Objective In a national study of Australian children aged 11–12 years old, we examined the (1) prevalence and characteristics of hearing loss, (2) its demographic risk factors and (3) evidence for secular increases since 1990.
Methods This is a cross-sectional CheckPoint wave within the Longitudinal Study of Australian Children. 1485 children (49.8% retention; 49.7% boys) underwent air-conduction audiometry. Aim 1: hearing loss (≥16 decibels hearing level (dB HL)) was defined in four ways to enable prior/future comparisons: high Fletcher Index (mean of 1, 2 and 4 kHz; primary outcome relevant to speech perception), four-frequency (1, 2, 4 and 8 kHz), lower frequency (1 and 2 kHz) and higher frequency (4 and 8 kHz); aim 2: logistic regression of hearing loss by age, gender and disadvantage index; and aim 3: P for trend examining CheckPoint and reported prevalence in studies arranged by date since 1990.
Results For high Fletcher Index, the prevalence of bilateral and unilateral hearing loss ≥16 dB HL was 9.3% and 13.3%, respectively. Slight losses (16–25 dB HL) were more prevalent than mild or greater (≥26 dB HL) losses (bilateral 8.5% vs 0.8%; unilateral 12.5% vs 0.9%), and lower frequency more prevalent than higher frequency losses (bilateral 11.0% vs 6.9%; unilateral 15.4% vs 11.5%). Demographic characteristics did not convincingly predict hearing loss. Prevalence of bilateral/unilateral lower and higher frequency losses ≥16 dB HL has risen since 1990 (all P for trend <0.001).
Conclusions and relevance Childhood hearing loss is prevalent and has risen since 1990. Future research should investigate the causes, course and impact of these changes.
- hearing loss
- risk factors
- secular trend
Statistics from Altmetric.com
What is already known on this topic?
The population prevalence of hearing loss in children is relevant both to their current functioning and the future burden of disabling presbycusis in older adults.
While slight/mild losses may have fewer obvious impact, detecting hearing loss at an early stage could help to prevent progression, for example through environmental modification.
What this study adds?
9.3% and 13.3% of Australian children aged 11–12 years old had bilateral and unilateral hearing loss ≥16 decibels hearing level (dB HL), respectively, across the main speech frequencies.
Most hearing loss was slight (16–25 dB HL) in degree.
While demographic characteristics did not convincingly predict hearing loss, prevalence for bilateral/unilateral lower frequency and higher frequency losses ≥16 dB HL has risen since 1990.
Suboptimal hearing in school children is important. In general, the greater its degree, the greater its immediate impact.1 While slight and mild losses may have few observable impact, detecting these losses in childhood could help to prevent progression of the loss and reduce any long-term impact. They may also represent the first stages of age-related hearing loss (presbycusis),2 with its profound impact in older adults.3–5 If so, rising rates of slight and mild losses in childhood could pose major additional burden in societies with ageing populations.
Prevalence of more severe losses is falling,6 but it is difficult to draw definite conclusions about secular trends in slight and mild losses, due to widely varying definitions, availability of information on conductive losses, test frequencies and participant characteristics. Systematic searches7 of Medline and Embase reveal some evidence to support a recent rise (online supplementary table e1). In terms of bilateral hearing loss ≥16 decibels hearing level (dB HL), prevalence in the US National Health and Nutrition Examination Survey (NHANES) increased significantly from 3.8% in 1988–1994 to 5.5% in 2005–2006 in children aged 12–19 years old,8 while a Canadian study on children aged 10–17 years old reported prevalence of 8.8% in 2009–2010.9 Unilateral losses are more common than bilateral losses in all studies: 11.1% and 14.0% in NHANES in 1988–1994 and 2005–2006, respectively,8 and 13.6% in Canada in 2009–2010.9
Supplementary file 1
Understanding the current prevalence and secular trend of hearing loss is necessary as some postulated risk factors are changing over time. First, daily noise exposure may be increasing. In NHANES, 34.8% of children aged 12–19 years old were exposed to music through headphones and earphones in 2005–2006 compared with 19.8% in 1998–1994,10 and reported personal stereo usage has been associated with a 70% increased risk of slight-mild sensorineural hearing loss.11 Second, factors related to the more general rise in non-communicable diseases could also contribute, including obesity,12 cardiovascular disease,13 diabetes,14 hypertension and dyslipidaemia.15 Consistently increasing levels of inflammation are a putative risk factor for presbycusis.16 Childhood obesity, which has risen dramatically over the last three decades,17 has been cross-sectionally associated with higher hearing thresholds and an almost doubling risk in low-frequency hearing loss.12 Third, social disparities—increasing in many societies—have been associated with noise-induced hearing threshold shift in children of low socioeconomic status.18 While gender balance is stable, sex should be considered because of its possible interactions with other risk factors.19
The Longitudinal Study of Australian Children’s (LSAC)20 recent cross-sectional CheckPoint biophysical module provides an opportunity to study current epidemiology in hearing loss at age 11–12 years.21 Here, we (1) describe the current prevalence and characteristics of hearing loss; (2) quantify its demographic risk factors; and (3) examine secular trends, drawing on international studies published in 1996–2015.
Study design and participants
The Child Health CheckPoint is a cross-sectional population-derived study nested within the national LSAC. LSAC recruited two nationally representative cohorts in 2004, of which the B cohort (5107 infants) is relevant to this paper. In a two-stage sampling design, 10% of all Australian postcodes were randomly selected, stratified by state and urban/rural domicile; in-age children were then randomly selected from the Australian Medicare database and followed biennially. The B cohort response rate was 57.2% (2004), of whom 75% were retained to wave 6 (2014).22
LSAC interviewers obtained consent at the wave 6 interview (10–11 years) to pass contact details to CheckPoint, which was open to all retained B cohort children aged 11–12 years between LSAC’s wave 6 (2014) and wave 7 (2016). Parents provided written consent and children verbal assent.21
From December 2014, CheckPoint contacted each family to ascertain interest and book a single appointment for the child, as the CheckPoint assessment centre visited each Australian state sequentially between February 2015 and February 2016. Each child underwent assessments of multiple body systems; we report on the ‘Listen Up’ station, which was offered in the first 30–45 min of either a 3.5-hour appointment at the main assessment centre (17 stations) or a 2.5-hour appointment at the ‘mini-centre’ (15 stations) in smaller regional cities.
Trained examiners conducted air-conduction pure-tone audiometry using an Oscilla USB-330 (V.3.3.4) computer-based audiometer with Oscilla headphones, and a standardised modified Hughson-Westlake audiometric technique. Testing of the first frequency began at 30 dB HL; if within normal limits, testing for other frequencies began at 20 dB HL. Participants were asked to remove hearing aids and/or cochlear implant speech processors if worn; testing of the first frequency began at 60 dB HL and of successive frequencies at 20 dB HL above the adjacent frequencies threshold. If participants’ hearing thresholds at two or more frequencies in at least one ear were >20 dB HL, parents’ written feedback stated that hearing was outside the usual range and they should consider a clinical audiology assessment. For the first 143 participants, only three frequencies were tested (1, 2 and 4 kHz) for each ear across an intensity range of −10 to 120 dB HL. As CheckPoint systems became faster and additional funding was sourced, testing at 8 kHz (n=1342), tympanometry (n=1090) and soundproof booths (n=930) were successively added.
Tympanometry (middle ear function)
The Oscilla TSM500 automatically calculated ear canal volume, middle ear pressure and compliance during a pressure sweep. Tympanograms were classified as types A (normal compliance), B (no or negligible compliance) and C (normal compliance, negative middle ear pressure), with criteria and interpretation detailed in table 1.11
CheckPoint recorded children’s age, sex and Socio-Economic Indexes for Areas disadvantage index.23 This composite neighbourhood index ranks postcodes nationally according to data from the 2011 five-yearly Australian Census. Contributing items include average household education levels, income levels, employment status and disability for that postcode. Higher scores reflect less disadvantage, with a national mean of 1000 and standard deviation (SD) of 100. Distribution of index scores is divided into five national quintiles.
Table 1 details all derived variables with their constructs and rationale. Our primary pure-tone average outcome was the high Fletcher Index (mean hearing threshold across 1, 2 and 4 kHz). As its range of frequencies maps most closely to the range of speech sound frequencies, it is likely to have the greatest functional relevance for oral communication.24 We also calculated the following indices to maximise cross-study comparability: four-frequency average (1, 2, 4 and 8 kHz), lower frequency average (1 and 2 kHz) and higher frequency average (4 and 8 kHz, believed to be most affected by noise exposure). We defined the following severity groupings: normal (−10 to 15 dB HL), slight (16–25 dB HL), mild (26–40 dB HL) and moderate or worse (≥41–60 dB HL) in line with the American Speech-Language-Hearing Association guidelines25 and other prevalence studies.1 8 We reported hearing abilities both by better and worse ear on high Fletcher Index. For all indices, we defined bilateral hearing loss as thresholds ≥16 dB HL in the better ear, and unilateral hearing loss as thresholds ≥16 dB HL in the worse ear but normal hearing (≤15 dB HL) in the better ear.1 8
Statistical analyses were performed in Stata V.14.0. For aim 1, we calculated the prevalence estimates of hearing loss with 95% CIs. To determine whether it was justifiable to combine those with and without likely middle ear pathology for analyses, we used analysis of variance (ANOVA) to compare the mean hearing thresholds between those with type A, B and C tympanograms. ANOVA was conducted for these comparisons despite a slight skew in the distribution of hearing thresholds, supported by application of the central limit theorem for a study of this size (n=1090 participants with tympanometry).26 For aim 2, we used logistic regression to estimate ORs with 95% CIs for hearing loss according to sociodemographic characteristics. For aim 3, we selected population studies8 9 from online supplementary table e1 that performed air-conduction audiometry in children of comparable ages using similar hearing loss definitions. We then plotted four secular trend lines (representing four definitions) for studies reporting data collected since 1990, each summarising published prevalence by midyear of data collection, with CheckPoint providing the final point. We report P value for trend using logistic regression.
Our main analyses included children irrespective of tympanogram type. In sensitivity analyses, we repeated aims 1 and 2 only including children with type A tympanograms in both ears (n=956), and repeated aim 1 for children tested in soundproof booths (n=930).
Figure 1 presents the study flow from wave 6 of LSAC onwards. The analyses included 1485 children. The mean age was 11.4 years with approximately equal numbers of boys and girls. The mean disadvantage index was 1026.2, indicating a slight skew towards less disadvantaged children compared with the general Australian population (mean 1000) (online supplementary table e2 and figure 1).
Audiometry and tympanometry
Online supplementary table e3 shows that the mean hearing thresholds for individual frequencies and all four indices were similar when including all children (n=1485), and the subsamples with audiometry at 8 kHz (n=1342) and with tympanometry (n=1090), with the lowest (best) for the higher frequency average and highest (worst) for the lower frequency average. The mean hearing thresholds for children with type A, B and C tympanogram in better and worse ears differed significantly (all P≤0.001), but these differences were small with high degrees of overlap of the distributions (online supplementary figure e1 and online supplementary table e3).
Prevalence of hearing loss (aim 1)
Table 2 shows the prevalence of bilateral and unilateral hearing loss using four indices. Regarding our primary outcome (high Fletcher Index), the prevalence of bilateral and unilateral hearing loss (≥16 dB HL) was 9.3% and 13.3%, respectively. Most hearing loss was slight (16–25 dB HL) (bilateral 8.5%, unilateral 12.5%). Using four-frequency averages, the prevalence of bilateral and unilateral hearing loss was 7.3% and 14.3%, respectively. Lower frequency losses were more common than higher frequency losses (bilateral 11.0% vs 6.9%; unilateral 15.4% vs 11.5%) (table 2).
In sensitivity analyses, the prevalence of bilateral and unilateral hearing loss in children with type A tympanograms (n=956) (online supplementary figure e2) and children tested in soundproof booths (n=930) (online supplementary table e4) was similar to our overall prevalence estimates.
Sociodemographic risk factors (aim 2)
Girls appeared slightly more likely to have bilateral (OR=1.55, 95% CI 1.09 to 2.22, P=0.02) or unilateral (OR=1.38, 95% CI 1.02 to 1.86, P=0.04) losses using high Fletcher index. Bilateral/unilateral hearing losses were not significantly associated with age or disadvantage index (table 3). In sensitivity analyses, prevalence was similar by age, gender and disadvantage index in children with type A tympanograms (n=956) (online supplementary table e5).
Secular trends since 1990 (aim 3)
The four secular trend lines representing four definitions (bilateral lower frequency, bilateral higher frequency, unilateral lower frequency and unilateral higher frequency) are plotted in figure 2, with CheckPoint providing the final point for each line. For all four definitions, there was evidence of rising prevalence since 1990 for hearing loss ≥16 dB HL (all P for trend <0.001).
To investigate non-linear secular trends, we conducted post-hoc tests by running logistic regression models with categorical time points, time-squared and time-cubed, respectively. The results show no evidence of non-linearity (data available from authors on request).
Nearly 10% of Australian children aged 11–12 years old had bilateral hearing loss ≥16 dB HL across the main speech frequencies, with the majority slight (16–25 dB HL) in degree. Unilateral losses were more prevalent, and lower frequency losses were more common than higher frequency losses. While demographic characteristics did not convincingly predict hearing loss, prevalence estimates have substantively increased since 1990.
Strengths and limitations
Strengths include our population-based sample, standardised measurement using air-conduction audiometry and tympanometry, and classification of hearing loss via four indices allowing a thorough exploration of hearing thresholds. The children were old enough for good compliance with testing protocols27 28 and accurate ascertainment of even slight losses. Sensitivity analyses indicated that, in a best-case scenario (normal middle ear function, minimal external noise), the prevalence was robust. To our knowledge, we are the first to examine secular trends using different definitions, enabling us to make comparisons with ‘like’ studies. This revealed a consistent and concerning upswing in prevalence and will support future comparisons.
There were also limitations. First, the time and cost constraints of a ‘whole child’ assessment, where hearing was just one of multiple health domains measured, precluded additional frequencies (eg, 0.5, 3 and 6 kHz) included in some studies.8 29 While this could have provided a finer grained understanding of hearing profile, it is unlikely to have greatly altered prevalence. The 3 and 6 kHz do not add greatly to conclusions drawn from the adjacent frequencies,30 and we excluded the 0.5 kHz because it is less relevant to spoken speech and is most affected by the residual background noise (this could partly explain the larger upswing in lower frequency losses). Second, bone-conduction audiometry would have more accurately classified sensorineural, conductive and mixed losses, but would have reduced comparability with other population studies, which mostly did not use bone-conduction audiometry.8 9 28 Our tympanometry indicated little influence of middle ear status on mean hearing thresholds at this age. Third, under-representation of disadvantaged families may limit generalisability to the wider socioeconomic context. Finally, we compared secular trends across studies reporting similar but not identical definitions, since this proved impossible.
Interpretation in light of other studies
Compared with the selected comparable studies, we note an upward secular trend in prevalence since 1990. Design differences would not explain away these trends. Exclusion of children with abnormal tympanograms (type B or C) by the Canadian study8 9 would have reduced prevalence, but would not alter the overall trends. Additionally, our prevalence differed slightly regardless of including or excluding children with abnormal tympanograms. Second, as the first 555 children were not assessed in soundproof booths, background noise could have contributed to some false-positive results. However, prevalence when restricted to the 930 children tested in soundproof booths was slightly higher with more frequent identification of slight lower frequency losses (online supplementary table e4). Third, although racial/ethnic make-up differs somewhat between USA, Canada and Australia, this would not explain the changes over time seen within those countries, for example in the NHANES studies.8
Regarding risk factors, it is not surprising that age was not significantly associated with hearing loss due to our narrow range of 11–12 years. Other studies have variably noted differences in hearing loss prevalence by gender9 10 28 and socioeconomic status.8 We think that our few and inconsistent significant associations most likely reflect chance, but recommend further exploration with a wider range of indicators and preferably replication across studies.
It is widely accepted that non-communicable diseases are progressive, with first perturbations often starting early in life.31 32 Presbycusis >25 dB HL in either ear affects around 15%, 30% and 50% of individuals aged 40–49, 50–59 and 60–69 years, respectively.33 If slight childhood hearing losses are a harbinger of presbycusis, then this documented rise is concerning, particularly while the underlying pathology of hair cell death remains irreversible.
Implications flow. First, it is important to confirm and monitor secular trends in children, young adults and older adults. Second, to cater for multiple definitions across studies and to provide evidence to move towards a single ‘best’ outcomes-driven definition, we recommend setting up an international repository of deidentified person-level population hearing data, similar to the accepted standard for genomic studies. Third, longitudinal studies should document hearing trajectories to establish whether these children with slight losses are indeed the adults with disabling presbycusis of the future. We urge national studies including children approaching adolescence to consider implementing audiometric assessment in future waves; ideally, this would collect a wider range of aural parameters such as bone-conduction audiometry and information on tinnitus. Fourth, causal research, especially in slight losses, is needed to identify prevention and treatment strategies. Noise exposure from sustained headphone use shows inconsistent associations especially at younger ages with likely individual differences in susceptibility to noise-induced hearing loss.34 35 Other avenues of exploration include risk factors for non-communicable diseases (eg, inflammation and adiposity as outlined in the Introduction),12–15 36 ear diseases (eg, otitis media), infections (eg, congenital cytomegalovirus)37 and genetics (eg, late-onset genetic losses, polygenic influences and mitochondrial DNA mutations).38 Trials are needed to determine whether reducing the effects of slight and mild hearing losses (eg, school building design, sound field systems, teaching styles) improve functional and learning outcomes.39 All of these factors could be collected in population-based studies. Finally, age-specific current and future prevalence, burden and costs should be modelled to provide policymakers with the information they need to act.
Childhood hearing loss is prevalent and has risen since 1990. Future research should investigate the causes, course and impact of these changes.
This article uses unit record data from Growing Up in Australia, LSAC. The study is conducted in partnership between the Department of Social Services, the Australian Institute of Family Studies and the Australian Bureau of Statistics. The findings and views reported in this paper are solely those of the authors.
Contributors MW conceived the CheckPoint study with the CheckPoint team. MW was the primary student supervisor, along with RAB and VS, and oversaw all aspects of the study and the manuscript preparation. RSL contributed to hearing data collection and, under the guidance of PC, designed the hearing protocols. JW and CMPC conducted data extraction, cleaning and handling. JW performed data analysis and wrote the main paper. MW, PC, FKM and LG advised on statistical issues and interpretation. All authors critically reviewed the manuscript and had final approval of the submitted and published version of this paper. MW and JW had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Funding This work was supported by the National Health and Medical Research Council (NHMRC) of Australia (1041352, 1109355), The Royal Children’s Hospital Foundation (2014-241), the Murdoch Children’s Research Institute, The University of Melbourne, the National Heart Foundation of Australia (100660) and Financial Markets Foundation for Children (2014-055, 2016-310). The funding bodies did not play any role in the study. The following authors were supported by the NHMRC: VS (Early Career Fellowship 1125687), PC (Centre of Research Excellence in Child Language 1023493), RSL (Postgraduate Scholarship 1114567), FKM (Career Development Fellowship 1111160), LG (Early Career Fellowship 1035100) and MW (Senior Research Fellowship 1046518). VS was additionally supported by a Cottrell Research Fellowship from the Royal Australasian College of Physicians; CMPC by a Ter Meulen Grant from the Royal Netherlands Academy of Arts and Sciences; RAB by the HEARing Cooperative Research Centre, established and supported under the Cooperative Research Centres Program, an Australian Government Initiative; and MW by Cure Kids New Zealand.
Competing interests None declared.
Patient consent Obtained.
Ethics approval The Royal Children’s Hospital Human Research Ethics Committee (HREC33225) and The Australian Institute of Family Studies Ethics Committee (AIFS14-26) approved the study.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.