Association of Audiometric Age-Related Hearing Loss With Depressive Symptoms Among Hispanic Individuals

Key Points

Question  Does an association exist between objectively measured age-related hearing loss and depression in Hispanic adults?

Findings  In this cross-sectional study of 5328 Hispanic adults 50 years or older from 4 US communities, audiometric hearing loss was associated with clinically relevant depressive symptoms as assessed using the Center for Epidemiologic Studies Depression Scale, 10-item version, adjusting for hearing aid use, age, sex, educational level, study site, geographic background, cardiovascular disease, and antidepressant use. The odds of having clinically significant depressive symptoms increased approximately 45% for every 20-dB increase in hearing loss.

Meaning  Objectively measured hearing loss appears to be independently associated with clinically significant depressive symptoms in older Hispanic individuals.

Importance  Age-related hearing loss is highly prevalent and has recently been associated with numerous morbid conditions of aging. Late-life depression is also prevalent and can be resistant to available treatments. Preliminary studies examining the association between hearing loss and late-life depression have been limited by subjective hearing measures, small sample sizes, and primarily white populations.

Objective  To assess whether a cross-sectional association exists between objective audiometric hearing loss and depressive symptoms in older Hispanic adults.

Design, Setting, and Participants  This cross-sectional study uses 2008-2011 Hispanic Community Health Study/Study of Latinos data collected in Miami, Florida, San Diego, California, Chicago, Illinois, or the Bronx, New York, from 5328 Hispanic adults 50 years or older who had exposure, outcome, and covariate data. Data analyses were conducted from March 2018 to September 2018.

Exposure  Audiometric hearing loss (pure-tone average).

Main Outcomes and Measures  Center for Epidemiologic Studies Depression Scale, 10-item version (CESD-10) score of 10 or higher, which indicates clinically significant depressive symptoms.

Results  The median age (interquartile range) of the 5328 participants was 58 (53-63) years, and 3283 participants (61.6%) were female. The mean (SD) CESD-10 score was 7.7 (6.4). Of the 5328 included participants, 1751 (32.9%) had clinically significant depressive symptoms. The odds of having these symptoms increased 1.44 (95% CI, 1.27-1.63) times for every 20 dB of hearing loss, adjusting for hearing aid use, age, sex, educational level, study site, geographic background, cardiovascular disease, and antidepressant use. Compared with those for individuals with normal hearing (0 dB), the odds of having clinically significant depressive symptoms was 1.81 (95% CI, 1.48-2.22) times as high in individuals with mild hearing loss (median threshold, 32.5 dB), 2.38 (95% CI, 1.77-3.20) times as high in individuals with moderate hearing loss (median threshold, 47.5 dB), and 4.30 (95% CI, 2.61-7.09) times as high in individuals with severe hearing loss (median threshold, 80 dB).

Conclusions and Relevance  Objective hearing loss appears to be associated with clinically significant depressive symptoms in older Hispanic people, with greater hearing loss seemingly associated with greater odds of having depressive symptoms. Given the high prevalence of untreated hearing loss in older adults, hearing loss may be a potentially modifiable risk factor for late-life depression.

Age-related hearing loss (HL) has historically been considered a clinically inconsequential result of aging. At worst a quality of life issue, age-related HL had no strong evidence linking it with more concerning health conditions. However, in the last decade, mounting evidence has suggested that hearing may have broader implications, including for mental^1-4 and neurocognitive health.^5,6 Although causation between HL and neuropsychiatric outcomes has yet to be proven, there is early mechanistic evidence, including accelerated brain volume decrease, in those with HL.⁷ A recent article predicted that preventing or treating HL may result in the single largest percentage rate reduction in dementia prevalence of any known risk factor.⁸ This is based on the very high prevalence of age-related HL, which affects two-thirds of those older than 70 years, or 38 million Americans overall.⁹

Data suggesting a link between HL and depression have been more variable.¹⁰ Some, but not all, existing studies have shown associations between these conditions, but even those studies showing an association rely largely on subjectively determined (ie, self-reported) HL,^11,12 which is a suboptimal measure.¹³ For example, many of these studies evaluated responses having a simple yes or no answer to the question of whether an individual noticed HL. More recent studies implementing audiometric testing have shown inconsistent results^1-4 and have been limited to largely white cohorts.

Elucidating risk factors for late-life depression is important because it is prevalent and disabling in older adults. Late-life depression is also frequently resistant to empirical administration of medications (ie, antidepressants) that are not tailored to a specific cause in a given patient. Because HL is readily measurable, modifiable (via hearing aids or cochlear implants), and highly prevalent in later life, demonstrating a link between HL and late-life depression may provide a therapeutically actionable strategy to alleviate depression in the subgroup of patients with comorbid HL.⁹

In the present study, we examined whether an association exists between audiometrically measured HL and clinically significant depressive symptoms using data from the Hispanic Community Health Study/Study of Latinos (HCHS). We were particularly interested in evaluating whether this association exists in a racial/ethnic minority because few Hispanic people¹⁴ have been evaluated for this association to date.^14,15 Depression prevalence has also been reported to be higher in Hispanic people than in other ethnic/racial groups (eg, 26.1% in Hispanic participants compared with 21.5% in non-Hispanic white participants by 1 estimate).^16-18 Furthermore, depression may be underdiagnosed in Hispanic individuals owing to language, cultural, and health literacy barriers as well as to different (eg, more somatic) presentations.¹⁹ We hypothesized that HL in Hispanic individuals would be associated with higher odds of having depressive symptoms.¹

The HCHS is a multicenter, prospective, community-based cohort study conducted in the United States. The HCHS has a uniquely large sample of Hispanic people with audiometric hearing data (n = 16 415 before exclusions) and is the largest study to date (by a factor of 4; n = 5328 after exclusions vs n = 1332 in the previously largest study¹¹) to examine associations between HL and depression. The HCHS data set includes data from interviews, physical examinations, and specialty tests (eg, neuropsychological testing, electrocardiogram, laboratory blood workup, and actigraph recording results). Audiometry is included for all participants. Currently, only 1 wave of data, collected from 2008 to 2011, are publicly available, enabling cross-sectional analysis. The present study using the HCHS data was approved by the Columbia University Institutional Review Board, New York, New York, which also waived the need for obtaining informed patient consent because these data are publicly available.

The Figure is a flowchart showing participant data screening and exclusions. In total, 16 415 participants were included in the downloaded data set. Of those, 2260 participants were eliminated because their records were empty. In addition, 7980 participants younger than 50 years as well as 212 participants with early-onset HL were excluded to restrict the analysis to age-related HL. Participants with missing audiometry results, Center for Epidemiologic Studies Depression Scale, 10-item version (CESD-10) scores, or covariate data were also excluded. Thus, 5328 participants were included in the final analysis.

Pure-tone audiometry across frequencies 500 Hz to 8000 Hz was performed in dedicated sound-isolated rooms by trained examiners.²⁰ The 4 frequency pure-tone average (PTA) was calculated for each ear from the decibel hearing level threshold at 500, 1000, 2000, and 4000 Hz. The HL was based on the PTA of the better hearing ear and unilateral HL was excluded.

For regression modeling, HL was used as a continuous variable. However, for clinical purposes, HL is often described categorically. For this reason, model-based odds ratios (ORs) are presented for comparing specific categories of HL. The severity of HL was categorized as follows²¹: absent or normal, 0 to 25 dB; mild, 26 to 40 dB; moderate, 41 to 55 dB; moderately severe, 56 to 70 dB; severe 71 to 90 dB; and profound, 91 dB or worse. Because there were few participants with severe or profound HL, those categories were aggregated with the moderately severe HL category.

Depressive symptoms were measured with the CESD-10 instrument. Regression analysis primarily used a binarized CESD-10 score. Clinically significant depressive symptoms were defined using a CESD-10 score cut point of 10 or higher, which has been associated with greater disability and adverse health outcomes in older adults.^22,23 Those scores have also been shown to have good predictive accuracy compared with the standard CESD-20 cut point of 16 or higher.²⁴ Alternative outcomes were used as part of sensitivity analyses, which included defining depressive symptoms with a CESD-10 cut point of 16 or higher²⁵ as well as defining depressive symptoms as use of antidepressants.

In the multivariable models, we adjusted for covariates that might alter or confound the association between HL and depression. This included hearing aid use, demographics (age, sex, educational level in years, study site, and geographic background), and cardiovascular disease. Cardiovascular disease is a potential confounder because it may cause both HL²⁶ and depression.²⁷ A composite cardiovascular disease variable was created to address possible multicollinearity of risk factors, as previously described.²⁸ One point was assigned for each of the following 3 individual cardiovascular disease variables if present: coronary artery disease (any of prior electrocardiogram reports of old or possible old myocardial infarction, history of angina, history of heart attack, angioplasty, or cardiac surgery), measured hypertension, or self-reported stroke. An additional 1 or 2 points were added for laboratory-defined diabetes (1 point for impaired glucose tolerance and 2 points for diabetes). In the final model, we also adjusted for antidepressant use (yes or no by documented prescription of any antidepressant class plus self-declared use in the previous 4 weeks).

The likelihood of HL differs across race/ethnic groups.²⁹ However, data about race/ethnicity were missing in more than a third of study participants. For this reason, we did not consider race/ethnicity in the primary analysis. For participants in whom race/ethnic data were available (Asian, Hawaiian Native or Pacific Islander, black, white, or multiracial), we included race/ethnic group as an additional covariate in part of the sensitivity analysis.

The Shapiro-Wilk test was used to assess normality of the continuous variables. Spearman correlation was used to examine the association between continuous CESD-10 scores and hearing.

Regression modeling was used to assess associations between predictors and depression status. Logistic regression was used because outcomes of interest were binary and because the existing literature examining the association between HL and depression overwhelmingly uses logistic regression,^1-4 enabling comparisons with results from previous studies. A univariable model with HL as the predictor was created first, followed by multivariable analysis.

Because logistic regression analysis does not directly estimate prevalence ratios, and the odds ratio does not approximate the prevalence ratio when the outcome is common (as in our sample), we additionally fit log-linear regression models using the same outcome and predictors as the logistic models described above. Model-based prevalence ratios and their corresponding 95% CIs are reported in an additional analysis.

Hearing loss was included in each model as a continuous predictor. An initial Loess fit (binary depression status vs HL) was used to obtain empirical logits, which were plotted against PTA hearing to reveal an approximately linear relationship. Hence, PTA hearing was assumed to have a linear association with the log odds of depression in all logistic models.

Unless otherwise indicated, all estimates are presented with 95% CIs. Data analysis was performed from March 2018 to September 2018 using R, version 3.4.3 (R Foundation for Statistical Computing) with RStudio, version 1.1.423 (RStudio, Inc). No external packages were used for the main analysis. Log-linear models were performed using the core Stats package (glm function, with Poisson family and log link).

Baseline participant characteristics stratified by HL category are given in Table 1. The median (interquartile range) age of participants was 58 (53-63) years old. Most participants (3283 [61.6%]) were women, a reflection of enrollment patterns in HCHS. The majority of participants (4387 [82.3%]) had no HL, 749 (14.1%) had mild HL, 143 (2.7%) had moderate HL, and 49 (0.9%) had moderately severe or worse HL. The mean (SD) CESD-10 score was 7.7 (6.4). Clinically significant depressive symptoms (CESD-10 score ≥10) were present in 1405 (32.0%) of those without HL, 254 (33.9%) of those with mild HL, 64 (44.8%) of those with moderate HL, and 28 (57.1%) of those with moderately severe or worse HL.

In the univariable model, the odds of clinically significant depressive symptoms (defined as CESD-10 score ≥10) were 1.27 (95% CI, 1.14-1.41) times as high per 20 dB of HL (Table 2).³⁰ An increase in 20-dB HL is approximately equivalent to a 1-category worsening (ie, across categories of normal, mild, moderate, moderately severe, severe, and profound).

Adjusting for hearing aid use, the odds of clinically significant depressive symptoms (defined as a CESD-10 score ≥10) were 1.29 (95% CI, 1.16-1.44) per 20 dB of HL. In addition, adjusting for demographic factors, including age, sex, educational level, study site, and geographic background, increased the OR to 1.47 (95% CI, 1.30-1.66). Adding cardiovascular risk factors to this model had a negligible effect. Finally, adding antidepressant use to the model slightly decreased the clinically significant depressive symptoms OR to 1.44 (95% CI, 1.27-1.63). This latter model is referred to as the fully adjusted model (Table 2). In summary, the odds of clinically significant depressive symptoms increased 1.44 (95% CI, 1.27-1.63) times for every 20-dB increase in HL, adjusting for hearing aid use, numerous demographic factors, cardiovascular disease, and antidepressant use.

The fully adjusted model was then used to calculate the odds of clinically significant depressive symptoms for each category of HL (the median within-category value was chosen) compared with normal hearing (0 dB). The odds of clinically significant depressive symptoms were 1.81 (95% CI, 1.48-2.22) times as high for mild HL (32.5 dB), 2.38 (95% CI, 1.77-3.20) times as high for moderate HL (47.5 dB), 3.12 (95% CI, 2.12-4.62) times as high for moderately severe HL (62.5 dB), and 4.30 (95% CI, 2.61-7.09) times as high for severe HL (80 dB).

We included the interaction between HL and sex in the fully adjusted model to assess whether the association between HL and depression status was modified by sex but found no evidence of an interaction (interaction coefficient OR, 1.00; 95% CI, 0.99-1.02). Similarly, we found no evidence of an interaction between HL and antidepressant use (interaction coefficient OR, 0.99; 95% CI, 0.99-1.01). When stratified by antidepressant use, the odds of clinically significant depressive symptoms were 1.43 (95% CI, 1.25-1.63) times as high per 20 dB of HL in those who were not using antidepressants. By comparison, the odds were 1.59 (95% CI, 1.05-2.48) times as high for those who were using antidepressants.

We examined alternative outcomes as part of a sensitivity analysis. When antidepressant use was included as a surrogate outcome of depression, an association with HL was still observed. In the fully adjusted model with antidepressant use as the outcome (instead of a covariate), the odds of antidepressant use were 1.21 (95% CI, 1.01-1.44) times as high per 20 dB of HL (eTable 1 in the Supplement). When a CESD-10 cut-point score of 16 (instead of 10) was used as an alternative outcome, an even higher OR of 1.49 (95% CI, 1.27-1.75) was found, adjusting for the same covariates as the fully adjusted model (eTable 2 in the Supplement).

We also examined race/ethnicity as a covariate as part of the sensitivity analysis. We found no significant association between HL and race for either the univariable regression or the multivariable regression adjusting for demographic characteristics. Race/ethnicity was omitted from the main models because more than one-third of participants lacked race/ethnic data. In the fully adjusted model restricted to 3511 participants with available race/ethnic data, adding race/ethnicity as a covariate had no effect on the association between HL and depressive symptoms (before and after adding race/ethnicity OR, 1.48; 95% CI, 1.26-1.73).

Few participants (47 [0.9%]) used a hearing aid. Restricting the full model to the 5281 participants without hearing aids had virtually no effect on the association between HL and depressive symptoms (excluding hearing aid participants OR, 1.43; 95% CI, 1.26-1.62 vs with all participants OR, 1.44; 95% CI, 1.27-1.63).

Finally, log-linear regression was performed to directly estimate prevalence ratios. In the fully adjusted model, the prevalence of clinically significant depressive symptoms increased 1.23 (95% CI, 1.12-1.34) times for every 20-dB increase in HL, adjusting for hearing aid use, demographics, cardiovascular disease, and antidepressant use (eTable 3 in the Supplement).

Age-related HL loss has been recognized as a risk factor for many conditions of public health concern, most notably cognitive impairment and dementia.^5,6,31 The association between HL and depression, 2 common conditions of older life, has remained less rigorously defined. We found a strong and robust association between audiometric HL and clinically significant depressive symptoms as defined by a CESD-10 score of 10 or higher.²⁴

In the fully adjusted model, accounting for hearing aid use, age, sex, educational level, study site, geographic background, cardiovascular disease, and antidepressant use, the odds of clinically significant depressive symptoms increased nearly 1.5 times for every 20 dB of HL. This 20-dB change is approximately the difference between categories of HL, for example, mild vs normal, or moderate vs mild. An individual with mild HL (32.5 dB median threshold) had nearly twice the odds of having clinically significant depressive symptoms compared with someone with normal hearing (0-dB threshold). An individual with moderate HL had nearly 2.5 times the odds of having clinically significant depressive symptoms, whereas an individual with severe HL had more than 4 times the odds.

Hearing loss may be associated with depression through the mediating effect of socialization. Hearing loss may lead to social isolation³² and loneliness,^33,34 which, in turn, could result in depression. This pathway may be modifiable because loneliness has been shown to improve with certain treatments for HL.^34,35 Furthermore, in an early randomized clinical trial, HL treatment with hearing aids improved both social function and depressive symptoms.³⁶

Previous studies on the association between HL and depression have largely relied on nonaudiometric subjective HL assessments.^11,12 Although associations were found, subjective HL is not an ideal measure.¹³ Particularly in studies of depression, subjective HL may be exaggerated owing to reverse causation, whereby depressed individuals may have a worse subjective perception of their abilities.

Other studies that have examined objective audiometric HL have found weak¹ or absent^3,4 associations. A recent study by our group in a smaller, non-Hispanic cohort did find an association between audiometric HL and depressive symptoms.¹⁴ The present study provides a more robust statistical analysis, including adjustments for numerous confounders. One study in Chinese individuals also showed associations between audiometric HL and depression. However, sample sizes were smaller and variables were not analyzed as continuous, which limited the analysis.² To our knowledge, the sample size in the present study is the largest to date for studies examining HL and depression, by a factor of 4.¹¹ The present study also extends the earlier findings in Chinese² and largely white¹⁴ individuals to a different ethnic group (Hispanic people). These data suggest that the association between HL and depression may be generalizable across multiple races/ethnicities.

It is important to study the association between HL and depression in the Hispanic population, the fastest growing ethnicity in the United States. Hispanic people have accounted for half of all US population growth since 2000.³⁷ Depression may also be more prevalent in Hispanic people compared with other ethnic/racial groups.^16,17 Indeed, we found a relatively high prevalence of depressive symptoms in our Hispanic cohort. Finally, identifying potentially modifiable risk factors for depression in Hispanic individuals is important, particularly given increased challenges with pharmacotherapy. Latino individuals, for example, are half as likely as white individuals to start antidepressant therapy and are also more likely to discontinue such therapy within the first 30 days of treatment. This observation holds even when adjusting for increasing age.^38,39

Depression is a multifactorial disorder with numerous risk factors. Despite the strong association observed in the present study between HL and depressive symptoms, the correlation coefficient between our measures of HL and depressive symptoms was very modest. This indicates that only a small fraction of the overall variability in the CESD-10 scores is explained by HL. However, any association between HL and depression is important given the high prevalence of HL (80% in those older than 80 years)⁹ and low treatment levels (<20% use hearing aids or cochlear implants). This high prevalence and low treatment level could result in a high proportion of late-life depression being attributable to HL.⁴⁰

The present study has some limitations. Because of the cross-sectional design, the temporal order of the exposure and outcome cannot be determined. Thus, reverse causation, whereby depression causes HL, cannot be ruled out. This scenario could be imagined if participants with depression were exposed to known risk factors for HL, such as unprotected noise exposure. Although this is theoretically possible, noise-induced HL typically causes HL centered at 4000 Hz, which was rarely observed in the present study. It is also plausible that participants with depression may have showed worse outcomes on the hearing test because of lack of effort or attention. Although this cannot be entirely ruled out, the examiners were certified in conducting audiometric protocols, which included probing for reliability and consistency. The test-retest reliability of audiometry also approaches 99%.⁴¹

There were several other unavoidable limitations. We attempted to control for confounders between hearing and depression. However, it is also possible that residual confounding could be present. A randomized controlled trial would be needed to eliminate confounding. The oldest participants were 76 years, which does not fully capture the upper age range affected by age-related HL. However, large clinical studies often aggregate all ages over 80 years into a single group to protect anonymity.⁴² Thus, from an analytic standpoint, the upper age range in the present study was about as high as can be practically used. Finally, while the HCHS is a longitudinal study, prospective data collection has only recently begun and only the first wave is currently available. Thus, the current study is limited to cross-sectional analyses. Future studies should study longitudinal data.

This study also has several strengths. With more than 14 000 participants total, the HCHS is one of the largest data sets that contains objective audiometric data. This increased power facilitated finding associations that appear to be generalizable to the US population. Finally, the results remained robust in sensitivity analyses, such as including a CESD-10 cut-point score of 16 and antidepressant use, and in alternative model techniques, including log-linear regression.

We found an association between audiometric age-related HL and clinically significant depressive symptoms as assessed using the CESD-10 in older Hispanic adults. This association remained when adjusting for potentially confounding covariates. Future studies should examine longitudinal relationships and whether treating HL reduces the risk of late-life depression. This is particularly significant in light of the near-universal prevalence of age-related HL and low levels of treatment.

Accepted for Publication: September 27, 2018.

Corresponding Author: Justin S. Golub, MD, MS, Department of Otolaryngology–Head and Neck Surgery, Vagelos College of Physicians and Surgeons, NewYork-Presbyterian/Columbia University Irving Medical Center, 180 Fort Washington Ave, HP8-865, New York, NY 10032 (justin.golub@columbia.edu).

Published Online: December 6, 2018. doi:10.1001/jamaoto.2018.3270

Author Contributions: Dr Golub had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Golub, Brewster, Brickman, Luchsinger.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Golub, Brewster, Ciarleglio, Luchsinger.

Critical revision of the manuscript for important intellectual content: Golub, Brewster, Brickman, Kim, Luchsinger, Rutherford.

Statistical analysis: Golub, Ciarleglio.

Obtained funding: Golub, Luchsinger, Rutherford.

Administrative, technical, or material support: Golub.

Supervision: Brewster, Luchsinger, Rutherford.

Conflict of Interest Disclosures: Drs Golub and Kim reported receiving travel expenses for education and an industry conference paid for by Cochlear. No other disclosures were reported.

Funding/Support: Drs Golub, Brewster, Brickman, Kim, and Rutherford are funded in part by Collaborative and Multidisciplinary Pilot Research Awards (Phase I and II) from the Columbia University Irving Institute for Clinical and Translational Research. Drs Golub (K23AG057832) and Luchsinger (K24AG045334) are supported by awards from the National Institute on Aging.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.