Diabetes and Hearing Impairment in the United States: Audiometric Evidence from the National Health and Nutrition Examination Survey, 1999 to 2004 FREE

—The Editors

Hearing loss, reported by more than 17% of the U.S. adult population, is a major public health concern affecting more than 36 million people (1). Risk for hearing impairment is associated with male sex, lower education, industrial or military occupation, and leisure time noise exposure (2 - 4), and prevalent hearing impairment has been correlated with smoking (5). Prevalence varies substantially by age, sex, and race, and estimates exceed 30% among those age 65 years or older (1). In 1 community-based study, 46% of the population age 43 to 84 years was classified as hearing-impaired on the basis of audiometric examination (6). These high-prevalence estimates imply that many people are at risk for functional and psychosocial limitations associated with hearing impairment (7 - 8).

Diabetes mellitus affects an estimated 9.6% of the U.S. adult population (9 - 10) and is associated with microvascular and neuropathic complications affecting the retina, kidney, peripheral arteries, and peripheral nerves (11). The pathologic changes that accompany diabetes could injure the vasculature or the neural system of the inner ear, resulting in sensorineural hearing impairment. Two studies (12 - 13) described evidence of such pathologic changes, including sclerosis of the internal auditory artery, thickened capillaries of the stria vascularis, atrophy of the spiral ganglion, and demyelination of the eighth cranial nerve among patients with diabetes in whom autopsy was done. Clinical evidence supporting an association between diabetes and hearing impairment is limited to several small studies (14 - 18) or noise-exposed samples (19). Epidemiologic evidence from 1 population-based cohort study suggested a modest association (20). We used recent national survey data to examine the relationship between diabetes and hearing impairment. Specifically, we designed this analysis to determine whether hearing impairment is more prevalent among U.S. adults who report a diagnosis of diabetes than those who report no diagnosis and whether differences in prevalence by diabetes status occur predominantly in specific U.S. population subgroups.

Data from NHANES (National Health and Nutrition Examination Survey) were collected by the National Center for Health Statistics from 1999 to 2004 by using a complex, multistage, probability sample designed to represent the civilian, noninstitutionalized U.S. population. Half of the study participants (n = 11 405) age 20 to 69 years were randomly assigned to audiometric testing. Of the 5742 assigned, we included 5140 (89.5%) persons who completed the audiometric examination and the diabetes questionnaire in this analysis. Major reasons for not completing an examination included time limitation (n = 128 [2.2%]), physical limitation (n = 60 [1.0%]), communication problem (n = 42 [0.7%]), refusal (n = 81 [1.4%]), and equipment failure (n = 47 [0.8%]). Included among the 60 participants with a physical limitation is an unknown number of participants who were not tested because they could not remove their hearing aids; 7 of these participants reported diabetes.

As part of the NHANES survey, pure tone air conduction hearing thresholds were obtained for each ear at frequencies of 500, 1000, 2000, 3000, 4000, 6000, and 8000 Hz. Higher frequencies are perceived as higher pitches. Audiologists usually consider tones of 500 Hz or less to be low frequency, tones from 1000 to 2000 Hz to be of mid-range frequency, and tones of 3000 Hz or greater to be high frequency. The measurements were collected by trained audiometric technicians by using a calibrated audiometer that met accepted standards (Appendix).

We derived measures of hearing impairment for 2 ranges of frequency (low or mid and high) and 2 categories of severity (mild or greater and moderate or greater). To produce low- or mid-frequency pure tone averages, we averaged pure tone thresholds (the signal intensities needed to perceive the tones) measured at 500, 1000, and 2000 Hz for each individual and ear (21). We averaged pure tone thresholds measured at 3000, 4000, 6000, and 8000 Hz (22 - 23) for each individual and ear to produce high-frequency pure tone averages. For each frequency range, a pure tone average greater than 25 decibels hearing level (dB HL) defined hearing impairment of mild or greater severity, whereas a pure tone average greater than 40 dB HL defined hearing impairment of moderate or greater severity (24). For each combination of frequency range and severity, we defined hearing impairment in terms of the pure tone average in the worse ear, which designates persons with impairment in at least 1 ear. We also defined hearing impairment in terms of the better ear, which designates persons with impairment in both ears (a subset of the persons impaired in at least 1 ear). Table 1 shows functional descriptions of hearing impairment, by frequency range and severity. In addition, we classified participants as having self-reported hearing impairment if they reported having a little trouble hearing, having a lot of trouble hearing, or being deaf without a hearing aid (1).

Among the 5140 participants, the National Center for Health Statistics identified 24 participants with at least 1 audiometric nonresponse (that is, participants did not perceive the pure tone at any level of intensity). We classified these cases as impaired for a frequency range if the audiometric nonresponse occurred within the range. An examination of these participants' available pure tone thresholds corroborated their classification as impaired at both levels of severity.

Information on demographic characteristics, diagnosed diabetes, noise exposure, medication use, and smoking was obtained during in-home interviews. Education was assessed as the highest grade level or degree attained. Income–poverty ratio was defined as the ratio of reported total family income to the U.S. Census Bureau poverty threshold, which varies by family size and age of family members. Diagnosed diabetes was assessed with the question, “Other than during pregnancy (for women), have you ever been told by a doctor or health professional that you have diabetes or sugar diabetes?” Of the 5140 participants, 2259 received an additional random assignment to a fasting protocol and subsequent blood draw. Of the 2259 participants, 146 reported a diagnosis of diabetes. Of the remainder, 73 participants were classified as having undiagnosed diabetes (fasting plasma glucose level ≥7 mmol/L [≥126 mg/dL]) and 539 were classified as having impaired fasting glucose (fasting plasma glucose level ≥5.6 mmol/L [≥100 mg/dL], but <7 mmol/L [<126 mg/dL]). The remaining 1501 participants were defined as having normal glycemic status.

Occupational noise exposure was defined as reporting a history of loud noise at work that required speaking in a loud voice to be heard. Leisure-time noise exposure was based on participant recall of noise from firearms (outside of work) or other sources (such as loud music or power tools) for an average of at least once a month for 1 year. History of military service was determined from a question asking about ever having served in the U.S. Armed Forces. Use of ototoxic medications was assessed by a review of medication containers. Because the small proportion of adults reporting use of aminoglycoside antibiotics (0.03%), loop diuretics (1.5%), antineoplastic drugs (5.0%), and nonsteroidal anti-inflammatory drugs (7.3%) precluded analysis of these medications individually, we defined use of ototoxic medication as use in the past 30 days of any of these 4 drug classes.

Differences in the distribution of sociodemographic characteristics, military history, noise exposure (leisure time and occupational), ototoxic medication use, smoking, and diagnosed diabetes were tested by using the t test (for continuous characteristics) or chi-square test (for categorical characteristics). Unadjusted prevalence estimates and 95% CIs for the hearing impairment outcomes were assessed by diagnosed diabetes status. Prevalence estimates were additionally stratified by sociodemographic characteristics, military history, leisure-time noise exposure, occupational noise exposure, ototoxic medication use, and smoking to identify population subgroups that may be particularly vulnerable to diabetes-related hearing impairment. Age-adjusted prevalence estimates were computed by direct standardization to the 2000 U.S. Census population by using age categories of 20 to 49 years, 50 to 59 years, and 60 to 69 years. Statistical significance of the difference between unadjusted estimates was determined from chi-square test statistics for a general association, and the Cochran–Mantel–Haenszel chi-square test was used to determine the statistical significance of the difference between age-adjusted estimates. For the 2259 participants who had been randomly assigned to the fasting protocol, age-adjusted prevalence estimates of high-frequency hearing impairment were generated by glycemic status (diagnosed diabetes, undiagnosed diabetes, impaired fasting glucose, or normal). Odds ratios (with 95% CIs) for the independent association of diabetes with hearing impairment were estimated by using multiple logistic regression models, adjusting for age, sex, race or ethnicity, education, income– poverty ratio, leisure-time noise exposure, occupational noise exposure, history of military service, use of ototoxic medications, and smoking. Age was treated as a continuous variable in all regression models. Nonlinear effects of age on the logit of each outcome were examined by testing the addition of an age squared term to each model but were not statistically significant. By using the concordance index, we assessed predictive accuracy, which ranged from 80% to 90% for each of the 8 audiometrically assessed outcomes and was 72% for self-reported hearing impairment. Six of the 9 models passed the Hosmer–Lemeshow goodness-of-fit tests. Finally, the frequency-specific pure tone thresholds were examined graphically by averaging within-person thresholds over both ears and plotting the age-adjusted and age-specific mean thresholds stratified by diagnosed diabetes status.

We used SAS software, version 9.1 (SAS Institute, Cary, North Carolina), and SUDAAN, version 9.0.1 (Research Triangle Institute, Research Triangle Park, North Carolina), for all analyses and incorporated 6-year sample weights that were adjusted for oversampling of ethnic minorities, elderly persons, and those of low income; eligibility of half of the sample for audiometric testing; and nonresponse of eligible individuals who were not tested. We computed 6-year audiometric sample weights by assigning two thirds of the 4-year audiometric weight for persons sampled from 1999 to 2002 and one third of the 2-year audiometric weight for persons sampled from 2003 to 2004.

The U.S. Department of Health and Human Services funds NHANES and oversees the conduct and reporting of the NHANES. As employees or contractors of the U.S. Department of Health and Human Services, the authors had a direct role in the reporting and analysis of data and the decision to submit the manuscript for publication.

Table 2 shows characteristics of the U.S. population, stratified by low- or mid-frequency hearing impairment of mild or greater severity assessed in the worse ear. People with hearing impairment were older than those without hearing impairment by an average of 13 years, more likely to be non-Hispanic white, and more likely to have less than a high school level of education. People with hearing impairment were also more likely to report having served in the military, experienced occupational noise exposure, and used ototoxic medications. The effects of military history and ototoxic medications were explained by the older age of participants with these characteristics. People with hearing impairment were no more likely than those without hearing impairment to report an income–poverty ratio of 1.0 or less, leisure-time noise exposure, or current smoking, although associations with income–poverty ratio and leisure-time noise exposure were observed when we corrected for age. Finally, people with hearing impairment were more likely to report diabetes, an effect not explained by age in preliminary analyses. All other characteristics were associated with diagnosed diabetes in preliminary analyses (data not shown), suggesting that they should be treated as likely confounders when assessing the potential relationship between hearing impairment and diabetes.

Table 3 shows unadjusted and age-adjusted prevalence estimates of hearing impairment in the United States, by diagnosed diabetes status. The unadjusted prevalence estimates for all 9 outcomes were statistically higher among individuals with diabetes than those without diabetes. Differences in prevalence were attenuated but remained statistically significant after adjustment for age.

The Figure shows age-adjusted and age-specific mean pure tone thresholds (averaged first within participants over both ears), by diagnosed diabetes status. Persons with diabetes had higher thresholds at all frequencies than persons without diabetes, and the difference seemed to widen at frequencies greater than 2000 Hz (Figure, A). Although these curves represent population averages, 2 individuals with these profiles would have clinically significant differences in hearing impairment. Age-specific analyses (Figure, B through F) demonstrate the consistency of higher pure tone thresholds across the entire frequency range and across all age groups for people with diabetes. The curve for people with diabetes age 20 to 29 years (Figure, B) should be interpreted with caution because it is based on only 10 people (most of whom probably have type 1 diabetes), and an examination of the age distribution within this age group suggests that age differs between those with diabetes and those without.

Table 4 shows the prevalence of low- or mid-frequency hearing impairment of mild or greater severity assessed in the worse ear in specific subgroups. The prevalence of hearing impairment among people with diagnosed diabetes statistically exceeded the prevalence among those without diabetes in all groups except persons age 60 to 69 years. Statistically significant differences by diabetes status remained after age adjustment within most subgroups. Appendix Tables 1 to 8 show the results for the other audiometric outcomes and self-reported hearing impairment. Findings were similar for all 4 high-frequency hearing impairment outcomes (Appendix Tables 1, 3, 5, and 7). The low prevalence of low- or mid-frequency hearing impairment of moderate or greater severity assessed in the worse ear (Appendix Table 2) or better ear (Appendix Table 6) resulted in insufficient statistical power to detect statistically significant differences in subgroup specific prevalence by diabetes status.

Table 5 shows the age-adjusted prevalence of high-frequency hearing impairment, by glycemic status (normal, impaired fasting glucose, or diabetes). The prevalence of hearing impairment was statistically higher for those with impaired fasting glucose than those with normal fasting glucose for 3 of the 4 outcomes and was statistically higher for all 4 outcomes among persons with diabetes than those with normal fasting glucose levels. There was no difference in prevalence between persons with diagnosed diabetes and those with undiagnosed diabetes.

In multivariable analyses, people with diabetes had statistically significant increased odds of hearing impairment in worse and better ears at all levels of severity and frequency (Table 6). Estimates were generally similar across frequencies except those for hearing impairment of moderate or greater severity assessed in the better ear, in which the odds ratio estimate of low- or mid-frequency hearing impairment was higher, and that of high-frequency hearing impairment was lower than all the others. Additional adjustment for hypertension and cardiovascular disease did not substantively change the odds ratio estimates (data not shown).

We evaluated the association between diabetes and audiometrically assessed hearing impairment in the U.S. noninstitutionalized population by using nationally representative data. We estimate a prevalence of low- or mid-frequency hearing impairment of mild or greater severity of 28.0% among people with diabetes. The prevalence of hearing impairment was higher among individuals with diabetes in both sexes; all groups of race or ethnicity, education, and income–poverty ratio; and all age groups but the oldest (those 60 to 69 years). The higher prevalence was not limited to possibly predisposed subgroups, such as those who smoke, those with occupational or leisure-time noise exposure, or those taking ototoxic medications. The association between diabetes and hearing impairment remained in analyses that adjusted for other factors that may contribute to impairment.

The strength of the association of diabetes with hearing impairment that we observed is similar to that in 2 previous population-based studies (20 - 21). We report an odds ratio of 1.82 (CI, 1.27 to 2.60) for low- or mid-frequency hearing impairment of mild or greater severity assessed in the worse ear, whereas Helzner and colleagues (21) and Dalton and coworkers (20) reported odds ratios of 1.41 (95% CI, 1.05 to 1.88) and 1.42 (CI, 1.10 to 1.83), respectively, although the outcome of the latter study was based on a pure tone average in the worse ear of greater than 25 dB HL over frequencies of 500, 1000, 2000, and 4000 Hz. For the purposes of comparison, we replicated the definition of hearing impairment used by Dalton and coworkers (20) and observed an odds ratio of 1.89 (CI, 1.27 to 2.81). Our definition of diabetes differed from that used by Dalton and coworkers, who included cases of undiagnosed diabetes and attempted to exclude individuals with type 1 diabetes. Our analysis focused on people reporting a diabetes diagnosis. Because of the self-reported nature of our assessment, we could not restrict our analyses to people with type 2 diabetes, although 90% to 95% of diabetes in our nationally representative sample of adults with diabetes was probably type 2 (9).

Differences in age composition might account for the modest differences in the strength of association among these population-based studies. The adults in Helzner and colleagues' study (21) were 73 to 84 years of age, and those studied by Dalton and coworkers (20) were between 43 and 84 years of age. The relative contribution of diabetes to hearing impairment may be stronger among our substantially younger sample (age 20 to 69 years) before the cumulative effects of aging, noise exposure, and other factors have made substantial contributions to hearing impairment. Our graphical analysis of mean pure tone thresholds suggested that the separation in pure tone thresholds by diabetes status was less in those age 60 to 69 years. In addition, the ratio of the age-specific prevalence estimates for older versus younger participants appears to be smaller (Table 4), suggesting that the relative contribution of diabetes may be less as one ages. Evidence from another relatively young sample of Japanese men in the military demonstrated an 87% increased odds of impairment (using the definition by Dalton and coworkers (20)) for those reporting diabetes, which is consistent with our findings (25).

Gates and colleagues (26) did not find a statistically significant difference in pure tone average (over 250, 500, and 1000 Hz or over 4000, 6000, and 8000 Hz) by diabetes status; however, the mean age of their study cohort was 73 years, and effects of diabetes may be less likely to be observed in a sample of this age. Ma and coworkers (27) examined mean pure tone thresholds at 500, 1000, 2000, and 4000 Hz by using data from the Hispanic Health and Nutrition Examination Survey and observed a higher mean threshold for Mexican-American adults with diabetes, but only at 500 Hz.

Diabetes-related hearing loss has been described as progressive, bilateral, sensorineural impairment with gradual onset predominantly affecting the higher frequencies (15). We observed generally stronger associations between diabetes and high-frequency hearing impairment than between diabetes and low- or mid-frequency hearing impairment. No consistently stronger associations were observed with hearing impairment assessed in the better ear (bilateral impairment) or when assessing greater severity. When we examined hearing thresholds at specific frequencies, thresholds at every frequency were higher for people with diabetes than for people without diabetes. This pattern held across all age groups. These observations are consistent with a report of higher hearing thresholds across all frequencies among patients with diabetes who are age 40 years or younger compared with healthy age-matched control participants, even though the thresholds in either group were not in the range to be considered hearing impaired (28).

Several biological mechanisms might explain an association between diabetes and hearing impairment. Well-established complications of diabetes, such as retinopathy, nephropathy, and peripheral neuropathy, involve pathogenic changes to the microvasculature and sensory nerves (14,29). These pathologic changes may include the capillaries and sensory neurons of the inner ear, but evidence from human studies is limited. Postmortem observations of diabetic patients include thickening of capillaries in the stria vascularis (13,30) and demyelination of the eighth cranial nerve, 1 branch of which transmits auditory signals from the cochlea to the brainstem (12). Pathologic changes specific to the cochlea also include thickened walls of the vessels of the basilar membrane and greater loss of outer hair cells in the lower basal turn (30). Compromised cochlear function has been measured by using evoked otoacoustic emissions (a noninvasive method to assess damage to the outer hair cells of the cochlea) among patients with diabetes relative to healthy control participants (31). Other vascular changes include narrowing of the internal auditory artery (32). Certain rare genetic syndromes, such as the Alström syndrome (33), the Wolfram syndrome (34), and maternally inherited diabetes and deafness (35), feature diabetes and hearing impairment as characteristics. It is possible that more common genetic factors predispose to both diabetes and hearing impairment, but these have not been identified yet.

Potential limitations of the analysis include recall-based assessments of leisure time and occupational noise exposure. Self-reported noise exposure is subject to measurement error, so we cannot rule out residual confounding as contributing to some of the association we observed. This limitation may be more of a factor for high-frequency hearing impairment because this outcome incorporates pure tone thresholds observed across 3000 to 6000 Hz, the frequencies at which injury from excessive noise stimulus is most notable (36). Similarly, for most of our analyses, the assessment of diabetes was based on self-report, and persons with undiagnosed diabetes were considered to be nondiabetic. Given the greater prevalence of hearing impairment that we observed with greater dysregulation of glucose, we have probably underestimated the overall measures of association. Also, we cannot distinguish between type 1 and type 2 diabetes; however, almost all participants in this study have type 2 diabetes. In addition, our measure of ototoxic drug use does not incorporate information on dose or use in the past. Finally, we can make inferences only about the U.S. noninstitutionalized adult population. The prevalence of both diabetes and hearing impairment is probably even greater among institutionalized adults.

In summary, our data suggest that hearing impairment may be an underrecognized complication of diabetes. Although this analysis does not focus on possible mechanisms for the association of diabetes and hearing impairment, we have identified an important public health problem that can be addressed. With the high prevalence of hearing impairment among diabetic patients, screening for this condition may be justified (37 - 39).

Pure tone audiometry is a method of measuring hearing sensitivity across a range of frequencies. For each frequency, a pure tone signal is presented to the ear and the intensity of the signal is varied until the level at which the participant is just able to perceive the tone is identified. This level is the pure tone threshold for a particular frequency. A higher threshold indicates that a more intense signal was needed to perceive the tone and signifies greater hearing impairment. Audiologists express the intensity level of thresholds on a decibels hearing level (dB HL) scale based on the common logarithm of the ratio of the signal's intensity to a reference intensity of 10⁻¹² watts/m²(22). Zero dB HL represents the threshold of hearing at each frequency for young adults. An average threshold of greater than 25 to 40 dB HL is considered to be mild impairment, greater than 40 to 55 dB HL is considered to be moderate impairment, greater than 55 to 70 dB HL is considered to be moderately severe impairment, and greater than 70 to 90 dB HL is considered to be severe impairment.

Hearing measurements in the study were collected by trained audiometric technicians by using a calibrated Interacoustics audiometer (model AD226, Interacoustics USA, Eden Prairie, Minnesota) while study participants were in a sound-treated booth that met the American National Standard Institute (ANSI) Maximum Permissible Ambient Noise Levels for Audiometric Test Rooms (ANSI S3.1-1991). The audiometer's calibration was confirmed daily by using a Quest Model BA-201-25 (Quest Technologies, Oconomowoc, Wisconsin) bioacoustic simulator to verify the stability of the audiometric signal over time. Audiometers met the specifications of ANSI S3.6-1996 for type 3 audiometers. Standard audiometric headphones were used unless a potential for collapsing ear canals was noted during otoscopic examination, in which case, insert earphones were used.