Pure tone auditory thresholds and their association with cognition in the Canadian Longitudinal Study on Aging

Abstract

Age-related hearing loss (ARHL) is associated with cognitive decline and was identified as the strongest modifiable individual risk factor for dementia. However, inconsistencies in the reported strength of this association have been observed. The present study aimed to assess whether differences in the range of sound frequencies used to determine pure tone average (PTA) thresholds could explain the inconsistencies. Data from 13,654 older adults in the Canadian Longitudinal Study on Aging were used to compute four different PTA thresholds: Low frequency, Speech (corresponding to frequencies used in speech), High frequency, and Average (of all 8 tested frequencies). Two cognitive composite scores were computed for memory and executive function. Correlation and partial correlation analyses, controlling for age, sex, education, cardiovascular risk factors, symptoms of depression and use of hearing aids were conducted, followed by stepwise regression to establish the parameter combination that reflects the strongest hearing-cognition association. PTA was found to be negatively correlated with both cognitive scores for the four methods used to determine PTA and the association remained significant after controlling for covariates. The optimal parameter combination was frequencies 0.5 kHz, 1 kHz and 2 kHz for memory and 0.5 kHz and 3 kHz for executive function, controlling for covariates. These findings confirm a negative association between hearing and cognition. However, this association is weak and the discrepancies in the reported strength of the association cannot be fully explained by measurement of hearing thresholds.

Introduction

Age-related hearing loss (ARHL), also known as presbycusis, is the most common health disorder in the elderly after hypertension and arthritis^1–3. An estimated 1.57 billion people worldwide suffer from hearing loss, and it is projected to affect approximately 2.45 billon people worldwide by 2050⁴. Hearing loss is an important public health issue, especially as numerous studies have reported its association with cognitive decline^5–9. This is supported by meta-analyses^7,10 and analysis of large data banks¹¹ such as the Canadian Longitudinal Study on Aging, where a significant correlation between hearing and cognitive performance was found in a sample of 30,097 individuals¹². Additionally, hearing loss has been associated with higher incidence of mild cognitive impairment¹³ and dementia^9,14,15. Most notably, the Lancet Commission identified hearing loss as the main modifiable risk factor for dementia¹⁶.

However, inconsistencies in the reported strength of the association between ARHL and cognitive decline have been observed, which may contribute to significant heterogeneity in study outcomes⁷. One potential source of variability is the choice of auditory parameters used to determine hearing performance. The first issue resides in how pure-tone threshold audiometry¹⁷ is used to assess hearing sensitivity. Typically, a pure tone average (PTA) is computed from frequencies of 0.5, 1, 2, 3, 4, 6, and 8 kHz¹⁸. Frequency combinations are used to determine PTA using two (e.g. 1, 4 kHz¹⁹, three (e.g. 2, 3, 4 kHz²⁰; 1, 2, 4 kHz⁵; 0.5, 1, 2: kHz;²¹, or four (e.g. 2, 3, 4, 6 kHz²²; 0.5, 1, 2, 4 kHz²³; 0.5, 1, 2, 3 kHz²⁴ frequencies, with frequencies of 0.5, 1 and 2 kHz being the most common in clinical settings. Another issue that may contribute to between study variability is the ear that is used to determine PTA⁷. PTA of the better^15,23 or worst²⁵ ear, as well as the average of both ears^26–28, have all been used to assess the link between ARHL and cognitive decline.

Hearing loss is a complex phenomenon with multiple possible aetiologies, including noise-induced hearing loss, Meniere’s disease, and head injury²⁹. Low frequency hearing loss is relatively uncommon and may stem from several causes such as otosclerosis or early-stage Meniere’s disease^29,30. In contrast, high frequency hearing loss is more prevalent, with ARHL being the most common cause of hearing loss in older adults. ARHL usually begins in the high frequencies and progressively affects lower frequencies¹. Recent studies have seemingly favored frequencies associated with speech to assess hearing loss^{2,15,23,25,31} since cognitive decline in people with ARHL could contribute to a decrease in communication ability leading to social isolation and depression, increasing the risk of dementia³². Excluding higher frequencies in the determination of PTA, however, could lead to an overestimation of the hearing-cognition association, since ARHL usually begins in the high frequencies¹. This is supported by meta-analytic data suggesting that stronger hearing-cognition associations are linked with lower frequency loss (< 4 kHz)⁷. With regards to which ear is used to determine PTA, better-ear PTA is the most frequently used, as it may better reflect functioning, but a recent meta-analysis reported stronger associations when auditory function was assessed with both ears⁷.

Differences in the parameters used to assess auditory performance introduces variability that can affect the significance and strength of associations between hearing and cognition. Between-study variability is an important issue that may limit interpretation of results and clinical translation. Moreover, the interpretability of meta-analyses is also limited when methodological heterogeneity is not accounted for^33,34. Effect size variability has implications for future study design, as it may distort power analysis in optimal sample size estimation. This is particularly relevant considering recent studies suggesting that auditory interventions such as hearing aids can reduce cognitive decline and increase cognitive test scores^35,36. A better understanding of how different assessment parameters modulate the association between hearing and cognition is necessary for the design of optimal interventional trials.

The aim of the present study was to determine the degree to which the method used to calculate PTA threshold affects the strength of the association between hearing and cognition in older adults. More specifically, it was designed to ascertain: (1) How different combinations of pure tone frequencies to determine PTA threshold affect the association between hearing and cognition; (2) Whether computing PTA threshold from the better ear, the worst ear, or the average of both ears affects the association between hearing and cognition; and (3) What is the combination of pure tone frequencies that leads to the strongest association between hearing and cognition. Although the etiology of hearing loss may be relevant for understanding shared biological mechanisms with cognitive decline, large-scale datasets lack such information. The present study adopts a functional approach, focusing on hearing thresholds regardless of the underlying cause.

Methods

Sample selection

The database from the Canadian Longitudinal Study on Aging was used in this study, which includes over 50,000 participants aged between 45 and 85 years old³⁷. All participants completed multiple questionnaires, and part of the cohort - the comprehensive cohort (n = 30,097) - underwent in-depth physical, cognitive, and clinical assessments. A sample from the CLSA Baseline Comprehensive Dataset – Version 7.1 was selected (n = 13,652) based on the following criteria: 60 years and over, complete demographic and health data, complete hearing and cognitive assessment data. Participants with a diagnosis of Parkinson’s disease, dementia, and a history of head trauma were excluded due to the potential impact on their cognition and hearing.

The CLSA has worked collaboratively with all associated Research Ethics Boards across Canada to create a coordinated ethics process. Ethical review of the CLSA protocol was conducted by the research ethics board at each research site with the coordination of the McMaster Research Ethics Board (at baseline there were 13 REBs involved). Participation in the CLSA cohort is voluntary and all individuals provided written informed consent. The present study was a secondary analysis; all participant information was de-identified and participant consent was not required. The study was approved by the Ethics Board of Clinical Research of the University of Montreal (reference number: 2024–5827). All procedures were carried out in accordance with relevant ethical guidelines and regulations and were performed in accordance with the declaration of Helsinki. For more information about the CLSA, visit https://www.clsa-elcv.ca.

Auditory assessment

Audiometric assessment was performed using a Tremetrics RA 300 + digital screening audiometer in automatic test mode. The assessment was carried out in a quiet room, and Audiocup headphones (circumaural) were used to further reduce ambient noise. Tests were performed by CLSA research staff that were not audiologists at data collection sites. Pulsed stimuli were delivered in 5-dB increments for frequencies at 0.5, 1, 2, 3, 4, 6, and 8 kHz (three tone bursts over 1 s) to obtain bilateral hearing thresholds. A reliability test was performed for each ear. Hearing aids were removed for all testing procedures. Participants whose hearing devices could not be removed (Lyric or bone anchor hearing aid) were excluded from testing. Further details of the testing procedure can be found in the CLSA protocol (see CLSA Hearing-Audiometer DCS Protocol V3.0, 2014 for more details³⁸. For the present study, four PTA thresholds were calculated as follows: Low frequency (LH) PTA (0.5, 1, 2 kHz), Speech PTA (0.5, 1, 2, 4 kHz), High frequency (HF) PTA (3, 4, 6, 8 kHz), and Average PTA (0.5, 1, 2, 3, 4, 6, 8 kHz). In addition, separate PTA thresholds were calculated for the better ear, the worst ear, and the average of both ears.

Cognitive assessment

A selected number of neuropsychological tests were used to assess participant cognitive performance and were administered at home or at one of the Data Collection sites. The following tests were included in the present study: Rey Auditory Verbal Learning Test (RAVLT), Mental Alternation Test (MAT)³⁹, Stroop color-word interference test^40,41, Controlled Oral Word Association Test (COWAT), and Animal Fluency Test (AFT). Two composite scores were generated from the cognitive tests based on principal component analysis with maximum-likelihood estimation performed and validated in a previous study using the same dataset¹². All scores were first converted into a standardized z-score. Then, two composite scores (memory and executive function) were computed from the average of the sum of z-scores. The memory score was calculated based on RAVLT (immediate and delayed) and the executive function (EF) score was calculated based on MAT, Stroop, COWAT, and AFT. A detailed description of each test and main variables used in composite score calculations can be found in Table 1.

Table 1.

Description of cognitive tests and main variables.

Composite score	Test	Description	Variable
Memory	RAVLT	Participant is asked to recall a list of 15 words after five learning trials	Immediate recall score Delayed recall score
Executive functions	MAT	Participant is asked to say out loud numbers and letters in the correct order, while alternating between them (e.g. 1-A, 2-B, 3-C, etc.) as quickly as possible for 30 s	A score [0–51]: the number of correct alternations
	Stroop	Condition 1, the participant must read a list of words printed in different ink colors as quickly as possible. In condition 2, the participant must name the color of a circle as quickly as possible. In condition 3 (interference condition), the participant must name the ink color of a distinctly colored word (e.g. say blue for the word ‘green’ printed in blue ink) as quickly as possible	Condition 3: time to completion
	COWAT	Participant is asked to name as many words as possible beginning with a given letter (A, S, F) over a 60-second period	Sum of number of words across three letters
	AFT	Participant is asked to name as many animals as possible in 60 s, without repetition	Number of words

RAVLT: Rey Auditory Verbal Learning Test; MAT : Mental Alternation Test; COWAT: Controlled Oral Word Association Test; AFT: Animal Fluency Test.

Covariates

Six variables associated with hearing loss and cognition were included as covariates for all analyses: age, biological sex, educational level, cardiovascular risk, depressive symptoms, and use of a hearing aid. Advancing age has been shown to be strongly associated with hearing loss and cognitive decline⁴². Moreover, some studies suggest that men are significantly more likely to develop ARHL than women^43–45. Education level has also been shown to be associated with cognitive performance⁴⁶. In the present study, participants were asked to report the highest degree, certificate, or diploma they had obtained and a value from 1 to 6 was attributed (1: no post-secondary degree, certificate, or diploma; 6: university degree or certificate above bachelor’s degree). The incidence of cardiovascular risk factors is also associated with worse cognitive health and a higher risk of hearing loss^47,48. Similar findings were reported in a recent study using the CLSA dataset showing that cardiovascular risk factors such as hypertension, diabetes and smoking are associated with hearing loss⁴⁹. The following cardiovascular risk factors were included in the analysis: stroke history, diabetes, hypertension, and tobacco use. To avoid multicollinearity, a cardiovascular disease variable aggregating several risk factors was computed⁵⁰. One point (range 0–4) was assigned for each risk factor. Depression also appears to be associated with both cognitive impairment⁵¹ and hearing loss⁵². In the CLSA, depressive symptoms were measured using the 10-item Center for Epidemiological Studies Short Depression Scale (CES-D 10; range: 0–30). Finally, the use of hearing aids was determined by asking participants to report if they used any hearing aid.

Statistical analysis

All statistical analyses were performed using SPSS v26.0 and R-studio. Participants with missing data were excluded from the analyses. To determine how different PTA combinations affect the association between hearing and cognition, Pearson correlations (Bonferroni-corrected) were computed between each PTA measure and the two cognitive composite scores (memory, EF). Then, partial correlations were calculated with all covariates. To assess the impact of ear choice, the same correlational analyses were performed for the better ear, the worst ear, and the average of both ears. Fisher r-to-z transformation was used to calculate whether the difference between the coefficients in the partial correlation analysis was significant for the PTA average of both ears.

Stepwise linear regression analysis was used to determine the optimal combination of pure tone frequencies reflecting the strongest association between hearing and cognition. The criteria for variable entry and removal were set at a probability of F-to-enter < = 0.05 and a probability of F-to-remove > = 100. All pure tone frequencies for the average of both ears were entered in separate models for memory and executive function scores, controlling for the six covariates. To control for type-I error, the significance level for regression models was set at p < 0.01.

Results

Data description

The final sample consisted of 13,654 participants (mean age = 69.30, SD = 6.82), including 7,029 women (51.48%) and 6,625 men (48.52%). In this sample, 927 participants reported using any kind of hearing aid (6.8%). Independent samples t-tests revealed significantly better hearing in women for the Speech (t = 11.03, p < 0.001), High Frequency (t = 30.82, p < 0.001) and Average PTA thresholds (t = 21.20, p < 0.001) whereas better hearing was found in men for the Low frequency PTA threshold (t = 4.75, p < 0.001). Independent samples t-tests revealed significantly better cognitive performance in women for the memory (t = 12.17, p < 0.001) and executive function (t = 9.45, p < 0.001) composite scores.

A preliminary correlation analysis was performed to match that used by Phillips and collaborators¹² in a previous study of the CLSA. This was done to determine to what extent differences in participant selection criteria modified the association between hearing and cognition. Partial correlations with age as a covariate were performed between cognitive measures (RAVLT delayed recall, Stroop interference, AFT, COWAT, MAT) and PTA threshold (1,2,3,4 kHz). Results were similar between the two studies: all p values were under 0.001 while r values ranged from − 0.053 to -0.141 in the present study compared to -0.059 to -0.105 in Phillips et al.¹².

Correlations and partial correlations

Pure tone average thresholds are presented in Table 2. PTA was found to be significantly correlated with both cognition scores (all p < 0.001, r = -0.237 to -0.287) for the four methods used to measure hearing (Table 3). The associations remained significant after controlling for the covariates (all p < 0.001, r = -0.066 to -0.110) but were significantly smaller (all p < 0.001, Fisher r-to-z transformation). For the memory composite score, Fisher r-to-z transformation revealed significantly smaller partial correlation coefficients for the High Frequency method compared to the Low Frequency (z = -3.5, p < 0.001) and Speech (z = -3.1, p = 0.002) methods. For the executive function composite score, Fisher r-to-z transformation revealed significantly smaller partial correlation coefficients for the High Frequency method compared to the Low Frequency (z = -2.3, p < 0.02) and Speech (z = -2.67, p = 0.01) methods.

Table 2.

Pure tone averages.

	Bilateral	Better ear	Worst ear
Low frequency	19.87 (10.47)	17.01 (9.71)	22.72 (12.20)
Speech	23.77 (11.14)	20.82 (10.53	26.72 (12.66)
High frequency	41.62 (17.43)	37.38 (17.34)	45.97 (18.45)
Average	32.30 (13.22)	29.07 (12.86)	35.52 (14.42)

Table 3.

Results of correlation analysis (Peason r).

Memory score
PTA measures	Better ear		Worst ear		Bilateral
	No covariates	covariates	No covariates	covariates	No covariates	covariates
PTA LF	-0.238	-0.110	-0.217	-0.095	-0.237	-0.108
PTA speech	-0.270	-0.106	-0.244	-0.090	-0.266	-0.103
PTA HF	-0.270	-0.071	-0.250	-0.056	-0.267	-0.066
PTA average	-0.287	-0.095	-0.260	-0.077	-0.281	-0.089

EF score
PTA measures	Better ear		Worst ear		Bilateral
	No covariates	covariates	No covariates	covariates	No covariates	covariates
PTA LF	-0.226	-0.100	-0.214	-0.096	-0.229	-0.104
PTA speech	-0.261	-0.106	-0.242	-0.098	-0.261	-0.108
PTA HF	-0.262	-0.078	-0.246	-0.069	-0.261	-0.076
PTA average	-0.277	-0.097	-0.256	-0.087	-0.274	-0.096

PTA: pure tone average. All correlations are significant at p < 0.001.

Obtaining PTA with the better ear, worst ear, or the average of both ears resulted in significant correlations (Table 2) with the two cognitive scores (all p < 0.001) and the associations remained significant after controlling for the covariates (all p < 0.001). Fisher r-to-z transformation revealed no significant difference between partial correlation coefficients between the three conditions for any of the PTA threshold methods.

Stepwise regression

Stepwise regression analysis was performed for the two cognitive composite scores, controlling for the 6 covariates. For both memory and executive models, all predictors showed acceptable levels of multicollinearity (VIFs ranged from 1.01 to 3.34). Values were normally distributed (Q-Q plot) and respected homoscedasticity (residuals vs. predicted plot).

For memory, the first model included the 0.5 kHz frequency (F = 368.24, p < 0.001; r² = 0.19). Adding the 0.5 kHz to the 6 covariates improved the fit by an r² of 0.007. The second model added the 2 kHz frequency (F = 327.27, p < 0.001; r² = 0.19). Adding the 2 kHz frequency improved the fit by an r² of 0.002. The final model added the 1 kHz frequency (F = 291.64, p < 0.001; r² = 0.19). Adding the 1 Hz frequency improved the fit by an r² < 0.001. Beta coefficients for each variable are presented in Table 4.

Table 4.

Stepwise regression optimal model for each cognitive score.

	Memory		EF
Adjusted R2	0.19		0.19
	Standardized β	Sig (p)	Standardized β	Sig (p)
Age	-0.293	< 0.001	-0.263	< 0.001
Sex	0.153	< 0.001	0.126	< 0.001
Education	0.181	< 0.001	0.238	< 0.001
CVR	-0.049	< 0.001	-0.040	< 0.001
Depression	-0.042	< 0.001	-0.051	< 0.001
Hearing aid	0.026	= 0.005	-0.022	= 0.015
0.5 kHz	-0.087	< 0.001	-0.072	< 0.001
2 kHz	-0.081	< 0.001		< 0.001
1 kHz	0.036	= 0.019		< 0.001
3 kHz			-0.058	< 0.001

For executive function, the first model included the 0.5 kHz frequency (F = 367.27, p < 0.001; r² = 0.19). Adding the 0.5 kHz to the 6 covariates improved the fit by an r² of 0.006. The final model added the 3 kHz frequency (F = 326.07, p < 0.001; r² = 0.19). Adding the 3 Hz frequency improved the fit by an r² < 0.002. Beta coefficients for each variable are presented in Table 3. Figure 1 shows the correlations between the PTA (resulting from the regression) and the two cognitive scores, with and without the 6 covariates.

Fig. 1.

Scatterplot showing the association between hearing threshold for the average of both ears and measures of memory (PTA 0.5 kHz, 1 kHz, 2 kHz) and executive function (PTA 0.5 kHz, 3 kHz). r²: coefficient of determination; r²_cov: coefficient of determination with the six covariates.

Discussion

The present analysis of CLSA data confirms a weak negative association between hearing loss and cognitive performance, regardless of which ear is chosen to assess hearing performance. Similarly, variations in the method used to calculate PTA threshold had limited effect on the hearing-cognition association except for the High Frequency threshold, which was generally associated with weaker correlations. As expected, worse cognitive performance and hearing loss were highly correlated with increasing age, which emerged as the most prominent variable in the regression analysis. Sex, education, cardiovascular risk factors, symptoms of depression, and use of hearing aid also moderated the hearing-cognition association. These data are consistent with previous findings reporting small effect sizes, after controlling for known covariates, in the association between hearing and performance in a variety of cognitive domains^5,7,21.

The present data confirm a previous analysis of the CLSA dataset, despite important differences in the way data were analyzed. In the study by Phillips and collaborators¹², participants ranged from 45 to 85 years of age, PTA threshold was established with frequencies of 1,2,3 and 4 kHz, a multiple imputation method was used for missing data and different covariates were used in regression models¹². Notwithstanding these differences, effect sizes were similar between the two studies, underscoring the robustness of the findings. Significantly, the data show that the association between hearing and cognition does not change whether PTA is established with the better ear, the worst ear, or with the average of both ears. Similar findings were reported in a previous meta-analysis⁵³. However, in a previous analysis of CLSA data, Mick and collaborators reported that calculating PTA from the worst ear, compared to the better ear, better predicted self-reported hearing loss⁵⁴. Although asymmetric hearing abilities may underlie functional differences⁵⁴, the present data clearly establish that calculating PTA from either ear, or binaurally, has no statistically significant impact on strength of the association between hearing and cognition.

With regards to PTA frequencies, the High frequency method was associated with weaker correlations with the composite cognitive scores. This was partially confirmed by the regression analysis, where the strongest correlations between hearing and cognition were found when hearing is measured using lower frequencies. More specifically, hearing and executive function composite score showed the strongest correlation when hearing was assessed at PTT 0.5 kHz and 3 kHz. For memory, the strongest correlation was found when hearing was measured at PTT 0.5 kHz, 1 kHz and 2 kHz. This aligns with meta-analysis data showing that hearing-cognition associations are stronger when low-frequency loss (< 4kH) is present⁷. While the mechanism underlying the hearing-cognition association remains unclear, ARHL is characterized by a gradual bilateral hearing loss starting in the high frequency range¹. The present findings suggest that cognitive deficits are likely more prominent in a later stage of ARHL, when the low frequency range is impacted. It aligns with findings of a meta-analysis where moderate/severe peripheral hearing impairment was found to be associated with higher risk of cognitive impairment⁵³. Interestingly, one study using CLSA data found that older adults are unlikely to self-report sensory loss even though an impairment is detected with behavioral measures. It suggests that earlier mild hearing loss may be overlooked due to compensation ability and psychological factors⁵⁵. In the later stage however, more functional difficulties can be observed such as hindered communication, hearing difficulties in noisy environments, and slower auditory information central processing, all of which may contribute to social isolation and depressive symptoms^18,54,56,57. It should be noted that associations between hearing loss and cognition found in the present study were weak. Indeed, when controlling for covariates, hearing explained at most 1% of the variance. This was also reflected in the regression analysis, where adding significant frequencies to the covariates improved model fit negligibly. As a result, these association may have limited practical implications. However, as suggested in a recent meta-analysis where similarly weak association between hearing and cognition were reported⁷, these effect sizes are similar to other risk factors associated with cognitive impairment and dementia. It should also be noted that the weak associations reported in the present study diverge from some of the earlier work investigating the association between hearing loss and cognition. Indeed, some studies reported moderate effect sizes when PTA was correlated with measures of global cognition⁹ and tasks measuring specific cognitive capacities such as inhibition and cognitive flexibility²³. Inconsistent effect sizes and differences in outcome variability between cognitive domains have also been noted⁷. For example, in cross-sectional studies, the strength of the association between hearing loss and attentional abilities (r: -0.050 to -0.411) is more inconsistent than studies investigating language fluency (r: -0.040 to -0.160). Notably, associations were weaker with later publication date, likely due to better research practice and larger sample sizes⁵⁸. Weak associations in the present study may also be attributed to the inclusion of several covariates (age, sex, CVR, hearing aid, education, depression) that were often overlooked in previous studies⁷.

Considering the present findings, future studies should determine whether hearing loss in specific frequencies is associated with deficits in distinct cognitive domains, as well as assessing the association between hearing loss and cognition in individuals with high frequency or low frequency loss. When feasible, considering the small effect sizes found here and elsewhere^7,12, studies should prioritize large datasets. In turn, a better understanding of the association between hearing and cognition should (i) contribute to a better integration of the auditory elements in the early detection of pathologies; (ii) improve the identification of appropriate study populations and sample size estimation for research protocols; (iii) promote better assessments of the impact of interventions; and iii) guide the development of targeted and relevant interventions.

Finally, limitations to the present work should be acknowledged. The cognitive assessment was limited in terms of scope, where domains such as visuospatial ability, attention, and processing speed were not tested. This may be relevant as regression analysis revealed domain-specific sensitivity to associations with specific frequencies used to determine hearing ability. Furthermore, the memory composite score may not reflect real-world memory abilities, as it relied on a single measure (RAVLT). The assessment of hearing thresholds may not have been conducted in optimal conditions as it was not performed in a sound-treated environment. Interestingly, it has been reported that hearing-cognition associations are generally weaker when hearing is tested in sound-treated rooms⁷. Importantly, only peripheral hearing was measured in the present study, overlooking the contribution of central auditory processing. This is noteworthy given recent evidence suggesting that auditory brainstem responses, a measure of central auditory functioning, are associated with cognitive performance⁵⁹. This suggests that central auditory processing may have an important role in cognitive decline and that additional measures such as speech-in-noise may lead to a better understanding of the association between hearing and cognition. Finally, between-studies differences in outcome are certainly not limited to which frequencies were used to determine PTA. Sample size, participant characteristics such as age, sex and ethnicity, choice of covariates and cognitive tests, to name a few, contribute to the strength of the association.

Conclusion

There is growing interest in understanding the association between hearing and cognition considering recent data suggesting that simple auditory interventions can slow cognitive decline in older individuals^35,36. Although critical in establishing the nature of this association, the way auditory performance is assessed has generally lacked justification. In the present study, it was shown that hearing loss shows a weak but robust association with cognitive functions (memory and EF) in a large population-based sample. Moreover, it was shown that the strength of this association is mostly independent of which combination of frequencies or ear is chosen to measure hearing loss. These findings could facilitate between-studies comparison, provide justification for methodological choices related to hearing assessment and contribute to the development of standardized methods of measuring hearing loss in clinical trials.

Author contributions

Yi Ran Wang: Conceptualization, Methodology, Formal analysis, Writing- Original draft preparation, Visualization. Benoit-Antoine Bacon: Funding acquisition, Writing - Review & Editing; François Champoux: Conceptualization, Methodology, Validation, Supervision, Writing - Review & Editing. Hugo Théoret: Conceptualization, Methodology, Formal analysis, Validation, Supervision, Writing - Review & Editing.

Funding

This work was also supported by the Chaire Fondation Caroline Durand en audition et vieillissement (FC) and funding from Carleton University to BAB.

Data availability

Data are available from the Canadian Longitudinal Study on Aging (www.clsa-elcv.ca) for researchers who meet the criteria for access to de-identified CLSA data. The datasets used for the current study are available from the corresponding author on reasonable request.

Competing interests

The authors declare no competing interests.