Associations Between Cardiovascular Risk Factors and Audiometric Hearing: Findings From the Canadian Longitudinal Study on Aging

Abstract

Objectives: 

The objectives of the study were to determine, among a population-based sample of Canadian adults, if risk factors for cardiovascular disease (alone and in combination) were associated with hearing loss. Cross-sectional and longitudinal associations (the latter with about 3 years of follow-up) were examined. Risk factors considered included diabetes, dyslipidemia, hypertension, obesity, and smoking. We also aimed to determine if associations were modified by sex and age group (45 to 54, 55 to 64, 65 to 74, and 75 to 86 years old at baseline).

Design: 

A secondary analysis of data collected for the Canadian Longitudinal Study on Aging was performed. Data were collected in two waves, the first between 2012 and 2015, and the second between 2015 and 2018. Hearing was measured using screening air-conduction pure-tone audiometry. The outcome of interest was defined as the mid-frequency (1000, 2000, 3000, and 4000 Hz) pure-tone average for both ears. Diabetes was defined based on self-reported physician diagnosis, use of diabetes medications, or a hemoglobin A1c level ≥6.5%. Dyslipidemia was determined by blood lipid profile as defined using the Canadian guidelines for the diagnosis and treatment of dyslipidemia (low-density lipoprotein cholesterol ≥3.5 mmol/L or non-high-density lipoprotein cholesterol ≥4.3 mmol/L). Hypertension was determined by self-reported physician diagnosis or an average systolic blood pressure ≥140 mm Hg or an average diastolic blood pressure ≥90 mm Hg. Obesity was defined as a waist-to-height ratio ≥0.6. Smoking history was determined by self-report (current/former/never-smoker). Two composite measures of cardiovascular risk were also constructed: a count of the number of risk factors and a general cardiovascular risk profile (Framingham) score. Independent associations between risk factors for cardiovascular disease and hearing were determined using multivariable regression models. Survey weights were incorporated into the analyses. All results were disaggregated by sex. Effect modification according to age was determined using multiplicative interaction terms between the age group and each of the risk factor variables. A complete case (listwise deletion) approach was performed for the primary analysis. We then repeated the multivariable regression analyses using multiple imputation using chained equations to determine if the different approaches to dealing with missing data qualitatively changed the outcomes.

Results: 

In longitudinal analyses, hypertension and the general cardiovascular risk profile score were associated with greater loss of hearing over the 3-year follow-up period for both sexes. In addition, smoking in males and obesity in females were associated with faster rates of hearing decline. In cross-sectional analyses, smoking, obesity, diabetes, and composite measures were each independently associated with worse hearing for both sexes (although for females, obesity was only associated with hearing loss in the 55 to 64-year-old age group). The results were similar for the complete case and multiple imputation approaches, but more cross-sectional associations were observed using multiple imputation.

Conclusions: 

Diabetes, obesity, hypertension, and smoking were associated with hearing loss. Higher combinations of risk factors increased the risk of hearing loss. Further studies are needed to confirm age and sex differences and whether interventions to address these risk factors could slow the progression of hearing loss in older adults.

INTRODUCTION

In high-income countries, smoking, diabetes, hypertension, dyslipidemia, and obesity (which are referred to as “cardiovascular disease [CVD] risk factors” in this article) account for about 50% of the population attributable risk for CVD in men, and about 60 to 70% of the population attributable risk in women (^{Cheng et al. 2014}). CVD risk factors are associated with microvascular damage (^{Ziegler et al. 2019}) that can impair molecular transport (e.g., of oxygen, ions, glucose, metabolites, or proteins) across capillary walls, and/or cause inflammation and oxidative stress that may damage cochlear structure and function (^{van der Vaart et al. 2004}).

Observational studies have been performed to determine if risk factors for CVD increase the risk of developing age-related hearing loss (ARHL), with inconsistent results. The variability in outcomes may be partly due to differences in the prevalence of modifiable risk factors for hearing loss according to the study cohort (^{Brant et al. 1996; Cruickshanks et al. 1998; Zhan et al. 2011; Dawes et al. 2014; Engdahl et al. 2015; Wang et al. 2017; Roth et al. 2020}), differences in measures (e.g., subjective [^{Curhan et al. 2013}] versus behavioral measures of hearing [^{Cruickshanks et al. 2015}]), or differences in other participant characteristics (e.g., age range). Most studies have been cross-sectional. There are very few (if any) published reports of randomized controlled trials investigating therapies for risk factors for CVD or metabolic dysfunction on hearing. In 2016, a protocol was published for a clinical trial that will investigate if daily Aspirin affects the progression of hearing loss (^{Lowthian et al. 2016}); however, the results are yet to be published.

Few studies have examined if longitudinal associations between CVD risk factors and hearing are modified by sex or age group. Gates et al. (¹⁹⁹³) found that associations between cardiovascular disease events and hearing loss were stronger among females than males, and postulated that the difference was due to higher levels of other risk factors (e.g., noise exposure [^{Biswas et al. 2021}]) in male participants. Linssen et al. found that hypertension and large waist circumference were associated with worse hearing in young adults (age 24 to 42 years) but not older age groups. Large waist circumference was also associated with hearing declines in young adults only. Cardiovascular disease was associated with worse hearing and hearing declines only in middle age adults (age 43 to 62 years). Obesity was associated with faster hearing loss in older adults only (age 63 to 81 years). Analyses were not disaggregated by sex (^{Linssen et al. 2014}).

In contrast, there is a large literature showing sex and age influence associations between CVD risk factors and traditional cardiovascular diseases. CVD risk factors are more closely correlated with myocardial infarction, coronary artery disease, or stroke in women (^{Pekka et al. 1999; Appelman et al. 2015; Vogel et al. 2021}). Most CVD risk factors become less important predictors of myocardial infarction, coronary artery disease, and stroke with older age (^{Lind et al. 2022}). The underlying reasons could include an increased baseline risk of CVD in older individuals without CVD risk factors due to the effects of unmeasured age-related factors, longer survival of individuals with lower levels of risk factors, and differences in risk factor levels across different age groups (e.g., smoking rates are lower in older people) (^{Staessen et al. 1989; Weijenberg et al. 1996; Abbott et al. 2002; Centers for Disease Control and Prevention: CDC 24/7: Saving Lives Protecting People 2022; Lind et al. 2022}). Effect modification according to age may differ between the sexes. The onset (or diagnosis) of most cardiovascular risk factors is later in life for women than men (^{Vogel et al. 2021}), and thus, for a given age, the length of exposure time would be greater for men and associations might be seen earlier in life. Hormonal changes around or after menopause may also alter the effect of some of the risk factors in women (^{Vogel et al. 2021}). If sensorineural hearing loss is a peripheral vascular disease, similar age- and sex-interactions might be seen when assessing associations between CVD risk factors and pure-tone threshold average (PTA), and therefore were examined in this study.

Combinations of risk factors increase the risk of developing CVD, cerebrovascular disease, peripheral arterial disease, and aortic disease relative to individual risk factors alone (^{D’Agostino et al. 2008}). Most previous studies examining associations between risk factors for CVD and hearing loss, however, examined risk factors individually rather than in combination, with some exceptions (e.g., ^{Bae et al. 2020}). Risk factors for CVD co-exist to a greater degree than would be expected by chance alone (^{Wilson et al. 1999,2002; Paynter et al. 2022}), and for the purpose of estimating risk according to combinations of CVD risk factors composite risk scores can be used. Scoring systems such as the general cardiovascular risk function (Framingham score) (^{D’Agostino et al. 2008}) and the SCORE system (^{Conroy et al. 2003}) have been created to estimate the risk of cardiovascular disease outcomes based on an individual’s specific risk factor profile. Clinical and public health efforts to reduce the risk of CVD generally aim to address all relevant risk factors.

Our objectives were to determine if associations exist between PTA and smoking, hypertension, dyslipidemia, diabetes, obesity, and composite measures of cardiovascular risk, both cross-sectionally and over 3 years of follow-up among women and men participating in the Canadian Longitudinal Study on Aging (CLSA). The secondary goal was to determine if the associations were modified by age.

MATERIALS AND METHODS

The Study Sample

The CLSA is a 20-year longitudinal closed cohort study consisting of approximately 50,000 Canadians aged 45 to 85 years at the time of baseline analysis (^{Raina et al. 2009}). It consists of two sub-cohorts, the “comprehensive” and “tracking” cohorts. There are approximately 20,000 individuals in the tracking cohort who respond to health surveys via telephone but were excluded from the present study because they did not undergo audiometry. Our analysis was restricted to the approximately 30,000 members of the comprehensive cohort whose information was obtained through in-person interviews and physical assessments (including audiometry) performed at one of 11 data collection sites across Canada.

Recruitment for the comprehensive cohort began in 2012. The baseline or first wave of data (“T0”) was collected between 2012 and 2015, while the first wave of follow-up (“T1”) was collected between 2015 and 2018. At the time of our analysis, further follow-up data were not available and hence the follow-up time was, on average, approximately 3 years.

The following CLSA datasets were used for the analysis: Baseline Comprehensive Dataset version 4.1, Follow-up 1 Comprehensive Dataset version 3.0, and Comprehensive Baseline Sample Weights Version 1.2.

Approval was obtained for the secondary analysis of data from the CLSA from the University of Saskatchewan Biomedical Research Ethics Board (Application ID: 1643).

Sampling Frames of the Comprehensive Cohort of the CLSA

CLSA comprehensive cohort participants were recruited by mail using Provincial Healthcare Registration Databases (14%) and by random digit dialing telephone sampling (86%). The CLSA excluded certain types of individuals from the two sampling frames. Sampling was restricted to individuals living within 25 km of the nearest data collection site (except for data collection site locations with lower population densities, where the radius was 50 km). Persons living in the three northern Canadian territories, persons living on First Nations reserves and other First Nations settlements in the provinces, full-time members of the Canadian Armed Forces, and persons living in institutions (e.g., long-term nursing homes, hospitals, and penitentiaries) were also excluded from the sampling frames (^{CLSA 2017}). Individuals perceived to have cognitive impairment at the time of recruitment (i.e., individuals who CLSA interviewers judged unable to understand the purpose of the study and/or provide reliable data) were excluded, as were individuals who could not respond to questions in English or French. Persons living in census tracts with high proportions of persons with lower levels of education were over-sampled (^{Raina et al. 2009}). The overall response rate (i.e., the percentage of individuals who were successfully contacted by CLSA staff and agreed to participate) was 10%. Single-stage sampling was performed (i.e., there was no geographical clustering).

Inclusion Criteria for This Analysis

Cross-sectional and longitudinal associations were analyzed in the study. Cross-sectional associations were calculated from data collected at baseline (i.e., at T0). Longitudinal associations were calculated using hearing data collected at T0 and T1, and predictor/covariate data at T0. At T0, there were 30,097 CLSA participants, while at T1, 27,765 remained in the study. Of the 2332 participants who had no data at T1, 1280 had voluntarily withdrawn, 510 had died, and 542 had no data for unknown reasons. The 30,097 participants at baseline were included in the cross-sectional analyses, and the 27,765 participants who were still members of the cohort at T1 were included in the longitudinal analyses.

Exclusion Criteria

Some participants had missing data for one or more variables included in the multivariable analyses. All results reported in the article reflect analyses drawn only from individuals who had complete data for all variables included in the multivariable models (i.e., a complete case or listwise deletion approach was used). To assess the risk of bias resulting from excluding participants with missing data, we performed a sensitivity analysis, repeating all multivariable regressions using a multiple imputation by chained equations (MICE) approach. The regression results obtained using the multiple imputation approach were qualitatively compared to the results obtained using data from participants with complete data.

Measures

Pure Tone Average.

Pure-tone hearing detection thresholds were determined using a digital screening audiometer in the automatic test mode and Audiocup headphones (Tremetrics RA300+) (^{Tremetrics RA300 and RA300+ Operations Manual 2011}). Each test was administered in a quiet room by a CLSA technician. Hearing aids were not worn during audiometric testing. Biological calibration listening checks were performed daily on the machines. Pulsed stimuli (three tone bursts over 1 sec) at 500, 1000, 2000, 3000, 4000, 6000, and 8000 Hz were presented in 5-dB increments from 0 to 100 dB HL using a modified Hughson-Westlake method (^{Margolis et al. 2015}). Further details of the automated testing procedure can be found in the operations manual of the audiometer (^{Tremetrics RA300 and RA300+ Operations Manual 2011}). “No responses” (recorded by the audiometer as error code “EE”) were recoded as 105 dB HL for the purpose of our analyses. According to CLSA protocols (CLSA Hearing-Audiometer DCS Protocol V3.0 Doc SOP_DCS_0020, 2014), individuals were not tested if they used devices (e.g. Lyric aids) that could not be removed during testing.

The primary measure of hearing used in our study was the bilateral mid-frequency PTA calculated as the average of hearing thresholds at 1000, 2000, 3000, and 4000 Hz in both ears. In a previous study (^{Mick et al. 2021}), we found that, across all 5-year age categories, the average 500-Hz threshold was unexpectedly worse than the average thresholds at 1000 and 2000 Hz, likely due to the interference of low-frequency background noise during testing (sound proof booths were not used in the CLSA). We also found that binaural PTAs predicted two “real-world” measures of hearing ability (self-rated hearing and hearing aid use) better than worse-ear averages, which in turn were superior than better-ear averages.

Diabetes

Participants were classified as having diabetes if they met any of the following three conditions: (1) their HbA1c was equal to or greater than 6.5% (^{Lipscombe et al. 2018}); or (2) they responded yes to the question, “Has a doctor ever told you that you have diabetes, borderline diabetes or that your blood sugar is high?” or (3) they responded yes to the question, “Are you currently taking medication for diabetes?” following the definition used by (Kim et al. 2017). Most participants with diabetes were identified using the question regarding physician diagnosis.

Non-fasting blood samples were obtained from participants at the CLSA data collection sites according to the CLSA phlebotomy protocol (^{Canadian Longitudinal Study on Aging 2015}).

Samples were shipped in a cryogenic shipping container to the Calgary Laboratory Services laboratory. HbA1c levels were measured using a Roche Diagnostics Cobas 8000 series analyzer (^{Roche Diagnostics GmbH 2013}).

Dyslipidemia

A complete serum lipid profile (i.e., total cholesterol, HDL cholesterol, and triglycerides) was measured using a Roche Diagnostics Cobas 8000 series analyzer (^{Roche Diagnostics GmbH 2015}). Phlebotomy and shipping protocols were as described above. LDL cholesterol was estimated using the Friedewald formula (^{Friedewald et al. 1972}).

Dyslipidemia was defined based on the Canadian guidelines for the diagnosis and treatment of dyslipidemia (^{Anderson et al. 2016}). Individuals were considered to have dyslipidemia if their low-density lipoprotein cholesterol was ≥3.5 mmol/L (63.1 mg/dL) or if their non-high-density lipoprotein cholesterol (i.e., total cholesterol minus high-density lipoprotein cholesterol) was ≥4.3 mmol/L (77.5 mg/dL), corresponding to the thresholds at which the Canadian Cardiovascular Society recommends considering initiation of pharmacotherapy for dyslipidemia.

Hypertension

Participants were classified as having hypertension if they met any of the following three conditions: (1) they responded “yes” to the question, “Has a doctor ever told you that you have high blood pressure or hypertension?” or (2) their average systolic blood pressure was ≥140 mm Hg at the time of their T0 data collection site visit; or (3) their average diastolic blood pressure was ≥90 mm Hg at the time of their T0 data collection site visit. Systolic and diastolic blood pressure were measured six times at the data collection site by a trained technician who used an automated sphygmomanometer. Average systolic and diastolic blood pressures were calculated from the final five of six measurements. Details on the CLSA blood pressure measurement protocol can be found at https://www.clsa-elcv.ca/doc/542 (^{Canadian Longitudinal Study on Aging 2014}).

Obesity

Obesity was calculated using the waist-to-height ratio, which is a measure of central obesity and is more strongly correlated to mortality and cardiovascular disease than the body mass index (^{Ashwell et al. 2012}). Waist circumference was measured using a measuring tape, according to the protocol found at https://www.clsa-elcv.ca/doc/1037 (^{Canadian Longitudinal Study on Aging 2016a}). Standing height was determined using a stadiometer following the protocol found at https://www.clsa-elcv.ca/doc/1112 (^{Canadian Longitudinal Study on Aging 2016b}). A very high risk of obesity was indicated by a waist-to-height ration of ≥0.6. A waist-to-height ratio of 0.5 is commonly used as the cutoff between “healthy” and “unhealthy” body size in adults, but some authors consider values between 0.5 and 0.6 to be an in-between range where corrective action should only be considered, rather than definitively taken (^{Ashwell et al. 2012}). We chose to use the more restrictive definition of obesity (i.e., waist-to-height ratio ≥0.6) to identify those at the highest risk of obesity-related health problems.

Smoking.

The primary smoking variable was defined using the question, “What is your smoking status?” Participants could choose from the following responses: “Yes (I currently smoke),” “No (I don’t smoke and I never have,” or “Former (I don’t smoke now but I have in the past).” Based on their response, they were classified as a current smoker, non-smoker, or former smoker, respectively. A secondary smoking variable was analyzed that estimated the lifetime number of cigarettes smoked while the participant smoked daily. Participants who had ever smoked were asked to estimate the number of years that they smoked on a daily basis, and average number of cigarettes they smoked per day during those times (possible responses included 1 to 5 per day, 6 to 10 per day, 11 to 15 per day, 21 to 25 per day, and 26+ per day). Those who smoked 26 or more cigarettes per day were asked to estimate the exact number they smoked per day. Lifetime number of cigarettes smoked while smoking daily was calculated as follows: number of cigarettes smoked daily × 365 days/year × number years smoked daily. The number of cigarettes smoked daily was estimated as the middle value for the categorical responses (i.e., for the 1 to 5 per day category, the number was estimated to be three cigarettes per day; for the 6 to 10 per day category, the number was estimated to be eight cigarettes per day, etc.). Participants who reported never smoking daily were assigned a value of zero. Cigarettes that were smoked on a non-daily basis are not included in the calculation.

Composite CVD Risk Scores.

Two composite measures of metabolic risk were created. The first was simply a count of the number of CVD risk factors, from 0 to 5, experienced by the participant. The second was the general cardiovascular disease risk function (Framingham score) developed by D’Agostino and colleagues using data from the Framingham study (^{D’Agostino et al. 2008}). We applied the score to the CLSA data to calculate a measure of CVD risk for each participant. The general CVD risk function was calculated based on age, total cholesterol, high-density lipoprotein levels, systolic blood pressure, smoking status, and presence or absence of diabetes (^{D’Agostino et al. 2008}). The score estimates the 10-year risk (%) of developing CVD. Risk is considered low if the score is less than 10%, moderate if it is 10 to 19%, and high if it is 20% or higher (^{Bosomworth 2011}).

Other Participant Characteristics.

Information about other participant characteristics was ascertained through questionnaires administered by trained interviewers. Basic demographic variables included age in years at baseline, sex, and cultural/racial background. Participants were asked to identify their sex at T0 and their gender at T1. A comparison of responses revealed that 99.7% of participants who reported female sex also reported female gender, and 99.8% of participants who reported male sex also reported male gender. We used information from only the sex variable in our analyses. Participants were asked the following question about cultural/racial background: “People living in Canada come from many different cultural and racial backgrounds. Are you…” Possible responses included White, Chinese, South Asian, Black, Filipino, Latin American, Southeast Asian, Arab, West Asian, Japanese, Korean, North American Indian, Inuit, Metis, and “Other”(^{Canadian Longitudinal Study on Aging 2018}). Ninety-four percent of the study population identified as white, so we dichotomized cultural background to “white” or “any minority.” The highest level of education was coded as (1) less than secondary school graduation; (2) secondary school graduation; (3) post-secondary education, no degree; and (4) post-secondary degree. Total household income (from all sources, before taxes and deductions, in the past 12 months) was coded as (1) less than $20,000; (2) $20,000 to $50,000; (3) $50,000 to $100,000; (4) $100,000 to $150,000; (5) $150,000 or more; or (6) refused to respond. Alcohol use was determined by the question “About how often during the past 12 months did you drink alcohol?” The original eight-category CLSA item representing frequency was transformed into a variable with five levels: 0 = never, 1 = less than once per month, 2 = one to four times per month (individuals who responded “about once a month,” “2 to 3 times a month,” or “once a week”), 3=multiple times per week (individuals who responded “4 to 5 times a week” or “2 to 3 times a week”) and 4 = daily. Menopause was assessed among females with the question: “Have you gone through menopause, meaning that your menstrual periods stopped for at least 1 year and did not restart?” with response options “Yes,” “No” or a self-report of having had a hysterectomy. Follow-up time (in years) was calculated as the difference between the two hearing test dates at T0 and T1. The data collection site attended by each participant at T0 was recorded as being either Victoria, Vancouver, Surrey, Calgary, Winnipeg, Hamilton, Ottawa, Montréal, Sherbrooke, Halifax, or St. John’s. Finally, the total Physical Activity Scale for the Elderly (PASE) score was calculated (^{Logan et al. 2013}). PASE contains a series of questions regarding 12 types of physical activity over the past 7 days. The activities included: (1) sitting activities; (2) walking outside; (3) muscle strengthening or endurance exercises; (4) light sport or recreational activities; (5) moderate sport or recreational activities; (6) strenuous sport or recreational activities; (7) light housework; (8) heavy housework; (9) home repairs or chores; (10) lawn work or yard care; (11) caring for another person; and (12) employment or volunteer work. The total PASE score was computed by multiplying either the time spent in each activity (hours/week) or participation in an activity (yes/no) by empirically derived item weights and then summing the twelve sub-scores. We followed the scoring protocol described by (^{Washburn et al. 1993}).

Statistical Analysis

Step 1. Comparisons of Participants With Complete and Incomplete Data

Characteristics of males and females with complete data were tallied. Next, differences in characteristics between individuals with complete data and incomplete data were determined using two-tailed chi-square tests for categorical variables, and two-tailed Student t test for continuous variables. For the purpose of comparison, individuals with incomplete data were classified into one of three following mutually-exclusive groups: Participants who died between T0 and T1; participants who were alive at T1, but had no T1 data; and participants who had partially missing data at T0 and/or T1.

Step 2. Comparisons of Mean PTA and Mean Difference in PTA Over Time According to Presence of Each Risk Factor

Mean PTA at T0, and mean difference in PTA between T0 and T1, were calculated for groups of participants with and without each of the CVD risk factors so that results could be reported in units familiar to hearing professionals (dB HL, and dB/year). Directly age-standardized means were calculated to adjust for age differences between comparison groups, given the importance of age as a confounder. Directly age-standardized means were calculated to adjust for age differences between comparison groups. The standard population used in calculations was the population of Canadian adults aged 45 years and older in 2016, using 1-year increments. (^{Statistics Canada 2016}). Comparisons were also stratified by age at baseline (45 to 54 years, 55 to 64 years old, 65 to 74 years old, and 75 to 86 years).

Step 3. Multivariable Linear Regression Models

Ordinary least squares linear regression models were used to estimate the independent associations between each of the CVD risk factors (and composite scores) and PTA at baseline (cross-sectional analyses). Repeated measures linear mixed models were used to estimate associations between each of the risk factors (and composite scores) at baseline and change in PTA between T0 and T1 (longitudinal analyses). In the repeated measures linear mixed models, time (as a continuous linear variable) between the T0 and T1 data collection site visits was interacted with each risk factor variable. The effect estimate of the product term represented the average difference in rate of change in PTA versus the reference. The linear mixed-effects models included a random intercept term for capturing the baseline subject-specific variability. Since the responses were measured at only two-time points, there was also limited scope for specifying the structure of covariance matrix for random effects. The choice of unstructured covariance matrix was made for this model to avoid any constraints on values. Robust standard errors were estimated.

Independent Variables.

All models examining associations between individual (single) metabolic risk factors for CVD and PTA included baseline sociodemographic (age, education) and economic (income) variables, all of the individual metabolic risk factors (the primary smoking indicator variable, obesity, dyslipidemia, hypertension, diabetes), lifestyle characteristics (alcohol consumption, PASE score) and CLSA data collection site location as independent variables. The secondary smoking variable was not included in models examining the primary smoking variable, and vice versa. Menopause status was included in all female models. In the models examining associations between composite risk factor scores at PTA, we did not adjust for the individual risk factors that comprised each of the composite scores but included all other independent variables, except the secondary smoking variable.

Age-risk Factor Interactions.

To test the potentially moderating effects of age on the associations between PTA and metabolic risk factors for CVD associations, we created a categorical age variable with four levels (45 to 54 years, 55 to 64 years, 65 to 74 years, and 75+ years old at baseline). Product terms were created between each of the age category indicator variables and each of the risk factor variables. The set of interaction terms for each risk factor (or composite score) was added to each multivariable regression model, one set at a time. Analysis of variance was performed on the model with interaction terms (without removing any main effects variables) to determine a global partial F test for the presence of interaction. If the test was significant (α = 0.05), parameter estimates for the CVD risk factor were reported for each age group.

Sex Interactions.

Sex interactions were not tested because the female and male models were not identical (female models included a term for menopause).

Step 4. Sensitivity Analysis: Multiple Imputation

Listwise deletion substantially reduced our sample to 21,492 for our complete case analysis. There were several forms of missing data including voluntary withdrawal from the study, death, unprovided data, and unknown reasons. Since participants in the complete case group were comparatively healthier, younger, and had better baseline PTA compared with incomplete groups and the results from complete case analysis could be biased, we performed multiple imputation by chained equations (MICE) technique to impute (M = 30 datasets) missing values across all variables in each analytic model as a part of sensitivity analysis and repeated all of the multivariable regression models including data for all participants. See the supplemental file in Supplemental Digital Content 1, https://links.lww.com/EANDH/B122 for details on multiple imputation methodology.

Survey Weights.

Inverse probability sampling weights were used in the descriptive analysis of males and females with complete data (step 1) and in all of the comparisons of PTA according to metabolic risk factor for CVD (statistical analyses, step 2) and multivariable analyses (statistical analyses, steps 3 and 4) to improve the national representativeness of estimates. CLSA statisticians assigned each participant a sample weight that was inversely proportional to the probability of inclusion in the study according to their sex, age, education level, and the province of residence. The most recent (2020) CLSA inflation weights were used in calculations of age-standardized PTA (step 2), and the most recent (2020) analytic weights were used in regression calculations (steps 3 and 4) (Canadian Longitudinal Study on Aging 2020). The Taylor linearization method was used for estimating standard errors that accounted for the survey weights.

Statistical Software and Level of Significance.

Stata SE version 16.1 was used to perform the analyses. The level of statistical significance was set at α = 0.05, except for analyses of associations between the two composite cardiovascular risk scores (which are conceptually similar) and PTA, and between smoking and PTA, where a Bonferroni correction was applied and the level of significance was set at α = 0.025.

RESULTS

Step 1. Comparisons of Participants With Complete and Incomplete Data

Of the 30,097 participants enrolled in the CLSA at T0, 21,492 had complete data for all variables used in the study (71.4%), and Table 1 summarizes their characteristics. At baseline, the mean PTA for females was 17.3 dB HL, and over the follow-up period, it increased on average by 2.3 dBHL (0.79 dB/year). The mean baseline PTA for males was 21.8 dB HL, and over the follow-up period, it increased by 2.7 dB (0.94 dB/year). The average age at baseline was 59.4 years for females and 59.0 years for males. Females had an average baseline general cardiovascular risk function (Framingham) score of 8.9% (low 10-year risk of CVD) while males had an average score of 16.8% (moderate 10-year risk) (^{Bosomworth 2011}). On average, females had 1.6 CVD risk factors while males had 1.7. The prevalence of risk factors was as follows: obesity: females 25.5%, males 30.6%; diabetes: females 19.6%, males 15.7%; dyslipidemia: females 32.5%, males 30.6%; hypertension: females 36.2%, males 42.0%; former smoking: females 42.7%, males 46.4%; and current smoking: females 9.7%, males 11.7%. Male former daily smokers, on average, reported smoking more daily lifetime cigarettes (9.7 × 10⁴) than female daily smokers (7.6 × 10⁴). Male current smokers smoked about the same number of daily lifetime cigarettes as females (1.9 × 10⁶).

TABLE 1. - Characteristics of females and males with complete data for both

Variable	Category	Females	Males
N		10,752	10,740
Baseline (T0) PTA* [mean dB HL, 95% CI])		17.3 (17.0–17.6)	21.8 (21.5–22.2)
Difference in T0 PTA per 1-yr age increment (mean dB, 95% CI)		0.73 (0.70–0.75)	0.87 (0.84–0.90)
Change in PTA between T0 and T1 (mean dB, 95% CI)		2.3 (2.2–2.4)	2.7 (2.6–2.9)
Rate of change in PTA (mean dB/yr, 95% CI)		0.78 (0.73–0.83)	0.94 (0.89–1.00)
PTA category at baseline (T0) (mean %, 95% CI)	Normal (≤25 dBHL)	79.4 (78.4–80.4)	66.7 (65.4–67.9)
	Mild (25.1–40 dBHL)	14.7 (13.8–15.5)	21.3 (20.2–22.4)
	Moderate (40.1–60 dBHL)	5.2 (4.7–5.8)	10.2 (9.5–10.9)
	Severe ≥60 dBHL)	0.7 (0.5–0.9)	1.8 (1.5–2.1)
Socio-demographic and economic characteristics
Age (mean years) (95% CI)		59.4 (59.2–59.7)	59.0 (58.8–59.3)
White culture/ethnicity (mean %, 95% CI)		96.4 (95.9–97.0)	95.5 (94.9–96.1)
Education (mean %, 95% CI)	Post-sec degree graduate	63.5 (62.1–64.8)	66.4 (64.9–67.9)
	Some post-sec education	9.1 (8.4–9.9)	9.3 (8.5–10.10
	Secondary school grad	12.2 (11.4–13.0)	10.4 (9.6–11.2)
	<Secondary school	15.2 (13.9–16.6)	14.0 (12.5–15.5)
Household income (mean %, 95% CI)
	Refused	3.5 (3.1–4.1)	2.6 (2.2–3.0)
	Don’t know	3.3 (2.8–3.9)	1.1 (0.9–1.3)
	≥$150,000	14.7 (13.8–15.5)	20.0 (19.0–21.0)
	$100,000–149,999	17.7 (16.7–18.7)	22.6 (21.5–23.7)
	$50,000–99,999	32.6 (31.4–33.8)	33.6 (32.3–34.9)
	$20,000–49,999	22.4 (21.4–23.5)	16.7 (15.7–17.8)
	<$20,000	5.8 (5.2–6.6)	3.5 (2.8–4.4)
Cardiovascular risk factors
Obesity (mean %, 95% CI)		25.5 (24.3–26.7)	30.6 (29.3–31.9)
Diabetes (mean %, 95% CI)		19.6 (18.5–20.7)	15.7 (14.9–16.8)
Dyslipidemia (mean %, 95% CI)		32.5 (31.3–33.8)	30.6 (29.3–31.9)
Hypertension (mean %, 95% CI)		36.2 (35.0–37.5)	42.0 (40.6–43.4)
Smoking status (mean%, 95% CI)
	Never	47.6 (46.5–49.1)	41.8 (40.4–43.1)
	Former	42.7 (41.1–43.7)	46.4 (46.3–49.1)
	Current	9.7 (9.0–10.7)	11.7 (9.5–11.7)
Lifetime cigarettes smoked when smoked daily (mean × 1,000, 95% CI)	Former smokers	75.7 (72.3–79.1)	97.2 (92.8–101.7)
	Current smokers	188.8 (174.3–203.4)	193.9 (179.7–208.1)
No. of metabolic risk factors (mean No., 95% CI)		1.62 (1.59–1.65)	1.74 (1.71–1.77)
General cardiovascular risk function (mean %, 95% CI)		8.9 (8.7–9.1)	16.8 (16.5–17.2)
Lifestyle characteristics
Alcohol (mean %, 95% CI)
	Never	15.4 (14.4–16.5)	12.7 (11.3–13.2)
	Less than once/mo	15.4 (14.5–16.4)	9.1 (8.3–10.0)
	1–4 times/mo	29.9 (28.8–31.1)	26.5 (25.3–27.7)
	Multiple times/mo	28.5 (27.4–29.6)	34.7 (33.4–36.0)
	Daily	10.8 (10.1–11.5)	17.5 (16.5–18.6)
PASE (mean score, 95% CI)		139.9 (138.0–141.9)	166.5 (164.1–168.8)

Percentages and means are weighted averages, calculated using survey weights assigned by the Canadian Longitudinal Study on Aging.

^*PTA: pure-tone threshold average at 1000, 2000, 3000, and 4000 Hz in both ears.

CI, confidence interval; PASE, Physical Activity Scale for the Elderly.

Hearing, health, and socioeconomic indicators generally deteriorated with increasing levels of missing data increased (see Supplementary table 1 in Supplemental Digital Content 2, https://links.lww.com/EANDH/B123). At baseline, for example, 27% of individuals with complete data were obese, compared to 32% of participants missing some data at either time point, 34% of participants who dropped out of the CLSA between T0 and T1, and 43% of participants who died between T0 and T1, respectively. Similar trends are seen for all of the metabolic risk factors for cardiovascular disease except dyslipidemia. Participants with missing data were older and had lower socioeconomic status than participants with complete data.

Step 2. Comparisons of Mean PTA and Mean Differences in PTA Over Time According to Presence of Each Risk Factor. Results are presented in Supplementary Tables 2 and 3 in Supplemental Digital Content 2, https://links.lww.com/EANDH/B123

Step 3. Multivariable Linear Regression Models

Cross-Sectional Results

Diabetes, current smoking, and higher composite risk scores were associated with worse PTA for both sexes. Obesity was also associated with worse PTA among males and females in the 55 to 64-year-old age group (Tables 2 and 3; full output in Supplementary table 4, Supplemental Digital Content 2, https://links.lww.com/EANDH/B123). The magnitude of statistically significant effect estimates for metabolic risk factors were compared to the mean difference in baseline PTA per 1 year of age (females: 0.73 dB per 1 year of increased age; males: 0.87 dB per 1 year of increased age; Table 1) to give a sense of clinical relevance. The results are as follows: females: obesity: 1.37 dB/0.73 dB/yr = 1.88 years (i.e., the average ΔPTA associated with having diabetes is equivalent to the average ΔPTA associated with an age difference of 1.88 years); current smoking: 1.45 dB/0.73 dB/year = 1.99 years; 1 additional metabolic risk factor for CVD: 0.32 dB/0.73 dB/year = 0.44 years; change of 10% in Framingham score: 0.43 dB/0.73 dB/year = 0.59 years; males: obesity: 0.97 dB/0.87 dB/year = 1.11 years; diabetes: 0.95 dB/0.87 dB/year = 1.09 years; current smoking: 1.71 dB/0.87 dB/year = 1.97 years; one additional metabolic risk factor for CVD: 0.56 dB/0.87 dB/year = 0.64 years; Increase of 10% in Framingham score: 0.39 dB/0.87 dB/year = 0.45 years.

TABLE 2. - Independent cross-sectional and longitudinal associations between CVD risk factors and PTA

	Females		Males
	Cross-sectional	Longitudinal	Cross-sectional	Longitudinal
	β (dB) (95% CI, p)	β (dB/yr) (95% CI, p)	β (dB) (95% CI, p)	β (dB/yr) (95% CI, p)
Single risk factors (indicator variables)
Obesity*	0.19 (−0.49 to 0.89, 0.574)	0.12 (0.01 to 0.23, 0.035)	0.97 (0.18 to 1.75, 0.016)	0.05 (−0.09 to 0.18, 0.485)
Diabetes*	0.89 (0.08 to 1.70, 0.031)	−0.03 (−0.17 to 0.11, 0.691)	0.95 (0.10 to 1.81, 0.029)	0.006 (−0.11 to 0.12, 0.925)
Dyslipidemia*	−0.24 (−0.78 to 0.30, 0.384)	0.06 (−0.04 to 0.17, 0.217)	0.46 (−0.21 to 1.14, 0.175)	−0.15 (−0.25 to 0.01, 0.064)
Hypertension*	0.16 (−0.43 to 0.76, 0.587)	0.18 (0.08 to 0.28, <0.001)	−0.37 (−1.03 to 0.30, 0.278)	0.22 (0.10 to 0.33, <0.001)
Smoking*
Former	−0.03 (−0.52 to 0.51, 0.990)	0.05 (−0.06 to 0.15, 0.399)	0.56 (−0.11 to 1.23, 0.103)	0.32 (0.22 to 0.43, <0.001)
Current	1.45 (0.29 to 2.60, 0.014)	−0.08 (−0.28 to 0.12, 0.447)	1.71 (0.58 to 2.85, 0.003)	0.29 (0.05 to 0.53, 0.018)
Composite scores
No. of metabolic risk factors present†	0.32 (0.05 to 0.60, 0.022)	0.06 (0.01 to 0.11, 0.021)	0.56 (0.26 to 0.86, <0.001)	0.06 (−0.02 to 0.11, 0.136)
General cardiovascular risk function (per 10% increment)‡	0.43 (0.37 to 0.50, <0.001)	0.28 (0.21 to 0.35, <0.001)	0.39 (0.36 to 0.43, <0.001)	0.15 (0.11 to 0.19, <0.001)

Cross-sectional results are based on ordinary least squares multivariable linear regression, while longitudinal results are based on repeated measures linear mixed regression models. For cross-sectional results, the beta coefficient represents the marginal difference in pure-tone threshold average (dB HL) versus the reference group without the risk factor at T0. For longitudinal results, the beta coefficient represents the marginal difference in change in pure-tone threshold average (dB HL/yr) versus the reference group without the risk factor. Bold indicates statistical significance.

^*Cross-sectional models for each sex were adjusted for age, education, white race/ethnicity, income, obesity, diabetes, dyslipidemia, hypertension, smoking, alcohol consumption, PASE score, and CLSA data collection site. Female models also adjusted for menopause status. Cross-sectional results are obtained from a single model for all individual risk factors (for each sex). Separate longitudinal models were performed on one risk factor at a time. Time (as a continuous variable) between data collection site visits was interacted with each individual risk factor. The time#risk factor measures of association are reported. Longitudinal models also adjusted for each of the other individual risk factors, age, education, income, alcohol consumption, PASE score, and CLSA data collection site. Female models also adjusted for menopause status.

^†Cross-sectional models for each sex were adjusted for age, white race/ethnicity, education, alcohol consumption, PASE score, and CLSA data collection site. Female models also adjusted for menopause status. The longitudinal models for each sex-adjusted for time#(no. of risk factors present), along with the other variables listed for the cross-sectional model above. Female models also adjusted for menopause status.

^‡Cross-sectional model adjusted for age, education, white race/ethnicity, alcohol consumption, PASE score, and CLSA data collection site. Female models also adjusted for menopause status. The longitudinal model adjusted for time#(no. of risk factors present), along with the variables listed for the cross-sectional models. Female models also adjusted for menopause status.

CI, confidence interval; CLSA, Canadian Longitudinal Study on Aging; CVD, cardiovascular disease; PASE, Physical Activity Scale for the Elderly; PTA, pure-tone threshold average.

TABLE 3. - Age group-stratified results for cross-sectional associations between CVD risk factors and PTA for which there was a significant interaction between the risk factor and age group (females only)

	Beta Coefficient (95% Confidence Interval, p)
	P for Interaction*	45–54-yr-old Model	55–64-yr-old Model	65–74 yr-old Model	75–86-yr-old Model
Obesity†	0.016	0.40 (−0.81 to 1.62, 0.514)	1.37(0.14 to 2.59, 0.029)	−0.78(−2.09 to 0.53, 0.242)	−1.63(−3.64 to 0.37,0.110)
General CV risk function (Framingham score) (/10% change)‡	0.039	0.20 (0.08 to 0.31, 0.001)	0.24 (0.13 to 0.35, <0.001)	0.10 (0.01 to 0.19, 0.033)	0.07 (−0.10 to 0.25, 0.417)

Effect estimates were calculated using multivariable ordinary least squares linear regression models performed separately for each age group. Bold indicates statistical significance.

^*p-value for the product term between the four-category age variable and the risk factor when the product term was added to the main effects model.

^†The model was adjusted for age, education, white race/ethnicity, income, diabetes, dyslipidemia, hypertension, smoking, alcohol consumption, PASE score, menopause, and CLSA data collection site. The number of participants with obesity in each age group is as follows: age 45–54 yrs: obese: 485; not obese: 2402. Age 55–64 yrs: obese: 962; not obese: 2822. Age 65–74 yrs: obese: 737; not obese: 1828. Age 75–86 yrs: obese: 414; not obese: 1102.

^‡The model was adjusted for age, education, white race/ethnicity, income, alcohol consumption, PASE score, menopause, and CLSA data collection site.

CI, confidence interval; CLSA, Canadian Longitudinal Study on Aging; CVD, cardiovascular disease; PASE, Physical Activity Scale for the Elderly; PTA, pure-tone threshold average.

Interactions With Age (Cross-Sectional Models).

In females, age significantly modified the cross-sectional associations involving obesity and the general CVD risk function score. Generally, the associations were stronger in younger age categories and particularly in the 55 to 64-year-old category (Table 3).

Longitudinal Results.

Hypertension and higher general cardiovascular risk function (Framingham) scores were associated with faster declines in hearing among both sexes. For women, obesity and number of metabolic risk factors for CVD were also associated with faster hearing loss, while for men, former and current smoking were also associated with faster rates of hearing loss (Table 2). Age did not significantly modify any of the longitudinal associations. Statistically significant effect estimates (units: dB/year) were calculated as a percentage of the average rate of change in PTA for all members of the cohort (females: 0.78 dB/year; males: 0.94 dB/year (Table 1)) to give a sense of clinical relevance. Results are as follows: females: obesity: 0.12/0.78 = 15% (i.e., the Δrate in PTA associated with obesity is equivalent to 15% of the average rate of PTA for all women in the cohort); hypertension: 0.18/0.78 = 23%; an increase in 1 metabolic risk factor for CVD: 0.06/0.78 = 8%; a change of 10% in Framingham score: 0.28/0.78 = 36%. Males: hypertension: 0.22/0.94 = 19%; former smoking: 0.32/0.94 = 34%; current smoking: 0.29/0.94 = 31%; an increase of 10% in Framingham score: 0.15/0.94 = 16%.

Analysis of Smoking Using # Lifetime Cigarettes Smoked Daily.

For males, a greater number of lifetime cigarettes smoked daily predicted worse hearing cross-sectionally (β = 0.07 dB HL/10,000 cigarettes, p < 0.001) and over time (β = 0.009 dB/year per 10,000 cigarettes, p < 0.001). For females, the cross-sectional association was not significant (p = 0.064). The p-value for the longitudinal association was 0.043, which was not significant after Bonferroni correction was applied (α = 0.025) (see Supplementary Table 5 in Supplemental Digital Content 2, https://links.lww.com/EANDH/B123).

Step 4. Sensitivity Analysis: Multiple Imputation (MI) With Chained Equations

Supplementary Table 6 in Supplemental Digital Content 2, https://links.lww.com/EANDH/B123, summarizes significant associations as determined using the complete case approach (i.e., including only participants with complete data) versus the MI approach (i.e., imputing missing data for those with missing observations). The MI approach confirmed all of the significant associations observed in the complete case analysis. In addition, there is an association between dyslipidemia and slower audiometric hearing loss over time among males (β = −0.19 dB/year, p = 0.001).

DISCUSSION

In summary, in cross-sectional analyses, diabetes, obesity, current smoking, and both composite measures of CVD risk were independently associated with worse PTA for both sexes. In females, obesity was only associated with PTA among the 55 to 64-year-old group. The association between general cardiovascular risk function (Framingham score) was stronger among females aged 45 to 64 years old compared to older female age groups. Parameter estimates for CVD risk factors appeared to be important when compared to the parameter estimate for age.

In longitudinal analyses, hypertension and higher general cardiovascular risk function (Framingham) scores were independently associated with faster rates of PTA increases for both sexes. In addition, obesity and the presence of a greater number of risk factors were associated with faster rates of pure-tone hearing loss in females, and former and current smoking were associated with faster rates of pure-tone hearing loss in males (versus males who had never smoked). Parameter estimates for CVD risk factors appeared to be important when compared to the average rate of pure-tone hearing loss for the males and females cohort as a whole. Dyslipidemia was the only CVD risk factor that showed no associations with hearing loss, which is consistent with other studies which have shown no or small associations between dyslipidemia and hearing. (^{Axelsson & Lindgren 1985; Fuortes et al. 1995; Shargorodsky et al. 2010; Helzner et al. 2011; Simpson et al. 2013}) Engdahl et al. found that higher levels of LDL cholesterol were associated with better hearing (^{Engdahl et al. 2015}). In the complete case analysis, the association between dyslipidemia and change in hearing over time was almost protective in males (β = −0.15, p = 0.064). The protective association was significant when multiple imputation was used.

The observed cross-sectional associations could be explained by CVD risk factor-mediated changes in PTA that occurred over the long term, before the observation period. Diabetes was associated with hearing loss cross-sectionally but not longitudinally for both sexes, as was obesity for males, and current smoking in females. The findings raise the possibility that in these cases, the underlying mechanism(s) driving the associations may have ceased to operate before the observation period, or continued to occur but imperceptibly. Linssen et al. found that in the Maastricht study (age range 24 to 83 years) obesity was associated with faster rates of hearing loss in young adults (age 24 to 42 years) and not older adults (^{Linssen et al. 2014}). ^{Kim et al. (2017}) found that in the Kangbuk Samsung Health Study (age range 18 years and older), diabetes was more strongly associated with hearing decline among participants younger than 50 years old (hazard ratio 1.62 [95% CI 1.29 to 2.04]) than participants 50 years and older (hazard ratio 1.27 [95% CI 1.07 to 1.51]). Younger age of onset of type 1 diabetes is associated with poorer cardiovascular outcomes and higher mortality, with the highest excess risk in women (^{Rawshani et al. 2018}). Atherosclerosis begins in childhood or adolescence and its rate of progression is determined by diabetes and other risk factors for CVD (^{McGill Jr et al. 1995,1998; Strong et al. 1999}).

For females, current smoking status was strongly associated with worse PTA values at baseline but not with faster increases in PTA over time. In contrast, smoking status was associated with worse hearing both cross-sectionally and longitudinally for men. The difference could be due to lower lifetime exposure to cigarette smoke in female smokers, who smoked fewer daily cigarettes than males in the CLSA, consistent with other studies that show that in general, women smoke less, have a later age of onset of smoking, and have lower nicotine dependence (^{Shiffman & Paton 1999; Thompson et al. 2015}). The analysis of the secondary smoking variable (number of cigarettes smoked daily) hinted that higher levels of lifetime smoking might be associated with faster declines in both sexes. Previous cross-sectional studies have also shown weak evidence of a dose-response relationship between smoking and poorer hearing (^{Cruickshanks et al. 1998; Dawes et al. 2014}). The finding that smoking is more strongly associated with hearing loss in males is consistent with a small clinical study by Lisowska et al. (n = 84), who found that although smoking was associated with decreases in click- and distortion-product-evoked otoacoustic emission levels among males but not females. In this study, smoking was not associated with pure tone thresholds (^{Lisowska et al. 2017}). Pre-natal exposure to smoking, and smoking itself, were associated with greater auditory attention deficits for male adolescents compared to female adolescents (^{Jacobsen et al. 2007}). Wang et al. found that smoking status was associated with audiometric hearing loss among middle-aged men, but not men of other age groups or women. Only 2% of the women in the study (based in China) smoked, however, likely reducing the statistical power to detect significant results in women (^{Wang et al. 2021}). In contrast, Demir et al. found that hearing loss may occur at lower rates of smoking (measured in lifetime pack-years) in women than in men (^{Demir et al. 2019}). The findings of our study are in contrast to a large meta-analysis that demonstrated that the relative risk of current smoking (versus not smoking) on coronary heart disease was 25% higher for women than men (^{Huxley & Woodward 2011}). The biological mechanisms underlying sex differences in CVD risk associated with smoking are not known.

Although smoking was more strongly associated with hearing loss in males, composite measures of CVD risk were more strongly associated with hearing loss in females, which is consistent with the large body of literature showing that overall, metabolic risk factors are more closely associated with CVD risk in women (^{Vogel et al. 2021}). Clear sex or gender differences do not emerge in the hearing loss literature, but most studies have not disaggregated by sex or gender. Goderie et al. examined associations between CVD risk factors and speech discrimination among a cohort of adults in the Netherlands, and none of the associations were modified by sex (^{Goderie et al. 2021}). A Japanese study showed that higher levels of hemoglobin A1c were associated with hearing declines among both females and males (^{Nagahama et al. 2018}). Helzner et al., analyzing data from the Health ABC study, found that diabetes was associated with hearing loss in both white males and females (but not Black males or females), hypertension was associated with hearing loss in white males (but not in other race/sex groups), and current smoking was associated with hearing loss in Black females (but not in the other race/sex groups) (^{Helzner et al. 2005}). Since 94% of CLSA participants self-identified as white, we did not have the power to disaggregate according to race or ethnicity. We encourage other authors to present sex-disaggregated results so that meta-analyses can be performed to clarify whether there are significant sex-related differences in associations between metabolic risk factors for CVD and hearing loss.

In females, the stronger associations between PTA and obesity in the 55 to 64-year-old age group, and the stronger associations between PTA and Framingham scores in the 45 to 65-year-old age group are, to our knowledge, unique findings. At older ages, the larger variance in PTA arising from important competing risk factors might mask the associations with metabolic risk factors for CVD if the other factors are imperfectly measured and thus not completely controlled in statistical models. In the case of ARHL, chronological and biological age are not perfectly correlated, and thus adjusting for chronological age does not necessarily adjust for the biological age of the ear (^{Jylhävä et al. 2017}). Similarly, the proportional effects of metabolic risk factors on the risk of developing CVD (e.g., ischemic heart disease, hypertensive heart disease, and stroke) decline with age due to competing risks that increase with age (^{Berry et al. 2012; Singh et al. 2013}). It is unclear why obesity was associated with worse hearing among females aged 55-64 years but not females aged 45-54 years. The relatively low prevalence of poor hearing (^{Appelman et al. 2015; Mick et al. 2019}) may have reduced the statistical power to detect an association in younger age groups.

An obvious question arising from the findings is whether, at the individual or population levels, improvements in CVD risk factor profiles might slow hearing loss. The question cannot be directly answered from the current study, but future CLSA studies, restricted only to participants with hypertension, dyslipidemia, or diabetes, could be performed to determine if medications for these conditions modify the risk of hearing loss progression. In the current study, former smokers had a lower risk of hearing loss relative to current smokers in cross-sectional analyses, and a greater number of lifetime cigarettes was predictive of worse hearing, providing some support to the notion that quitting smoking could have beneficial effects on hearing, with the caveat that the study is not designed to assess causal relationships. Hu and colleagues reported similar findings in a longitudinal study from Japan, although males and females were not disaggregated (^{Hu et al. 2019}). There do not appear to be any published clinical trials that directly examine the question of whether treatments to improve the CVD risk factor profile affect hearing outcomes. Importantly, individuals can improve their cardiovascular risk profile, for example with health behavior modifications (e.g., exercise, diet) and medications (e.g., medications to treat hypertension, dyslipidemia, and hyperglycemia). Future trials might examine the effects of metabolic risk factor interventions on hearing outcomes, following the protocol of Lowthian et al. (²⁰¹⁶).

Given that associations were present across the entire age range in our study, screening programs for ARHL might be targeted at middle-aged adults in their forties or fifties if one of the goals is to counsel participants about ways to address potentially modifiable risk factors that could potentially reduce the severity of ARHL. It should be re-emphasized that the present study design does not support a causal interpretation of the significant associations. There would be a little downside, however, to encouraging adults participating in a screening program to address risk factors for CVD (especially if they have multiple risk factors), with the knowledge that there could potentially be a benefit to hearing.

Limitations

Threats to causal inference include residual or unmeasured confounding, for example, the lack of adjustment for noise exposure or occupational history, since noise is associated with both cardiovascular disease and hearing loss (^{World Health Organization; Regional Office for Europe, 2011}). We cannot definitively rule out common causes or incomplete adjustment for confounders as explanations for significant associations, even though the regression models included a large number of independent variables. In cases where CVD risk factors were associated with change in PTA cross-sectionally but not longitudinally, we cannot completely rule out reverse causation, but it is unlikely because a biological pathway from pure-tone hearing loss to an unhealthy cardiovascular risk profile has not been demonstrated. The follow-up time was relatively short (about 3 years) and may not have been long enough to detect all meaningful longitudinal associations. The CLSA has a broad age range but some of the associations of interest may be stronger in even younger cohorts. We did not have information on the length of time living with each CVD risk factor and the trajectory of each condition. Some of the risk factor data (e.g., lipid profile) were not available at T1 at the time of our analysis. The CLSA is a relatively privileged cohort in terms of health and socioeconomic status relative to the Canadian population. There is an over-representation of people who report white culture/ethnicity in the CLSA relative to the national population of seniors. The disparities are even more pronounced when comparing participants with complete data to the general population. Thus, our findings may not be generalizable to all populations and may underestimate the strength of associations because there were relatively few participants with a poor risk factor profile.

CONCLUSIONS

The study confirmed that CVD risk factors (except dyslipidemia), alone and in combination, increased the risk of audiometric hearing loss, both cross-sectionally and over time. Compared to the overall rate of hearing loss over time, parameter estimates for CVD risk factors in longitudinal analyses were reasonably large. Associations differed between males and females, with smoking, for example, being associated with faster increases in PTA for males but not females, and the opposite being true for obesity. Composite measures of CVD risk were associated with faster increases in PTA among women compared to men. Diabetes was associated with PTA in cross-sectional but not longitudinal analyses, raising the possibility that the effect of diabetes on cochlear function may occur at younger ages, as the youngest member of the cohort was 45 years old. Obesity and Framingham score were more strongly associated with PTA among younger females in the cohort (versus older females). Longer-term follow-up and analyses of different cohorts are needed to clarify the degree to which associations differ according to sex, age, and populations not well-represented in the CLSA. Further work is also needed to determine if treating risk factors for CVD modifies the risk of hearing loss. In the meantime, considering the vast number of other health benefits that would likely arise from improving the CVD risk factor profile, there would be little harm in counseling individuals that quitting smoking, improving diabetes and blood pressure control, and reducing obesity might prevent some degree of hearing loss.

ACKNOWLEDGMENTS

This research was made possible using the data/biospecimens collected by the Canadian Longitudinal Study on Aging (CLSA). Funding for the Canadian Longitudinal Study on Aging (CLSA) is provided by the Government of Canada through the Canadian Institutes of Health Research (CIHR) under grant reference: LSA 94473 and the Canada Foundation for Innovation. The present research was conducted using the CLSA Baseline Comprehensive Dataset version 4.1, Follow-up 1 Comprehensive Dataset version 3.0, and Comprehensive Baseline Sample Weights Version 1.2, under Application Number 2002002. The CLSA is led by Drs. Parminder Raina, Christina Wolfson, and Susan Kirkland. The authors had full editorial control over the design and reporting of the present research using CLSA data. The opinions expressed in this article are the authors’ own and do not reflect the views of the CLSA.

The study was funded by a College of Medicine (CoMRAD) grant from the University of Saskatchewan and an educational grant from Sonova AG for the purpose of independent medical research. The funding agreement with Sonova specified that the sponsor had no right to sensor the article or preclude publication.

P.T.M. conceived and designed the study, wrote the research protocol, applied for funding, obtained data from the CLSA, supervised the data analyses, analyzed the data, and wrote the article. R.K. designed the study, performed the data analyses, prepared tables and the supplementary materials, analyzed the data, and wrote the article. M.K.P.-F., C.J., N.P., and E.U. designed the study, analyzed the data, and edited the article. L.M. performed a literature review, analyzed the data, and edited the article. All authors discussed the study design, results, and implications, contributed to writing the article, and approved the final version of the article.