participant age range exceeded that provided in the CELF-5, TACL-4, SPELT-3, WRMT™-III, GORT-5, and CTOPP-2 for certain subscales. For the CELF-5, WRMT-III™, GORT-5, and CTOPP-2, our sample contained participants younger than the normative range; for the TACL-4 and SPELT-3, our sample contained participants older than the normative range. Thirty-nine participants contributed scores to the RC domain, as eight participants were unable to complete the WRMT™-III or GORT-5 due to age at the baseline assessment. The highest mean domain score was obtained for RC (M =  − 0.20; 95% CI [− 0.58, 0.18]) and the lowest mean domain score was obtained for PPN (M − 1.90; 95% CI [− 2.10, − 1.70]). Only scores for the PPW and PPN, not RL, EL, or RC, fell with a clinically significant deficit range (z ≤ -1.00).

Table 3 Construct domain scores.
Full size table

Comparison across domains

Linear mixed-effects models using the packages lme4 (v1.1.37)58 and jtools (v2.2.0)59 were completed to compare participant performance across the five construct domains (Fig. 2). While these five domains were our primary comparisons of interest, the impact of age, as well as the interaction between age and domain, on differences in participant performance were initially assessed using hierarchical modeling. Backward stepwise comparisons between models using the function anova() found no evidence of significant improvements in model fit with the inclusion of age \(\left({\chi }^{2}=1.34, p=0.25\right)\) or the interaction between age and domain \(({\chi }^{2}=1.48, p=.83)\). Therefore, a model with construct domain as the only predictor of scores was used to complete post-hoc comparisons between domain performance. In this single predictor model, domain was found to have a significant effect on participant score \(\left(F=80.92, p<.01,\, pseudo\, {R}^{2}=0.35\, for\, fixed\, effects\right).\)

Fig. 2
figure 2

Histograms of construct domain scores. Dashed vertical line: sample mean; solid line: z-score of 0; curve: approximation of a Gaussian distribution.

Full size image

The emmeans package (v1.11.1)60 and Tukey’s Honest Significant Difference (HSD) tests were used to compare participant scores across domains. RL scores were higher than PPW (mean difference = 0.71) and PPN scores \((mean\, difference=1.66)\). EL scores were higher than PPW \((mean\, difference=0.73)\) and PPN scores \((mean\, difference=1.68)\). PPW scores were higher than PPN \((mean\, difference=0.95\) scores but lower than all remaining domain scores (p < 0.01). PPN scores were lower than all other domain scores (p < 0.01). RC scores were higher than PPW \((mean\, difference=0.86)\) and PPN \((mean\, difference=1.81)\) scores. There were no significant differences in scores between RL, EL, and RC domains. Statistics for all pairwise comparisons can be found in Table 4.

Table 4 Linear mixed-effects model and post-hoc results (n = 39).
Full size table

Phonological processing and language skills as predictors of reading comprehension levels

To assess the unique variance provided by language and phonological processing as predictors of reading comprehension (Fig. 3), multiple linear regression analyses were completed. Our a-priori model and predictors of interests were limited to language and the phonological processing domains. However, hierarchical regression was also completed initially with age, as well as the interaction between age and the construct domains, as predictors to assess potential moderating effects of age on the relationship. Based on considerations for statistical power, the correlation and repeated measures ANOVA results herein, and prior theory, two series of regression models were evaluated. In all models, subscale and test scores in the domains RL and EL were averaged into a single “language” domain. This was done in accordance with established theory regarding their shared higher-level construct as well as with the high correlation \((r= .82)\) and comparable performance between the RL and EL domains found in this dataset (\(mean\, difference=\pm 0.02)\). In the first series of hierarchical comparisons, PPW and PPN were also combined into a single phonological processing domain (PP), given extent evidence that both domains fall within this overall latent ability61. However, these domains represented separate predictors in the second series of model comparisons. This was due to the fact there was a significant discrepancy in performance for these two domains within the dataset, such that performance in the PPW domain was significantly higher than performance in the PPN domain (\(mean\, difference=0.95)\). Language, PP, and PPW were found to be significantly correlated (p < 0.003) with RC scores following the Bonferroni correction for multiple comparisons (Table 5).

Fig. 3
figure 3

Scatterplots of construct domain score correlations.

Full size image
Table 5 Correlation matrix for reading comprehension regression models (n = 39).
Full size table

In the first series of models, the impact of three main effects and two interaction effects on reading comprehension were initially considered (language, PP, age, language x age, PP x age). Backward stepwise comparisons between models using anova() found no evidence of significant improvements in model fit with the inclusion of age \(\left(F=0.02, p=0.88\right)\) nor the interactions between age and the two construct domains \((range\, of\, p-values: .38-.93)\). Therefore, there was evidence to support our a-priori model with only domain predictors (language and PP) independent of additional age effects (Table 6). A significant proportion of variance in the RC domain scores was accounted for by this model (F(2,36) = 44.07, p < 0.01, R2 = 0.69). Language was the only significant predictor of RC scores (B = 0.93, p < 0.01). PP was not a significant predictor.

Table 6 Regression model for reading comprehension skills (Two predictors).
Full size table

For the second model, four main effects and three interaction effects were initially considered (language, PPW, PPN, age, language x age, PPW x age, PPN x age). Backward stepwise comparisons between models using anova() found no evidence of significant improvements in model fit with the inclusion of age \(\left(F=0.00, p=0.95\right)\) nor the interactions between age and the three construct domains \((range\, of\, p-values: .60-.94)\). Therefore, there was evidence to support our a-priori model with only domain predictors (language, PPW, and PPN) independent of additional age effects (Table 7). A significant proportion of variance in the RC scores was accounted for by this model (F(3,35) = 33.30, p < 0.01, R2 = 0.72). Language was the only significant predictor of reading comprehension scores (B = 0.82, p < 0.01). Neither PPW nor PPN were significant predictors.

Table 7 Regression model for literacy skills (Three predictors).
Full size table

Spectral resolution and phonological processing skills

Analysis of the relation between spectral resolution, as measured using SMD threshold, and PPW and PPN performance was completed (Fig. 4) using Kendall’s Tau62 as SMD thresholds were not normally distributed (p < 0.0554). SMD threshold was not significantly correlated with PPW performance (r =  − 0.14, p = 0.19) nor PPN performance (r = 0.10, p = 0.36).

Fig. 4
figure 4

Scatterplot of spectral resolution and phonological processing skills. PPW = Phonological Processing (Word); PPN = Phonological Processing (Nonword); SMD = Spectral Modulation Detection.

Full size image

Discussion

The following research questions were addressed, and these were the findings:

  1. 1.

    Are oral language, phonological processing, and reading comprehension abilities in this sample of children with severe-to-profound hearing loss using CIs significantly below age-normative levels on standardized measures of these parameters? The results indicated receptive and expressive language, as well as reading comprehension abilities were within the broad range of typical performance and not significantly below age-normative levels. However, phonological processing abilities fell significantly below normative levels.

  2. 2.

    Are there significant discrepancies between oral language, phonological processing, and reading comprehension abilities in this sample of children with severe-to-profound hearing loss using CIs? The results indicate that phonological processing abilities were significantly below measures of oral language and reading comprehension ability.

  3. 3.

    Are there significant associations among oral language, phonological processing, and reading comprehension levels in this sample of children with severe-to-profound hearing loss using CIs? While the results indicate that oral language, phonological processing, and reading comprehension abilities were positively correlated, only oral language ability was a significant predictor of reading level.

  4. 4.

    Is spectral resolution associated with phonological processing abilities in this sample of children with severe-to-profound hearing loss using CIs? The results indicate that spectral resolution was not significantly correlated with phonological processing abilities for words or nonwords.

As noted in the introduction, phonological processing has long been viewed as a crucial skill for developing language and as a necessary path to reading comprehension, often weighted at least equally with oral language comprehension14,15. Using existing paradigms, a priori predictions for oral language comprehension and for reading comprehension would posit poor abilities in these linguistic domains for deaf children with poor phonological processing skills63,64,65. And, in fact, there are also substantial data to support this thesis in children with typical hearing who are poor decoders66,67,68. However, one interpretation of the literature could plausibly suggest that although phonological processing relates to word recognition and to efficient decoding, this does not necessarily ensure reading comprehension. A straightforward example of this can be seen in hyperlexia, wherein decoding and word recognition abilities are seen in the absence of reading comprehension69. The results of this study indicate that the reverse can also be true in children with CIs: Reading comprehension can be achieved despite relatively poor phonological processing skills.

In terms of reading comprehension, for more than 30 years, “the simple view of reading,” has been a foundational model of reading acquisition and reading disability14,15,70. There is ample evidence in the literature supporting this hypothesis in children with and without hearing loss, including those who are advanced, average, and poor readers21. Similarly, the literature on language and reading acquisition in deaf children has long illuminated core challenges in this population, with the overwhelming majority of these children struggling to reach proficiency in reading comprehension and in oral language3,4,5,70.

Children with CIs are an important population to reexamine the relative contribution of phonological processing to oral language and reading comprehension given the auditory signal differences in CIs as compared to the auditory system of a child without hearing loss. Specifically, although CIs do provide significant access to the auditory signal, spectral resolution is significantly poorer among children with CIs in comparison to children without hearing loss24. Access to the speech signal available to children with CIs is limited to, at most, up to 12 independent spectral channels across 12 to 22 intracochlear electrodes71. Furthermore, temporal fine structure critical for perception of spectral and temporal contrasts with phonemic transitions is poorly represented by the envelope-based signal processing strategies used in modern CI systems9,24,71. Given limitations in the discrete number of spectral channels, channel independence, and envelope-based CI signal coding, we should expect an attenuation of the spectral signal underlying speech perception and associated phonological processing inputs72. Thus, one would also expect children with CIs to display significant deficits not only in phonological processing, but also in oral language and reading comprehension. However, as we and others have shown, this is not necessarily the case8,39,73.

The results of this study indicate unusual and unexpected asymmetries in phonological processing, oral language, and reading comprehension that are not directly aligned with predicted contributions of phonological processing skills to reading comprehension. Specifically, comparisons of mean performance levels in language, reading, and phonological processing in the CI children herein show language skills and reading comprehension skills within the typical range (uniformly less than ½ deviations below the normative mean) whereas phonological processing skills, both with real words (e.g., CTOPP-2 Blending words M =—1.55 deviations) and with pseudowords (nonwords, e.g., CTOPP-2 Nonword Repetition M =—2.33 deviations) are significantly poorer. Indeed, phonological processing falls within the clinically significant deficit range for individual participants and for the group (e.g., more than 1.5 or even 2 SD below normative means). Longitudinal studies of phonological processing acquisition indicate that children with significant hearing loss, including those with CIs, do make progress in phonological processing skills, but do not attain typical ability levels and even rudimentary milestones are acquired much later8,73. It is certainly possible, and testable, that reaching a minimal “threshold” level of phonological processing abilities within the limits of CI signal processing is a necessary precursor for language acquisition and reading comprehension.

Furthermore, our results suggest that the relation between oral language, phonological processing, and reading comprehension abilities is different among pediatric CI users relative to the literature on their peers without hearing loss. Although performance in the domains of language and phonological processing for words was positively correlated to reading comprehension, language ability alone was a significant predictor of reading. Given the relative weakness of phonological skills in comparison to language skills for this sample, we theorize that children with CIs may rely more on language skills, for example vocabulary or morphosyntax, in developing reading comprehension. This is not to say phonological processing is not important to acquisition and maturation of all skills related to reading. The contribution of phonological processing abilities may be more important for literacy skills outside of reading comprehension, for example phonics and/or reading fluency which were not included in the reading comprehension composite score herein.

Additionally, it should be noted that our language and phonological processing composites are substantially correlated in this sample. These correlations likely affected model regression coefficients and may suggest that there is a broader general ability that accounts for shared variability within these domains. The fact that nonword scores were particularly low and unrelated to reading comprehension further underlines the potentially different relations between phonological processing, language, and reading skills among children with CIs. It suggests that the association between phonological processing for words and reading may reflect the bootstrapping of other linguistic abilities, such as semantic knowledge, given the absence of a significant relation for nonword tasks. Of particular interest is the possibility of alternate neural cognitive processes and pathways linking these language and phonological processing domains to reading comprehension68.

Although the relation between spectral resolution and speech understanding is not clearly established for children with CIs22,23,74, the degraded spectral signal of CIs has been suggested as a contributing factor in the deficits in phonological skills for pediatric CI users25,26. The exploratory analyses of SMD threshold herein do not provide evidence for a relation between spectral resolution and phonological processing skills. However, the question still remains as to whether improvements in spectral resolution could impact phonological skills in this population. A future direction of this line of research would be to address these relations using longitudinal data. Additionally, it is possible that specific phonological tasks (e.g., elision) exhibit strong relations with spectral resolution. Because we instead used aggregated measures of phonological processing for this analysis, this may account for the discrepancy between our results and previous findings on the relation between spectral resolution and phonological skills25,26.

This study included a total of 47 children with CIs, and although this is one of the larger studies of this population, replication with a larger sample is certainly warranted. Because this sample is well characterized from surgical procedures for the CIs, audiological and speech processing measures, and well accepted and validated measures of language, reading comprehension, and phonological processing (with multiple estimates of each parameter), we hypothesize that replications will yield similar results. Additionally, this study’s sample, while well characterized, may not generalize to all children who use CIs. Most participants did not have additional disabilities or development differences, were consistent bilateral device users (12 + hours per day), received consistent early-intervention supports, and primarily utilized auditory-oral communication. Additionally, while recruited from many locations in the United States, this sample was largely homogenous in terms of race, ethnicity, and language background.

An additional limitation is the variability in normative age ranges in standard and scaled scores across assessments. For example, the SPELT-3 does not include normative values for children over the age of 10 and the Phoneme Isolation subscale on the CTOPP-2 does not include normative values for children under the age of seven. This resulted in missing data and variability in the number of assessments included within composite scores for participants. While these age-related restrictions for norms occurred at both extremes of participant age range, it’s possible that this, and any other missing data, have affected results herein.

There are number of intriguing directions for future research. First, the finding that deaf children with CIs can acquire typical levels of language ability and reading comprehension skills despite strikingly poor phonological processing skills that are usually associated with significant language and reading impairments provides a unique opportunity for uncovering alternative and compensatory behavioral and neural pathways to language acquisition and to reading ability. Key questions include identifying the nature of phonemic underpinnings for language and reading as well as the neural processing pathways for speech in prelingually deaf children. Additionally, reconsidering foundational models of neuroplasticity and, potentially, alternate multisensory (audiovisual) phoneme representations in children with CIs is warranted. It will be very interesting to determine which processes, auditory and visual, are being employed to “bootstrap” language acquisition and reading ability.

There are also intriguing opportunities to rethink and evaluate language intervention and reading instruction in these children. Detailed behavioral and instructional intervention records were unavailable across all participants. However, the ubiquity of phonemic awareness, decoding, and other forms of phonological processing instruction in children with hearing loss strongly suggests that many of these children received relatively extensive exposure to these services. Thus, important questions for future research include systematic study of whether this instruction indirectly contributes to language acquisition and reading abilities despite the relatively low phonological processing levels attained.

Finally, naturalistic transactional interventions based on emphasizing key aspects of speech, morphological, and syntactic supports have an extensive evidence base in children with learning challenges, including autism, Down syndrome, speech disorders, and language disorders. The findings herein, where children with CIs attain typical or near typical language and reading levels, suggest that future research can illuminate how these very effective interventions and instructional methods may be applicable to deaf children with CIs.

Conclusions

Normative assessments of phonological processing, language, and reading comprehension were administered to 47 deaf children with CIs (mean age = 8.33 years). The results indicated that phonological processing abilities in children with CIs are strikingly and significantly below expected levels and that there are strong associations between phonological processing, language, and reading skills. Another result indicated that language abilities and reading comprehension scores were often within the typical range and significantly above phonological processing levels. Although there have been reports of some prelingually deaf children with CIs attaining typical or near typical levels of reading8,39, this finding is intriguing because hearing children with similar deficits in phonological processing would be highly likely to display significant deficits in language and reading. It is noteworthy that the sparse signal inherent in CIs did not prevent many of these prelingually deaf children from acquiring relatively typical language and reading comprehension scores despite poor phonological processing ability. This suggests that unknown alternate pathways of skill acquisition are being utilized. Unlike previous studies of reading in prelingually deaf children using CIs, including those indicating relatively strong reading abilities, a persistent weakness in phonological processing that often results in poor reading in hearing children is highlighted herein.