Early intervention influences 9-year speech, language, cognitive and quality-of-life outcomes in deaf or hard-of-hearing children

Abstract

Objectives.

Early identification of congenital deafness enables early intervention, but evidence on the influence of age at fitting of hearing aids (HAs) or cochlear implants (CIs) on outcomes in school-aged children who are deaf or hard of hearing (DHH) is limited. This study 1) described developmental outcomes and health-related quality of life in DHH children; and 2) examined the relationships among demographic factors, including age at fitting of HAs or CIs, and outcomes.

Design.

This prospective cohort study included participants in a population-based study who were followed up at 9 years of age. Children who are DHH and who first received hearing habilitation services before 3 years of age from the government-funded national hearing service provider in the states of New South Wales, Victoria, and Southern Queensland in Australia were invited to enroll in the study. At 9 years of age, enrolled children were assessed using standardized measures of language, cognitive abilities, and speech perception. The children also completed questionnaire ratings on their quality of life. Parents provided demographic information about their child, family, and education; and completed ratings on their child’s quality of life. Audiological data were retrieved from the client database of the hearing service provider and records held at CI centers. Descriptive statistics were used to report quantitative outcomes. The relationships among demographic characteristics, including age at fitting of HAs or CIs, and children’s outcomes were examined using structural equation modelling.

Results.

A total of 367 children, 178 (48.5%) girls, completed assessments at age 9.4 (SD 0.3) years. On average, performance was within one SD of the normative mean for language, cognitive functioning, and health-related quality of life; but much below norms for speech perception. The modelling result is consistent with verbal short-term memory having a mediating effect on multiple outcomes. Better verbal short-term memory is significantly associated with no additional disabilities, earlier age at CI activation, use of an oral communication mode in early intervention, and higher maternal education. In turn, verbal short-term memory directly and positively affects speech perception, language and health-related quality of life. Maternal education directly and positively affects language outcomes, and indirectly via its effects on nonverbal IQ and verbal short-term memory. Better language is directly associated with better quality of life.

Conclusions.

This study found evidence consistent with early hearing intervention having a positive effect on speech perception and language via its effect on verbal short-term memory. Children who had better language also had better quality of life. The importance of early hearing for cognitive development lends support to early detection and early hearing intervention, including streamlining pathways for early CI activation. Strategies for intervention in language and communication development may benefit from tailoring programs to meet the needs of individuals with different memory profiles for optimizing outcomes.

Introduction

Permanent childhood hearing loss, which affects about 3 per thousand children (Mehra et al., 2009), has major adverse developmental and health impacts on children’s lives (Kronenberger et al., 2014; AuBuchon et al., 2015; Tomblin et al., 2015; Roland et al., 2016; Ching et al., 2017; Nittrouer et al., 2017; Wong et al., 2017). Consequently, universal newborn hearing screening (UNHS) programs have been implemented to detect children who are deaf or hard of hearing (DHH), so that intervention can be initiated to improve outcomes (Joint Committee on Infant Hearing, 2007, 2013, 2019). Earlier fitting of hearing aids (HAs) or cochlear implants (CIs) was associated with better language development at pre-school years (Geers et al., 2009; Niparko et al., 2010; Sininger et al., 2010; Tomblin et al., 2015; Dettman et al., 2016; Ching et al., 2017). However, there is limited evidence on the efficacy of early intervention for improving school-age outcomes of DHH children at a population level (Thompson et al., 2001; Nelson et al., 2008; Pimperton & Kennedy, 2012; Duchesne & Marschark, 2019). In a summary of evidence from the systematic review conducted by the US Preventive Services Task Force, Thompson et al. (2001) concluded that the efficacy of UNHS to improve long-term language outcomes remains uncertain. They identified important methodological limitations in previous studies on benefits of early intervention, including among others the use of convenience samples. The systematic review called for population-based studies to prospectively evaluate whether long-term language outcomes of children improve as the age of identification decreases. Nelson et al. (2008) provided an update of the review with two additional studies (Kennedy et al., 2006; Wake et al., 2005). The updated review reiterated the need for more research on the effectiveness of early detection and intervention on not only language but also long-term functional outcomes, including quality of life.

The primary outcomes of interest in the current study were language and health-related quality of life (HRQOL), and secondary outcomes included verbal short-term memory and speech perception. In a trial of UNHS in Wessex, Kennedy et al. (2006) reported that DHH children confirmed by age 9 months (n = 45) had better receptive and expressive language at age 6–10 years, compared to those confirmed later (n = 56). On average, children who were confirmed before 9 months of age had receptive language ability at school age that was −1.76 standard deviation (SD) below their hearing peers, and expressive language ability of the early-confirmed group was −0.59 SD below hearing peers. An advantage of 0.6 SD was observed for early confirmation (Kenney et al., 2006; Stevenson et al., 2011). When 27 of the early-confirmed group and 33 of the later-confirmed group were followed up at 13–19 years (Pimperton et al., 2017), the mean composite Z scores for receptive language were −1.6 SD and −2.4 SD respectively, and the mean expressive information Z scores were 0.004 and −0.33 SD for the two groups. There were no significant differences in receptive or expressive language abilities between the early and later-confirmed groups. Wake et al. (2016) reported a comparison of outcomes of three approaches to detection of congenital hearing loss. On average, 69 children identified via UNHS had receptive language, expressive language, and receptive vocabulary at age 5 years around −1.3 SD below hearing peers. For children without intellectual disability, the benefits of UNHS relative to opportunistic screening were around 0.5 SD, 0.9 SD, and 0.8 SD for receptive language, expressive language and receptive vocabulary respectively. As children grow, the effectiveness of early intervention may alter with outcomes being measurable in a wider range of domains than was possible at a younger age.

Quality of life refers to an “individual’s perception of health status, psycho-social status and other aspects of life” (World Health Organization, 1998). Although overall quality of life was not assessed in the Wessex trial, Stevenson et al. (2011) provided evidence that early confirmation has no significant impact in reducing behavioral problems in DHH at ages 5 to 11 years. In a similar vein, Wake et al. (2016) reported that UNHS had no significant effect on DHH children’s HRQOL. On average, parent-reported HRQOL was at −0.4 SD of normative mean values, regardless of whether children had access to UNHS or risk-factor screening programs.

The literature on population-based cohorts revealed that language scores of early-confirmed children or children who had access to UNHS remain on average more than one SD below their hearing peers (Kennedy et al., 2006; Stevenson et al., 2011; Wake et al., 2016); and there is considerable individual variability in performance scores (Pimperton & Kennedy, 2012; Wake et al., 2016; Pimperton et al., 2017). Achieving a better understanding of mechanisms that underlie observed variations in outcomes is essential to devising effective intervention strategies.

This study examined whether the same factors impact language abilities at school age as in preschool age in a population-based cohort. Factors that have proven associations with language outcomes include child (i.e., sex, additional disabilities, auditory neuropathy, hearing level), intervention (i.e., age at fitting HAs or CIs, hearing aid prescription, communication mode in early intervention), and family (i.e., socioeconomic status, maternal education) characteristics. These affected language outcomes of the cohort at age 3 and 5 years (Ching, Dillon, et al., 2013; Leigh et al., 2015; Ching et al., 2017; Cupples et al., 2018). Ching, Dillon, et al (2013) showed that on average, the global language score (a weighted sum of expressive and receptive language, receptive vocabulary, speech production, and auditory functional scores) of children at age 3 years was higher for the female gender, those who had less severe hearing loss, who did not have additional disabilities, whose mothers had higher educational level, and whose CIs (for children with CIs) were activated earlier. At age 5 years, global language scores (aggregated from 20 measures including receptive and expressive language, vocabulary, oral reading and passage comprehension, speech production, auditory functional scores, social skills, and HRQOL) increased with decrease in age at HA fitting and age at CI activation (Ching et al., 2017). Cupples et al. (2018) examined factors influencing specific aspects in spoken language and functional performance of children at age 5 years. The study reported that higher non-verbal IQ was associated with better outcomes in most measures, including receptive and expressive language, receptive vocabulary, speech production, and auditory functional performance. Earlier age at HA fitting and use of oral communication were associated with better outcomes on directly administered language assessments. Less severe hearing loss and higher maternal education level were associated with better language, consonant production and auditory functional outcomes of children using HAs. Earlier age at CI activation and the absence of additional disabilities were associated with better outcomes of children using CIs. The contribution of verbal memory was not assessed.

The current study also drew on findings from the same cohort in earlier years to examine factors influencing HRQOL outcomes (Leigh et al., 2015; Wong et al., 2017; Wong et al., 2018). Leigh et al. (2015) reported that psychosocial development at age 3 years was within the range for hearing children, with large individual differences. There was a positive correlation between language ability and social development. The presence of additional disabilities was associated with weaker social skills. Wong et al. (2017; 2018) provided evidence on social and emotional development of the cohort at age 5 years, showing that on average, global psychosocial scores (a composite score of measures of social skills, emotion, and behavior) were at −0.67 SD of normative mean values of typically hearing peers, with high individual variability. Factors that affected psychosocial development included nonverbal IQ, presence of additional disabilities, language abilities and auditory functional performance. Children with HAs who had higher nonverbal IQ scores, better language and higher functional auditory performance scores achieved higher psychosocial functioning scores. Children with CIs who had no additional disabilities and who obtained higher functional auditory performance scores demonstrated better psychosocial functioning (Wong et al. 2017). Individual research reports on DHH children have also indicated that better language (Fellinger et al., 2009; Castellanos et al., 2018; Haukedal et al., 2018; Stevenson et al., 2018; Haukedal et al., 2020; Haukedal et al., 2023), higher maternal education (Spencer et al., 2012) and the use of an oral communication mode (Haukedal et al., 2020) were associated with better psychosocial and HRQOL outcomes.

Regarding research that examined the associations among outcomes domains, we note that studies on the effect of language on speech perception in noise have mixed results, with some showing that children with stronger language abilities have better speech perception abilities as measured using speech in noise tests (e.g. Ching, Zhang, Flynn et al., 2018; von Koss Torkildsen et al., 2019), whereas others have not (McCreery et al., 2017). Of direct relevance to this study is evidence on the role of speech perception on language development (Desjardin et al., 2009; Chen et al., 2015; Davidson et al., 2019). Desjardin et al (2009) showed that speech perception in quiet accounted for significant unique variance in mean length of utterance in 18 DHH children, 11 of whom used HAs and 7 used CIs, at ages 2.5 to 6 years. Chen et al (2015) examined the influence of demographic variables on speech perception and vocabulary abilities in a sample of 115 Mandarin-speaking children with CIs, ages 2.5 to 7.1 years, and found that the strongest individual contributor to children’s vocabulary ability was their speech perception. Davidson et al (2019) reported on speech perception and receptive language abilities of a group of 117 children who used CIs, ages 5 to 9 years, showing that after controlling for the effects of nonverbal IQ, age at CI activation, gender, and maternal education, speech perception skills accounted for unique variance in receptive vocabulary.

Previous research has demonstrated that early exposure to sound is vital for the development of cognitive abilities that relate to the representation and processing of sequential information, which in turn is associated with language development (Conway et al., 2009; Conway et al., 2011; Harris et al., 2013; Pisoni et al., 2011). Findings from the same cohort in this study measured at age 5 years are consistent with a link between nonverbal IQ and language (Cupples et al., 2018); and also between nonverbal IQ and speech perception (Ching, Zhang, Flynn et al., 2018). Ching, Zhang, Flynn et al. (2018) found that nonverbal IQ accounted for significant variance in performance of 98 children using HAs and 80 children using CIs. For children using CIs, age at CI activation and maternal education also accounted for variance in speech perception. The contribution of verbal memory was not investigated at that young age. In a longitudinal study on 112 children with CIs, Pisoni et al. (2011) examined the influence of digit span and verbal rehearsal speed, measured at 8–9 years of age, on language outcomes measured at 15–18 years. They found that digit span scores were significantly below that of a normative age-matched sample, and that early digit span and verbal rehearsal speed were significantly positively correlated with language scores measured later, even when controlling for age at CI activation.

There is also evidence of a positive association between language ability and HRQOL (Stevenson et al., 2018; Haukedal et al., 2018; Haukedal et al., 2020). Stevenson et al. (2018) showed that better language and reading comprehension abilities were associated with reduced emotional and behavioral difficulties in 57 DHH children who were assessed at 6–10 years, and again at 13–20 years as part of the Wessex trial. The positive association between language abilities and HRQOL is also shown in two recent studies on HRQOL in DHH children aged between 5 and 13 years who had no additional disabilities. Haukedal et al. (2018) reported that better communication and language abilities in 45 children using HAs were associated with higher parent-reported HRQOL. Similar findings were reported by Haukedal et al. (2020) on 84 children using CIs.

Based on associations among outcomes domains shown in the literature and those found in the current research cohort measured at 3 and 5 years of age, we hypothesized multiple relationships between child and family characteristics on the one hand and outcomes domains on the other; together with hypothesized associations among outcomes domains. Figure 1 shows a schematic representation of the hypothesized relationships.

Figure 1.

Schematic representation of structural relationships hypothesized among variables.

Our aims were to: (1) describe the cognitive, speech perception, language, and HRQOL outcomes in 9-year-old DHH children using HAs or CIs; and (2) examine the system of relationships among child, family, and intervention factors, including age at fitting of HAs or CIs, and outcomes.

MATERIALS AND METHOD

This report is part of a population-based study, the Longitudinal Outcomes of Children with Hearing Impairment (LOCHI) study (Ching, Leigh, et al., 2013). The study commenced during a narrow time window when Australian states were at different stages of implementing UNHS such that children across states had differential access to UNHS, but uniform access to free pediatric audiological services provided by a government-funded organization. National protocols ensured consistency in assessment and amplification from diagnosis, and referrals for CI, medical, and educational intervention. The service network records hearing acuity and treatment in its database, which served as the sampling frame. Families of DHH children born between March 2002 and August 2007 in New South Wales, Victoria, and Queensland who first received hearing intervention before 3 years of age were invited to participate. This study was approved by the institutional Human Research Ethics Committee. Parents/caregivers provided written informed consent.

Participants

Children enrolled in the LOCHI study who had received, by 7 years of age, a device of the same type (i.e., HA or CI) that they were using at the time of assessment.

Procedure:

After a participant turned age 9 years, a researcher contacted their family to arrange for assessments by research speech pathologists at home, in school or at a hearing service center. Children wore their personal CIs or HAs during assessments. Assessors were blinded to a child’s age at diagnosis, age at HA fitting or CI activation. The direct assessments of outcomes required approximately 2 – 3 hours to complete, over one or two sessions with breaks when required. The assessor also went through the items of an HRQOL questionnaire verbally with each child and noted down the ratings provided by the child. The speech perception tests were conducted at hearing centers by research audiologists. Questionnaires were sent to parents for completion.

Measures

Language and cognitive assessments.

Children were assessed using standardized measures with known validity and reliability (see Table 1 for details). Language abilities were assessed using the core language subtests of the Clinical Evaluation of Language Fundamentals – 4^th edition (CELF-4; Semel et al., 2003), including concepts and following directions (CFD), recalling sentences (RS), formulating sentences (FS), word classes – receptive (WCR) and word classes – expressive (WCE). The CELF-4 is a standardized test of spoken English that has been used widely to assess language abilities of DHH children (Tomblin et al., 2015; Wake et al., 2016). It includes verbal tasks that enable children to demonstrate understanding of and ability to produce English language structures. It gives standard scores for each subscale. The Peabody Picture Vocabulary Test – 4^th edition (PPVT-4; Dunn & Dunn, 2007) and the Expressive Vocabulary Test – 2^nd edition (EVT-2; Williams, 2007) were used to measure vocabulary skills. The PPVT-4 is a standardized test of receptive vocabulary, using a four-alternative forced-choice, picture-pointing format in administration. The test has been used widely for assessing vocabulary of DHH children (Tomblin et al., 2015; Wake et al., 2016). The EVT is a standardized test of expressive vocabulary that gives an overall score of expressive vocabulary. Given the importance of written language processing as children get older, measures of reading skills were included. Three subtests from the Woodcock Johnson III^® Diagnostic Reading Battery (WJ III DRB; Woodcock et al., 2004) were used. The WJ III DRB has been used widely for assessing DHH children (Mayer et al., 2021). The word identification (WI) subtest assessed children’s ability to recognize and produce letters and words. The word attack (WA) subtest measured children’s ability to apply phonological decoding skills to read non-words aloud. The passage comprehension (PC) subtest assessed children’s understanding of words, phrases, and/or short passages using word-picture matching and cloze procedures. Good split-half reliability is reported in the test manual for each of the three subtests.

Table 1.

Key measures for the study.

Domains	Administration	Measure	Information about Measure and Scoring
Language	Direct administration	Clinical Evaluation of Language Fundamentals – 4th edition (CELF-4)	Standard scores were derived using published norms.
		Concepts and following directions (CFD)	Recall and follow spoken directions of increasing length and syntactic complexity.
		Recalling sentences (RS)	Recall and repeat sentences of increasing lengths and syntactic complexity.
		Formulating sentences (FS)	Produce syntactically and semantically correct sentences based on a picture and a word spoken by an examiner.
		Word classes - receptive (WCR)	Select two words that go together from a set of four words.
		Word classes - expressive (WCE)	Explain how the two selected words are related.
Receptive vocabulary	Direct administration	Peabody Picture Vocabulary Test, 4th edition (PPVT-4)	Indicate which one of four pictures best showed the meaning of a spoken word. Standard scores were derived using published norms.
Expressive vocabulary	Direct administration	Expressive Vocabulary Test – 2nd edition (EVT-2)	Name an item or action shown in a picture or give a synonym. Standard scores were derived using published norms.
Reading	Direct administration	Woodcock Johnson III Diagnostic Reading Battery (WJ III DRB)	Standard scores were derived using published norms.
		Subtest: Word identification (WI)	Oral reading of words.
		Subtest: Word Attack (WA)	Oral reading of non-words.
		Subtest: Passage Comprehension (PC)	Understanding of words, phrases, and/or short passages using word-picture matching and cloze procedures.
Nonverbal IQ	Direct administration	Wechsler Nonverbal Scale of Ability (WNV)	Two subtests version (matrices and spatial span). Standard full-scale scores were derived using published norms.
Short-term memory	Direct administration	Comprehensive Test of Phonological Processing – 2nd edition (CTOPP-2) Subtest: Memory for digits	Measures forward digit recall. Standard scores were derived using published norms.
Sentence perception	Direct administration	Sentences in collocated babble noise, presented at 0° azimuth from one metre in a sound treated room (BEST-S0N0)	Speech reception threshold for 50% correct repetition of keywords in recorded sentences, expressed in terms of decibels (dB) signal-to-noise ratio (SNR).
		Sentences in spatially separated noise: sentences from 0° and noise from +90° and −90° azimuth (BEST-S0N90)	Speech reception threshold in dB SNR.
Consonant perception	Direct administration	Nonsense syllables in collocated noise presented at 10 dB SNR (VCV-N)	Percent correct consonant score.
HRQOL	Parent proxy-report	Pediatric Quality of Life Inventory – version 4.0 (PedsQL 4.0)	23 items. The total score is the sum of physical, social, emotional, and school functioning scores. Higher scores indicate better HRQOL. Z-scores were derived using published norms.
	Self-report	Pediatric Quality of Life Inventory – version 4.0 (PedsQL 4.0)	23 items. The total score is the sum of physical, social, emotional, and school functioning scores. Higher scores indicate better HRQOL. Z-scores were derived using published norms.

Abbreviations: IQ, Intelligence quotient; HRQOL, health-related quality of life; dB SNR, signal-to-noise ratio in decibel.

Non-verbal IQ was measured using the Wechsler Nonverbal Scale of Ability (WNV; Wechsler & Naglieri, 2006). The WNV is a standardized test of nonverbal cognitive ability. It gives a full-scale IQ score. In addition, we measured verbal short-term memory using the memory for digits subtest (DS) in the Comprehensive Test of Phonological Processing – 2^nd edition (Wagner et al., 2013). The subtest is a standardized test of capacity of the phonological loop. Recorded digits are presented at a rate of two per second, and forward-only recall is measured. It gives an overall score on phonological short-term memory.

All assessments were completed using standard methods. Direct assessments were video-recorded, and randomly selected samples constituting at least 10% of the total number of assessments were subjected to a second, independent scoring. The interobserver reliability was 97%. We used published norms to derive standard scores from raw scores.

Speech perception.

Speech perception was measured using recorded sentence stimuli (Bench & Doyle, 1979; Dawson et al., 2013) and nonsense syllables presented in babble noise in a sound-treated room at hearing centers. For both types of speech material, digital recordings of native Australian English speakers were used as stimuli. The sentence test is an open-set test requiring children to repeat as much of the stimulus sentence as possible. Each list comprised 16 sentences with 50 keywords for scoring. Speech stimuli were presented via a computer-controlled sound card (Babyface RME) and a power amplifier connected to three loudspeakers positioned at 0°, +90°, and −90° azimuth respectively at a distance of 0.75 meter from subject position at ear level. Calibration of each of the loudspeakers was conducted using one-third-octave-band noise centered at 1 kHz with a sound pressure level equivalent to that of the speech stimuli. For the sentence tests, digital recordings of the stimuli were presented from the loudspeaker positioned at 0° azimuth in two listening conditions. In the S0N0 condition, 8-talker babble was presented from the same loudspeaker. In the S0N90 condition, uncorrelated four-talker babble was presented at +90° and −90° azimuth from both sides, constituting an eight-talker babble effectively. The sentences were presented at 65 dB SPL, and the level of the babble noise was adjusted adaptively depending on the number of keywords that were correctly repeated. The noise level was increased if 50% or more of the keywords were repeated correctly. The noise level was initially presented at 10 dB SNR, then adjusted by 4 dB for the first four sentences, followed by 2 dB steps for the remaining sentences in the list. Following each adjustment was a 2 sec period of noise at the new level and a beep cue before the next sentence was presented. The speech receptive threshold for 50% correct (SRT) was calculated as the mean signal-to-noise ratio (SNR) for the remaining sentences in the list. A custom-designed software (SRT4.12 software, Cochlear Limited) was used to control the presentation of stimuli, implement the adaptive procedure, and compute the final SRT for each list in each listening condition. Prior to measurement, the children completed a practice list. Two sentence lists were completed for each condition.

The consonant test comprised of 21 English consonants in a vowel-consonant-vowel context where the vowel is [a], presented in 4-talker babble noise (VCV-N). Each consonant occurred 4 times in a test list. Stimuli in noise were presented from 0° azimuth, using the same software as for the sentence material. The speech stimuli were presented at 65 dB SPL. The child was required to repeat the nonsense syllables heard, and the number of consonants correctly repeated was scored. The research audiologist completed scoring on-line, and a speech pathologist double-scored the test either on-line or based on video or audio-recording of the test. Prior to measurement, a practice list was completed. Two lists were completed at a signal-to-noise ratio of 10 dB.

Quality of life.

Children and parents completed the Pediatric Quality of Life Inventory – version 4.0 Generic Core Scale (PedsQL 4.0; Varni et al., 2001). The PedsQL 4.0 was designed to measure the core dimensions of health as well as role (school) functioning. The scale has good reliability and validity (Varni et al., 2003), and it has been used widely in assessing HRQOL of DHH children (Roland et al., 2016; Wake et al., 2016; Haukedal et al., 2022). The child (8–12 years) self- and proxy-versions were used. The inventory comprises 23 items from four domains: physical health (8 items), emotional functioning (5 items), social functioning (5 items) and school functioning (5 items). Each item is rated on a 5-point Likert scale: 0 = never a problem; 1 = almost never; 2 = sometimes a problem; 3 = often a problem; 4 = almost always a problem. Items were reverse-scored and rescaled to a 0–100 scale, where higher scores indicate better HRQOL. For scale and total scores, the mean was computed as the sum across all items divided by the total number of items answered. A psychosocial health summary score was calculated as the mean score over the items answered across the emotional, social, and school functioning scales. Published norms were used to calculate Z-scores (i.e. number of population standard deviations above or below the mean for typically developing children of the same age).

Parents provided information about child sex, additional disabilities, communication mode in early education (oral or oral plus sign), and the mother’s educational level (school, vocational training, university). An Index of Relative Advantage/Disadvantage score was determined for each participant using the census-based Socio-Economic Indexes for Areas (Australian Bureau of Statistics, 2006), with higher scores reflecting greater advantage. Age at HA fitting or CI activation, hearing level, and presence of auditory neuropathy were retrieved through chart review.

Statistical Analysis

Outcomes of DHH children (Aim 1):

Descriptive statistics (mean and SD, median and interquartile range) were used to report quantitative outcomes.

Relationship among variables (Aim 2):

Structural equation modelling (Beran & Violato, 2010; Kline, 2015) was used to evaluate the inter-relationships (pathways) among hearing level, age at fitting of HA or CI, additional disabilities, auditory neuropathy, primary communication mode, maternal education, socio-economic status and cognitive function, speech perception, language, and HRQOL. Structural equation modelling is a generalization of multiple linear regression. Whereas linear regression models have one dependent variable and one or more explanatory variables, structural equation models (SEMs) can have more than one dependent variable and variables that act as both dependent and explanatory variables. SEMs can also include latent variables, which represent abstract constructs that are not observed directly, but only observed indirectly through other variables called their indicators (Table 3). By including more than one indicator of a latent variable, a model can accommodate the idea that none of the indicators measure the underlying concept perfectly. A structural equation model consists of a set of equations involving multiple parameters and fitting the model to a set of data involves determining the parameter values that make the predicted covariances between the observed variables in the model match the sample covariances as closely as possible. Several methods were used to assess how well the model matched the data. The Chi-squared test evaluates the probability that discrepancies between the model and the data were the result of sampling error, with values <0.05 indicating an unlikelihood that discrepancies are just sampling error. The Comparative Fit Index indicates how well the model fits the data compared to a null model that assumes no correlation between the variables, with values >0.95 indicating a good fit to the data (Hu & Bentler, 1999; Kline, 2015). The Root Mean Squared Error of Approximation indicates the relative magnitude of the discrepancies between the observed and model-predicted co-variances, per degree of freedom, with values <0.06 indicating a good fit to the data (Hu & Bentler, 1999; Cook et al., 2009; Beran & Violato, 2010).

Table 3.

Outcomes and latent variables.

Latent variable	Domain	Measure (Indicator)	Normative mean (SD)	Full cohort Mean (SD)	n	Hearing Aid, Mean (SD)	n	Cochlear Implant, Mean (SD)	n
Language	Receptive and Expressive Language	CELF-4: CFD	10 (3)	6.6 (3.4)	294	6.7 (3.4)	189	6.5 (3.5)	105
		CELF-4: RS	10 (3)	6.7 (3.9)	296	7.0 (3.8)	189	6.3 (4.1)	107
		CELF-4: FS	10 (3)	7.4 (4.1)	293	7.6 (4.1)	189	7.0 (4.3)	104
		CELF-4: WCR	10 (3)	7.8 (3.3)	297	7.7 (3.3)	189	8.1 (3.5)	108
		CELF-4: WCE	10 (3)	8.0 (3.5)	297	7.9 (3.4)	189	8.1 (3.8)	108
	Receptive vocabulary	PPVT-4	100 (15)	87.5 (20.8)	313	88.7 (18.8)	199	85.3 (23.8)	114
	Expressive vocabulary	EVT-2	100 (15)	89.7 (18.7)	314	91.2 (17.4)	202	87.1 (20.8)	112
	Reading	WJ III DRB: WI	100 (15)	97.4 (17.9)	299	98.0 (17.3)	194	96.2 (19.0)	105
		WJ III DRB: WA	100 (15)	98.4 (16.2)	288	98.3 (15.6)	188	98.5 (17.2)	100
		WJ III DRB: PC	100 (15)	89.4 (14.2)	297	89.5 (14.2)	194	89.2 (14.3)	103
IQ	Nonverbal IQ	WNV	100 (15)	99.8 (15.1)	304	100.2 (15.2)	194	99.0 (15.0)	110
STM	Verbal short-term memory	CTOPP-2: DS	10 (3)	9.1 (3.3)	243	9.6 (3.1)	163	8.1 (3.5)	80
Speech perception in noise	Sentence	BEST-S0N0 (dB SNR)	−2.1 (0.9)	2.3 (2.8)	282	1.8 (2.8)	185	3.3 (2.7)	97
		BEST-S0N90 (dB SNR)	−5.7 (2.0)	0.9 (3.6)	282	0.1 (3.5)	185	2.3 (3.3)	97
	Consonant	VCV-N (%)		70.8 (17.5)	249	74.1 (17.3)	168	63.9 (15.9)	81
HRQOL	HRQOL: Parent proxy-report	PedsQL 4.0 Total	0 (1)	−0.3 (1.0)	263	−0.3 (1.0)	174	−0.4 (0.9)	89
		Physical health	0 (1)	0.3 (0.9)	263	0.3 (0.9)	174	0.2 (1.0)	89
		Emotional functioning	0 (1)	−0.4 (1.1)	263	−0.4 (1.1)	174	−0.5 (1.2)	89
		Social functioning	0 (1)	−0.3 (1.1)	263	−0.3 (1.1)	174	−0.3 (1.0)	89
		School functioning	0 (1)	−0.2 (1.0)	263	−0.3 (1.0)	174	−0.1 (1.0)	89
		Psychosocial health	0 (1)	−0.4 (1.1)	263	−0.4 (1.1)	174	−0.4 (1.1)	89
	HRQOL: Child self-report	PedsQL 4.0 Total	0 (1)	−0.8 (1.0)	260	−0.8 (1.1)	177	−0.6 (0.9)	83
		Physical health	0 (1)	−0.3 (1.1)	260	−0.4 (1.1)	177	−0.2 (1.0)	83
		Emotional functioning	0 (1)	−0.6 (1.1)	260	−0.6 (1.1)	177	−0.5 (1.0)	83
		Social functioning	0 (1)	−0.3 (1.1)	260	−0.4 (1.2)	177	−0.2 (1.0)	83
		School functioning	0 (1)	−0.5 (1.1)	260	−0.6 (1.1)	177	−0.2 (0.8)	83
		Psychosocial health	0 (1)	−0.5 (1.1)	260	−0.6 (1.2)	177	−0.4 (1.0)	83

Abbreviations: IQ, Intelligence quotient; STM, short-term memory; HRQOL, health-related quality of life; CELF-4, Clinical Evaluation of Language Fundamentals – 4^th edition; CFD, concepts and following directions in the CELF-4; RS, recalling sentences in the CELF-4; FS, formulating sentences in the CELF-4; WCR, word class – receptive in the CELF-4; WCE, word class – expressive in the CELF-4; PPVT-4, Peabody Picture Vocabulary Test 4^th edition; EVT-2, Expressive Vocabulary Test 2^nd edition; WJ III DRB: Woodcock Johnson III Diagnostic Reading Battery; WI, Word Identification subtest in the WJ III DRB; WA, Word Attack in the WJ III DRB; PC, Passage Comprehension in the WJ III DRB; WNV, Wechsler Non-verbal Test of Ability; CTOPP-2: DS, Digit span subtest in the Comprehensive Test of Phonological Processing 2^nd edition; BEST-S0N0, sentences in collocated noise; dB SNR, decibels signal-to-noise ratio for 50% correct; BEST-S0N90, sentences in spatially separated noise; VCV-N, vowel-consonant-vowel nonsense syllables in noise at 10 dB SNR; PedsQL 4.0, Pediatric Quality of Life Inventory version 4.0.

Several iterations of the model were evaluated. All shared the same basic structure shown in Figure 1. The final model differed from the initial one in that some indicators (additional measures of HRQOL and verbal memory, and a measure of speech perception in quiet) in the initial model were deleted because the correlations observed between them and some other variables were not predicted with sufficient accuracy by the model.

Models were fitted in R, version 3.5.2 (R Core Team, 2016) using the lavaan package, version 0.6–3 (Rosseel, 2012). Models were fitted using ‘full information maximum likelihood’ estimation (Arbuckle et al., 1996). This approach handles missing data without dropping subjects who have incomplete data and without imputing missing values.

RESULTS

A total of 367 DHH children (mean age 9.4 years [SD: 0.3]; 178 girls [48.5%]) provided data for analyses. Of the 472 enrolled at age 5 years (Ching et al., 2017), we followed up with 455 children whose families had not withdrawn from the study by the time their children turned age 9 years. Thirty-two could not be contacted, 14 withdrew, 35 were not available due to family circumstances or health reasons, 5 were not using spoken English for communication, and 2 were deceased. Participants were similar to non-participants with respect to age and severity of hearing loss. Table 2 contains descriptions of participants’ demographic characteristics and hearing devices used.

Table 2.

Characteristics of the study sample.

Characteristics	Full cohort No. (%)	Hearing Aid, No. (%)	Cochlear Implant, No. (%)
Children	367	231	136
Age at test, mean (SD), years	9.4 (0.3)	9.4 (0.3)	9.5 (0.3)
Sex
Male	189 (51.5)	122 (52.8)	67 (49.3)
Female	178 (48.5)	109 (47.2)	69 (50.7)
Ethnicity
Western	255 (69.5)	157 (68.0)	98 (72.1)
Non-Western	107 (29.2)	72 (31.2)	35 (25.7)
Unknown	5 (1.4)	2 (0.9)	3 (2.2)
Birthweight, mean (SD), grams	3042.3 (944.4)	3025.0 (954.0)	3074.9 (929.5)
Degree of hearing loss, BE4FA (dB HL)
Mild (≤40 dB HL)	76 (20.7)	76 (32.9)	-
Moderate (41–60 dB HL)	108 (29.4)	105 (45.5)	3 (2.2)
Severe (61–80 dB HL)	51 (13.9)	39 (16.9)	12 (8.8)
Profound (>80 dB HL)	48 (13.1)	11 (4.8)	37 (27.2)
No measurable hearing	84 (22.9)	-	84 (61.8)
Age at HA fitting, median (IQR), months	5.0 (2.6 – 14.9)	6.0 (2.9 – 16.4)	3.2 (2.2 – 11.8)
Age at CI activation, median (IQR), months	20.7 (12.0 – 34.8)	-	20.7 (12.0 – 34.8)
Hearing device
Unilateral HA	21 (5.7)	21 (9.1)	-
Bilateral HA	210 (57.2)	210 (90.9)	-
Unilateral CI	11 (3.0)	-	11 (8.1)
CI+HA	26 (7.1)	-	26 (19.1)
Bilateral CI	99 (27.0)	-	99 (72.8)
Auditory Neuropathy	37 (10.1)	15 (6.5)	22 (16.2)
Additional disabilities
Present	116 (31.6)	81 (35.1)	35 (25.8)
Missing	24 (6.5)	14 (6.1)	10 (7.4)
Communication mode in early education
Oral only	261 (71.1)	172 (74.5)	89 (65.4)
Combined (oral and sign)	36 (9.8)	23 (10.0)	13 (9.6)
Missing	70 (19.1)	36 (15.6)	34 (25.0)
Changed modes, or never attended	71 (19.3)	42 (18.2)	29 (21.3)
Mother’s education, highest level attained
High school or below	113 (30.8)	69 (29.9)	44 (32.4)
Certificate, diploma	102 (27.8)	64 (27.7)	38 (27.9)
University degree	144 (39.2)	94 (40.7)	50 (36.8)
Missing	8 (2.2)	4 (1.7)	4 (2.9)
SES, median (IQR)	8.0 (5.0 – 9.0)	8.0 (5.0 – 9.0)	7.0 (4.0 – 9.0)

Abbreviations: SD, standard deviation; IQR, inter-quartile range; HA, hearing aid; CI, cochlear implant; CI+HA, a cochlear implant in one ear and a hearing aid in the opposite ear; BE4FA, average of pure-tone thresholds at four frequencies in the better ear or non-implanted ear; dB HL, decibel Hearing Level; SES, socio-economic status in deciles, higher scores reflect greater advantage.

The present cohort included 116 children diagnosed with additional disabilities, 51 (44.0%) of whom have more than one additional disability. Sixteen children (13.8%) were diagnosed with cerebral palsy, 33 children (28.4%) with vision impairment, 24 (20.7%) with developmental delay, 17 (14.7%) with autism spectrum disorder, 11 (9.5%) with learning disorder, 12 (10.3%) with attention deficit disorder, and 64 (55%) with various medical conditions.

(Aim 1): Outcomes of children

Table 3 summarises outcomes of all children, together with descriptive statistics presented separately for children using HAs or CIs. The count of participant numbers for individual tests is also shown. Assessment data were missing for some participants because of either their inability to cope with the demands of directly administered formal tests, or their unavailability for testing. On other occasions, parents declined direct assessments of their children, or questionnaires were not returned.

On average, performance of the full cohort was within one SD of the normative mean for nonverbal IQ (−0.01 SD), verbal short-term memory (−0.3 SD), receptive word classes (−0.7 SD), expressive word classes (−0.7 SD), formulated sentences (−0.9 SD), receptive vocabulary (−0.8 SD), expressive vocabulary (−0.7 SD), reading of words (−0.2 SD) and nonwords (−0.1 SD), and passage comprehension (−0.7 SD). Performance in recalling sentences and following directions were at −1.1 SD of the normative mean. The HRQOL total scores were, on average, within one SD of the normative mean for parent proxy-reports (−0.3 SD) and child self-reports (−0.8 SD). On average, speech perception in noise was poorer than hearing peers (Ching et al., 2011), at −4.9 SD for co-located speech and noise and −3.3 SD for spatially separated speech and noise.

(Aim 2): Direct and indirect relationships among variables

The fitted structural equation model is shown in Figure 2. Following the usual convention, rectangles represent observed variables and ovals and circles represent latent variables. The unlabeled latent variables (circles) are error components. For example, the circle linked to DS represents influences on the DS score other than cognitive ability, including, but not limited to, random measurement error. Single-headed arrows (often called “paths”) mean that the variable at the tail of the arrow is assumed to affect the variable at the head of the arrow. Double-headed arrows represent covariances that are not explained by other variables in the model.

Figure 2.

The structural equation model of standardized path coefficients between family and child characteristics, nonverbal IQ and verbal short-term memory, speech perception in noise, language and HRQOL outcomes.

Note: In the path diagram, a latent variable is presented in the form of an ellipse, a measured indicator by a rectangle, error variances are small circles, a curved line with double arrows indicate covariance, and a directional path is depicted by a straight line with a single-headed arrow.

Abbreviations: HA, hearing aid; Better 4FAHL, four frequency pure tone average in the better ear in decibel hearing level; Age HA, age at fitting of first hearing aids; Age CI, age at activation of the first cochlear implant; MatEd, maternal education level; TAFE, technical education and training; Uni, University; EImode, communication mode in early intervention; Sign/comb, use of signing in communication or a combination of oral and signs in communication; ANSD, auditory neuropathy spectrum disorder; DSL, Desired Sensation Level prescription for hearing aids, SES, socio-economic status; WNV, nonverbal IQ measured by the Wechsler Nonverbal Test of Ability; CTOPP-DS, digit span test in the Comprehensive Test of Phonological Processing; CELF-CFD, concepts and following directions in the Clinical Evaluation of Language Fundamentals; CELF-RS, recalling sentences; CELF-FS, formulating sentences; CELF-WCR, word class – receptive; CELF-WCE, word class – expressive; EVT, Expressive vocabulary test; PPVT, Peabody Picture Vocabulary Test; WDRB-WI, word identification in the Woodcock Diagnostic Reading Battery; WDRB-PC, passage comprehension; WDRB-WA, word attack; Peds-P, Pediatric Quality of life (parent proxy-report); Peds-C, Pediatric Quality of Life (child self-report); QoL, health-related quality of life; BEST-S0N90, sentences in spatially separated noise; BEST-S0N0, sentences in collocated noise; VCV-N, vowel-consonant-vowel nonsense syllables in noise.

The latent variables speech perception, language and quality of life can be thought as the underlying abilities tested by the VCV-N, BEST-S0N0, BEST-S0N90 for speech perception; CELF-CFD, CELF-RS, CELF-FS, CELF-WCR, CELF-WCE, EVT, PPVT, WDRB-WI, WDRB-PC, WDRB-WA for language; and Peds-P, Peds-C for quality of life respectively (see Table 3). In Figure 2, the nonverbal IQ and digit span memory latent variables each have only one indicator, which means that their indicators’ error variances must be assumed instead of being estimated. We assumed error variances of zero, which was equivalent to simply replacing each of the two latent variables by its indicator.

The model chi-squared test indicated that discrepancies between the model and the data were too large to have occurred by chance (χ² = 497.96, df = 270, p <0.001), but despite this, we consider that the model is a reasonable fit to the data, with the Comparative Fit Index = 0.968 and Root Mean Squared Error of Approximation = 0.049. When the sample size is large (with N=367 in this study), even small differences can be statistically significant. The largest discrepancy between the correlation observed between any two variables and the corresponding correlation predicted by the model was only 0.12, observed in one instance.

Tables 4 and 5 contain estimates of the model’s path strengths together with p-levels and 95% confidence intervals. The main paths of interest are those representing the effect of predictor variables on latent variables (Table 4), and the effect of one latent variable on another (Table 5). Our modelling is consistent with verbal short-term memory having a critical mediating effect on multiple outcomes. Better verbal short-term memory is significantly associated with no additional disabilities, earlier CI activation, use of an oral communication mode, and higher maternal education. In turn, better verbal short-term memory is associated with better speech perception, language and HRQOL.

Table 4.

Path strengths between demographic characteristics and outcomes, with 95% confidence intervals within brackets.

	Latent variables
Characteristics	Nonverbal IQ	Digit Span	Speech perception	Language	QoL
Other disability	−0.38***(−0.48, −0.27)	−0.30***(−0.41, −0.20)	−0.10* (−0.20, 0.00)	−0.03 (−0.11, 0.05)	−0.21** (−0.33, −0.10)
Device = HA	−0.21 (−1.15, 0.72)	−0.75 (−1.75, 0.26)	−0.27 (−0.88, 0.34)	−0.09 (−0.60, 0.41)	0.31 (−0.64, 1.26)
Better 4FAHL	−0.10 (−0.64, 0.44)	−0.40 (−0.95, 0.15)	−0.19 (−0.55, 0.17)	0.04 (−0.24, 0.32)	0.02 (−0.49, 0.53)
Better 4FAHL x Device = HA	−0.04 (−0.25, 0.33)	0.05 (−0.24, 0.33)	−0.25*(−0.44, −0.06)	0.04 (−0.11, 0.20)	0.08 (−0.18, 0.35)
Log Age HA (HA)	0.01 (−0.08, 0.10)	−0.04 (−0.14, 0.06)	−0.02 (−0.11, 0.06)	−0.02 (−0.08, 0.04)	0.05 (−0.04, 0.14)
Log Age HA (HA) x better 4FAHL	−0.02 (−0.12, 0.09)	−0.03 (−0.13, 0.07)	−0.05 (−0.13, 0.04)	0.02 (−0.05, 0.09)	0.01 (−0.08, 0.11)
Log Age CI (CI)	−0.20 (−0.78, 0.38)	−0.67* (−1.31, −0.03)	−0.28 (−0.66, 0.11)	−0.06 (−0.39, 0.26)	0.30 (−0.30, 0.90)
MatED = TAFE	0.10 (−0.02, 0.22)	−0.18** (0.07, 0.28)	−0.01 (−0.11, 0.09)	0.07 (−0.01, 0.14)	−0.06 (−0.17, 0.05)
MatED = Uni	0.21** (0.08, 0.33)	0.25*** (0.13, 0.37)	0.05 (−0.05, 0.15)	0.10* (0.01, 0.19)	−0.11 (−0.23, 0.01)
EI Mode = Sign/comb	−0.03 (−0.15, 0.09)	−0.14* (−0.27, −0.02)	−0.12*(−0.22, −0.03)	−0.09* (−0.17, −0.01)	−0.08 (−0.18, 0.02)
EI Mode = Changed/never	−0.18** (−0.30, −0.07)	−0.11 (−0.22, 0.00)	−0.07 (−0.16, 0.03)	0.04 (−0.03, 0.12)	−0.07 (−0.18, 0.04)
ANSD	−0.03 (−0.13, 0.06)	−0.07 (−0.16, 0.02)	−0.03 (−0.10, 0.04)	0.05 (−0.01, 0.11)	−0.04 (−0.16, 0.08)
Prescription = DSL (HA)	0.02 (−0.09, 0.12)	0.01 (−0.09, 0.11)	−0.02 (−0.10, 0.06)	−0.05 (−0.11, 0.01)	−0.02 (−0.12, 0.07)
SES	0.05 (−0.06, 0.15)	−0.01 (−0.11, 0.09)	0.02 (−0.07, 0.10)	0.09* (0.01, 0.16)	−0.07 (−0.17, 0.03)

Abbreviations: HA, hearing aid; Better 4FAHL, four frequency pure tone average in the better ear in decibel hearing level; Log Age HA, log-transformed age at fitting of first hearing aids; Log Age HA (HA) x better 4FAHL, interaction between age at HA fitting and hearing level for children with hearing aids; Log Age CI, log-transformed age at activation of first cochlear implant; MatEd, maternal education level; TAFE, technical education and training; Uni, University; EImode, communication mode in early intervention; Sign/comb, use of signing in communication or a combination of oral and signs in communication; ANSD, auditory neuropathy spectrum disorder; DSL, Desired Sensation Level prescription for fitting hearing aid; SES, socio-economic status; IQ, intelligence quotient; QoL, health-related quality of life.

Notes:

^***p <0.001,

^**p < 0.01,

^*p < 0.05.

Table 5.

Path strengths among latent variables, with 95% confidence intervals within brackets.

	Nonverbal IQ	Digit span	Speech Perception	Language	QoL
Nonverbal IQ	-
Digit span	0.35*** (0.23, 0.47)	-
Speech Perception	0.13* (0.01, 0.24)	0.47*** (0.36, 0.58)	-
Language	0.26***(0.16, 0.36)	0.34*** (0.23, 0.45)	0.40*** (0.25, 0.55)	-
QoL	0.01 (−0.12, 0.13)	0.21* (0.04, 0.38)	0.10 (−0.13, 0.33)	0.54** (0.30, 0.77)	-

Abbreviations: IQ, Intelligence Quotient; QoL, health-related quality of life.

Notes:

^***p <0.001,

^**p < 0.01,

^*p < 0.05.

Nonverbal IQ also has a mediating effect, but not as strongly. Higher nonverbal IQ is significantly associated with no additional disabilities, attendance of early educational intervention, and higher maternal education. In turn, higher nonverbal IQ is directly associated with better speech perception and language.

The presence of additional disabilities affects outcomes via multiple mechanisms. It directly and negatively affects HRQOL, and indirectly affects HRQOL via its effect on cognitive abilities, speech perception, and language abilities.

Children who had early oral-only intervention had better speech perception abilities (path strength = 0.12 [95% CI: 0.22, 0.03], p <0.05) than those who used oral and sign in early intervention.

Children using HAs who had less severe hearing loss demonstrated better speech perception abilities (path strength = −0.25 [95% CI: −0.44, −0.06], p <0.05) than those with more severe hearing loss.

We hypothesized that speech perception affects language abilities, and that language abilities affect HRQOL, because these seem the most plausible directions of causality between these variables. Figure 2 and Table 5 show that the path strengths for each of these effects were high (0.40 [95% CI: 0.25, 0.55] and 0.54 [95% CI: 0.30, 0.77]), and each was highly significantly different from zero (p <0.001).

The path strengths between the latent variable of language and its measured indicators are all very high, ranging from 0.85 to 0.95 (Figure 2). This means that whatever affects the latent variable of language appears to have similar effects on all the specific abilities (indicators) that were measured.

Finally, higher SES background was weakly, but significantly, associated with better language (path strength = 0.09 [95% CI: 0.01, 0.16], p <0.05).

DISCUSSION

Outcomes of children

The first aim of this study was to describe the outcomes achieved by a population-based cohort of DHH children at age 9 years. Children’s performance was measured prospectively on a comprehensive set of assessments targeting cognitive abilities, speech perception, language and HRQOL. On average, children scored within one SD of the normative mean on verbal short-term memory. This result compares favorably with previous reports that showed deficits in children who use CIs or HAs (Pisoni et al., 2011; von Koss Torkildsen et al., 2019). The mean scores on receptive and expressive word classes and vocabulary, were within one SD of the normative mean, but children exhibited weaknesses in recalling sentences and following verbal directions. This relative advantage of vocabulary skills is consistent with findings in previous literature (Moeller, 2000; Geers et al., 2009; Geers & Sedey, 2011; Kronenberger & Pisoni, 2019). The overall scores on language abilities are broadly consistent with those reported on the cohort at earlier ages, noting that those children comprised 82.8 % of the present cohort (Ching et al., 2017; Cupples et al., 2018). The mean scores on oral reading of words and non-words and passage comprehension are similar to those of Mayer and colleagues (2021) in their evaluation of a sample of school-aged DHH children using the same measurement tool. Our results compare favorably with those reported by Kennedy et al. (2006), which showed mean receptive language composite scores of nearly 2 SD below population norms for a sample of 6- to 10-year-old children whose hearing loss was confirmed by age 9 months; those reported by Pimperton et al. (2017) for DHH teenagers at age 16 years; and those reported by Wake et al. (2016) for 5-year-old children.

The advantage shown by the present cohort over those in previous studies is likely related to their earlier ages at intervention, devices used, measures used, and other unknown factors. As shown in Table 2, participants in this study first received their HAs at a median age of 5.0 months. Hearing services and technology provided to children were consistent across all service centers, with multi-channel non-linear hearing aids adjusted according to hearing aid prescriptions and verified using real-ear measures according to national amplification protocols (Ching, Zhang, Johnson et al., 2018). Of the 136 children using CIs, 39 children (28.7%) first received their CIs at or below 12 months of age. The median age at first CI activation was 20.7 months. All children were provided with state-of-the-art technology with processor upgrades at no cost to families. Unlike previous studies that examined the effect of confirmation of hearing loss by age 9 months or access to UNHS on child outcomes (Kennedy et al., 2006; Wake et al., 2016) or studies that examined the effect of early implantation for children who first received CIs by ages 3 or 5 years (e.g. Geers & Sedey, 2011; Niparko et al., 2010), the present study cohort allows the effect of early fitting of HAs or activation of CIs in early life on later outcomes at school age to be estimated reliably. Also, it is noteworthy that unlike previous studies on outcomes that generally excluded children with additional disabilities (e.g. Geers & Sedey., 2011; Haukedal et al., 2022), the present population-based cohort necessarily included children with additional disabilities who were assessed using the same measures, thereby allowing an investigation of the effect of comorbidities on outcomes of DHH children who use HAs or CIs. Despite the aforementioned differences between the present cohort and those reported in previous studies, the present cohort is comparable to those reported in previous population studies in terms of the distribution of severity of hearing loss.

Speech perception abilities were, on average, much below those of typically hearing peers, consistent with results of DHH children in previous studies (Davidson et al., 2019; von Koss Torkildsen et al., 2019), including those of the present cohort at 5 years of age (Ching, Zhang, Flynn et al., 2018). The magnitude of deficit is similar to results described by von Koss Torkildsen and colleagues (2019), whose participant sample was similar to the current cohort in age and hearing devices used. The mean deficit observed in the present cohort at age 9 years is reduced relative to that measured at age 5 years – an improvement possibly attributable in part to maturation of central auditory processes, development of listening strategies, and changes in speech and language abilities with age (Lutfi, Kistler, Oh, Wightman, & Callahan, 2003; Blamey et al., 2001; Byrne, 1983; Davidson et al., 2019). Even with state-of-the-art HAs and CIs fitted at an early age, the present cohort did not approximate the speech perception in noise abilities of peers with typical hearing at 9 years of age. This highlights the need for hearing assistance technology and improvements in signal processing strategies in hearing devices to better support learning in noisy environments, including classrooms.

We found that the parent-proxy and self-ratings of HRQOL of children in this cohort were, on average, within the range of typically hearing peers. The overall scores for parent reports are broadly consistent with those reported by Wake et al. (2016) for DHH children detected via UNHS or risk-factor screening. On average, the total and subscale scores were within one SD of the normative mean for our cohort at age 9 years. This result compares favorably with previous research that reported deficits in social and school functioning in children ages between 6 and 18 years (Roland et al., 2016). Our results also compare favorably with recent studies that showed mean scores of 0.6 to 2.2 SD below normative means for 5- to 12-year-old DHH children with no additional disabilities who use HAs or CIs (Haukedal et al., 2022; Haukedal et al., 2023). We observed general agreement between parent-proxy and self-rating in this study, consistent with previous reports that showed no significant difference between ratings by proxies and the self-reported HRQOL in 5–13 year-old children who use CIs (Haukedal et al., 2020). In general, parent-proxy ratings were slightly better than children’s self-ratings on all subscales in the present cohort. It is noteworthy that children rated their physical health to be lower (Z-score: −0.3) than that rated by parents (Z-score: 0.3), in contrast to findings of Haukedal et al. (2020) that showed a difference in the opposite direction. This may be partly explained by the differences in age of the cohorts.

On average, the present results on HRQOL measured using PedsQL 4.0 compare favorably to those reported in the literature, suggesting potential benefits of early intervention for this contemporary cohort of DHH children. The results measured using PedsQL 4.0 allow a direct comparison of HRQOL between DHH children and their typically hearing peers, and with findings in previous reports. However, hearing-specific effects on quality of life may be underestimated as the tool was not specifically developed for those with hearing or communication difficulties (Ronner et al., 2020).

Direct and indirect relationships among variables

Our second aim was to investigate the relationships among a comprehensive set of outcomes and a range of factors by using structural equation modelling. The analysis takes advantage of the fact that the same measures were applied to both children using HAs and children with CIs, and the relationships among these and a range of child, family, and intervention-related variables were examined. Some intervention variables, specifically, age at CI activation, were necessarily different. We found that verbal short-term memory has a mediating effect on speech perception, language and HRQOL. Earlier CI activation was significantly associated with better verbal short-term memory (path strength = −0.67, 95% CI: −1.31, −0.03, p = 0.044). This finding is consistent with the connectome model of congenital deafness (Kral et al., 2016) which postulates that auditory deprivation has a direct impact on neurocognition that has a cascading effect on developmental outcomes, and sensory restoration by CIs offers the potential for improvement in outcomes. It is plausible that delayed access to understandable speech could affect verbal memory, but we are cautious about saying this is definitely so. The simple observed correlation between age at CI activation and verbal short-term memory was only −0.20, and in an alternative version of the model that included only children with CIs, the path strength was weaker (−0.13) and not significantly different from zero. Nonetheless, our current results are consistent with previous studies that demonstrated a connection between verbal memory and concurrent language processing (Kronenberger et al., 2018), longitudinal associations between verbal working memory, age of cochlear implantation, and language skills in preschool /early school years (Niparko et al., 2010; Kronenberger et al., 2020; Jamsek et al., 2022), and longitudinal associations between school-age verbal memory and adolescent language functioning (Pisoni et al., 2011). The potential role of earlier CI activation in supporting development of verbal short-term memory that has a critical positive effect on language and HRQOL reinforces the importance of streamlining management pathways so that DHH children who need CIs receive them in a timely manner. Without early detection via UNHS and early hearing intervention with HA fitting, early implantation would not have been possible.

The path strengths between age at HA fitting and the latent variables on outcomes are weak (Table 4). In principle, we would expect that age of HA fitting would have the greatest impact for those with more severe hearing loss. This was found when the effect of age at intervention on language outcomes of the present cohort was evaluated at age 5 years (Ching et al., 2017). For all children, there was little effect of age at first HA fitting on outcomes at 9 years. As most children with hearing loss greater than 80 dB HL had CIs when they were assessed, children using HAs generally had mild or moderate hearing loss at the time of assessment. It is possible that the children received sufficient auditory stimulation before amplification for cortical development to occur, such that children who received HAs later could make as good use of the auditory signals after amplification as children who received HAs earlier. For children using CIs, there was also little effect of age at HA fitting on outcomes. Perhaps sufficient numbers of these children had hearing loss so severe that amplification did not provide worthwhile auditory stimulation. Cortical development in response to auditory input would then not have commenced until the children received their CIs, which is consistent with the stronger effect of age at CI activation.

The fitted model is consistent with verbal short-term memory being affected by early intervention, and it, in turn, affecting multiple outcomes. This suggests that supporting verbal memory development may contribute to improving language learning. Strategies for language intervention may include additional approaches to enhancing verbal memory and sequential processing skills (Bedoin et al., 2018; Doosti et al., 2018; Torppa & Huotilainen, 2019). It may also be beneficial to tailor language intervention strategies to meet the needs of children with different verbal memory profiles (Cowan, 2014; Gray et al., 2019). There is a paucity of studies that evaluate language intervention approaches for DHH children to identify optimal strategies to meet individual needs.

Based on the paths that are significant in the model, maternal education directly and positively affects language outcomes, and indirectly via its influence on nonverbal IQ and verbal memory. As education increases access to human, cultural, and social capital that shape parenting practices (Harding et al., 2015), children of mothers with higher levels of education likely benefit from engagement in cognitively stimulating practices and enriched language environments (Jackson et al., 2017; Dwyer et al., 2019; Gonzalez et al., 2020). Although healthcare and intervention professionals cannot directly change maternal education, awareness of parental education level may inform the nature of intervention required by individual families. Provision of information and feedback about parenting practices has been shown to improve the quantity and quality of language use with children from low socio-economic backgrounds (Suskind et al., 2016).

Strengths of the study

This is the first modelling that examines the relationships among age at HA fitting or CI activation, child and family factors, and cognitive, speech perception, language, and HRQOL outcomes in a population-based cohort of school-aged DHH children. The directional pathways are consistent with early CI activation for facilitating cognitive development that underpins language and HRQOL in middle childhood. We measured performance in a range of indicators using validated standardized measures that relate to latent variables so that findings are not restricted to individual aspects of language or HRQOL but to the respective constructs of interest. In our model, all indicators loaded significantly on their latent variables.

Limitations

As the present cohort uses spoken language for primary communication, findings should not be generalized to children who use alternative modes. The relationships observed in middle childhood are assumed to reflect the long-term impact of early detection and hearing intervention on children’s outcomes. This will need to be confirmed using longitudinal data.

The current report describes a range of outcomes measured using standardized tools in a population-based cohort of DHH children including those with additional disabilities, thereby enabling quantification of the effect of additional disabilities on outcomes (Table 4). Although missing data mean that our conclusions are based on available results from about 70–85% of the cohort across measures, the findings allow for generalizability, comparisons across populations, and for documenting developmental changes in the cohort over time.

As this is an observational study, albeit longitudinal, we cannot be dogmatic that the significant associations found are indeed causative relationships. We formulated the model based on current knowledge about potential causative relationships, and the model is reasonably consistent with observed data. It is possible that some other model involving attributes that we did not measure might also be consistent with the data, without involving the causative paths we identified, or involving paths other than the ones that we identified as having strengths insignificantly different from zero. It is also possible that the direction of causality might be different: that later implantation adversely affects language, which impacts cognitive ability, or that both directions apply simultaneously.

Future directions and implications

This study has three major implications. Firstly, fully capitalizing on early detection will require optimizing pathways to early CI referral and activation to support cognitive development. This applies regardless of the direction of causality between cognitive ability and language. Secondly, research is needed to advance individualized intervention after fitting of HAs or CIs to reduce measurable socioeconomic (maternal education) disparities in language and cognitive outcomes. Thirdly, long-term follow-up of the existing cohort could confirm the influence of early hearing on developing cognitive abilities that support language, speech perception, and HRQOL. To this end, we are following up with the existing cohort on outcomes at 16 years of age.

CONCLUSION

This study found evidence consistent with early hearing intervention having a positive effect on the development of speech perception and language via its effect on verbal short-term memory. In addition, maternal education directly and positively affects language outcomes, and indirectly via its effects on nonverbal IQ and verbal short-term memory. Better language is directly associated with better HRQOL. The likely importance of early hearing intervention for cognitive development lends support to early detection and hearing intervention for childhood hearing loss, including streamlining clinical pathways for early CI activation. The critical mediating role of verbal short-term memory for multiple outcomes suggests that strategies for intervention in language and communication development may benefit from tailoring programs to meet the needs of individuals with different memory profiles for optimizing outcomes.