Modeling Hearing Loss Progression and Asymmetry in the Older Old: A National Population-Based Study

Abstract

Objective:

The progression and asymmetry of age-related hearing loss has not been well characterized in those 80 years of age and older because public datasets mask upper extremes of age to protect anonymity. We aim to characterize the progression, severity, and asymmetry of hearing loss in those 80 years of age and older using a representative, national database.

Methods:

Cross-sectional, multicentered US epidemiologic analysis using the National Health and Nutrition Examination Study (NHANES) 2005-2006, 2009-2010, and 2011-2012 cycles. Subjects included non-institutionalized, civilian adults 80 years and older (n=621). Federal security clearance was granted to access publicly-restricted age data. Outcome measures included pure-tone average air conduction thresholds and the 4-frequency pure tone average (PTA).

Results:

621 subjects were 80 years old or older (mean=84.2 years, range=80-104 years), representing 10,600,197 Americans. The average PTA was 38.9 dB (95% CI=37.8, 40.0). Hearing loss exhibited constant acceleration across the adult lifespan at a rate of 0.0052 dB/year² (95% CI = 0.0049, 0.0055). This model predicted mean PTA within 2 dB of accuracy for most ages between 20 and 100 years. From age 80 to approximately 100, the average PTA difference between the better and worse ear was 6.75 dB (95% CI=5.8, 7.1). This asymmetry was relatively constant (i.e., non-significant linear regression coefficient of asymmetry over age=0.07 [95% CI=−0.01, 0.2]).

Conclusions:

Hearing loss steadily and predictably accelerates across the adult lifespan to at least age 100, becoming near-universal. These population-level statistics will guide treatment and policy recommendations for hearing health in the older old.

Level of Evidence:

3

INTRODUCTION

Age related hearing loss, or presbycusis, is a major burden on the US and world population. Approximately two-thirds of the US population over the age of 70 has some level of hearing loss¹, which can have a significant impact on quality of life.² Recent studies illustrate an independent association of age-related hearing loss with morbid conditions, including cognitive impairment^3-6 , dementia^7-9, depressive symptoms¹⁰, social isolation¹¹, functional decline¹², falls¹³, and reduced temporal lobe volume.¹⁴ This has brought hearing loss to the forefront of public health discussions. For example, a recent review estimated that treating or preventing hearing loss may be associated with a 9.1% reduction in new dementia cases, representing the largest reduction among all known risk factors across the lifespan.¹⁵

Globally and in the United States, the aging population is increasing.¹⁶ This may further increase the prevalence of both hearing loss and associated morbidities. Previous epidemiologic studies have examined hearing loss progression, but group all ≥80 or ≥85 years olds into a single category. ^1,17-19 While certain studies have characterized the older-old, they have used convenience sampling in audiology practices, where hearing loss will be biased since those who seek audiologic care tend to have worse hearing.^20,21 There is no study that accurately characterizes the progression of hearing loss in participants aged 80 years and older in the US. This would require community-based sampling with audiometry. While data exists in the US, such as in the National Health and Nutrition Examination Study (NHANES), the true age of extremely old subjects is masked to protect anonymity. This makes it difficult to study the epidemiology of hearing loss in the older old. ²²

In this study, the true age of older old subjects was obtained in NHANES through a specialized security clearance process. This study expands on a prior brief report of basic prevalence statistics.²³ Herein, we use a cross-sectional analysis to model hearing loss progression over the lifespan with attention to the upper extremes of age. We also analyze sensorineural asymmetry. To our knowledge, this represents the first detailed characterization of US hearing loss epidemiology in the older old. This analysis will help inform the current national debate on hearing healthcare by studying a population both representative of the US population and most affected by hearing loss.

METHODS

Participants

Subjects were part of the National Health and Nutrition Examination Survey (NHANES), a biannual, cross-sectional epidemiological survey of a representative sample of non-institutionalized, civilian US individuals. Audiometric testing for individuals ≥80 years old was present in the 2005-06, 2009-10, and 2011-12 cycles (n = 621). Additional participants aged 20 years and greater were included to characterize trends over the adult lifespan (n = 4,844). Cycles were combined with a weighting calculation per NHANES protocols.²⁴

Demographic data included age and gender. Population numbers from the 2018 US census were used to estimate current numbers of people with hearing loss in the United States.²⁵

Data Access and Security

Due to the limited numbers of extremely old participants, the true age of those 80 (2009-2010, 2011-2012 cycles) or 85 (2005-2006 cycle) years and older in NHANES is restricted from public access to reduce the risk of identifying participants and breaching anonymity.²⁶ For example, in combination with other variables (e.g., rare disease status), a particular subject could be theoretically identified. The public dataset groups these participants into either an 80+ or 85+ age group.

To acquire the true age of participants in the restricted dataset, Special Sworn Status, which involves a months-long federal security clearance process, was obtained.²⁷ This security clearance, along with additional required training, involved government fingerprinting, background checks, 35 individual forms totaling 157 pages of paperwork, and three security-related online training modules that had to be completed twice. Subsequently, the scientific proposal, including proposed table shells and a data dictionary, went through four rounds of review with a CDC biostatistician.

All data analyses were performed under monitored surveillance at designated US Census Research Data Centers in accordance with US Census Bureau policies. Per regulation, true ages were binned to have at least 5 subjects per table cell in order to protect subject anonymity.

Audiometric Analysis

Identical audiometric testing protocols were used for all survey cycles. Air-conduction audiometry thresholds were recorded at 0.5, 1, 2, 4, 6, and 8 kHz in decibel hearing level (dB hearing level) for both ears. The four-frequency pure-tone average (PTA) was calculated by averaging thresholds at 0.5, 1, 2, and 4 kHz. Sample weighting, per NHANES protocols, was utilized to scale prevalence data to be representative of the non-institutionalized, civilian US population. Hearing was categorized as normal (PTA ≤ 25 dB), mild hearing loss (26-40 dB), moderate hearing loss (41-55 dB), and moderately-severe-or-worse hearing loss (≥56 dB).²⁸ More granular categories were not possible because they caused violation of the US Census Bureau requirement of having at least 5 subjects per table cell. NHANES does not include bone conduction audiometry, thus it was not possible to distinguish sensorineural from conductive hearing loss.

Statistical Analysis

Participants were divided into 5-year age brackets from 20-25 years to 95 years-and-older for analysis. Select analyses required a 90 years-and-older bracket due to the requirement to have at least 5 subjects per table cell. Analysis was performed in Stata (StataCorp, College Station, TX) on monitored US Census Bureau Federal Statistical Research Data Center computers (New Haven, CT and New York, NY). Univariable linear regression was used to model the rate of hearing loss over age. The accuracy of the model was assessed by running it in a different cohort, the Hispanic Community Health Study (methods previously described).^4,10 Statistical significance was defined at the α=0.05 level. Analyses were performed between June 2019 and December 2019.

RESULTS

Of 5,465 participants with available audiometry data in the included cycles, 621 were age 80 years or older, representing 10,600,197 Americans, and 4,844 were 20 to 79 years old, representing 253,715,691 Americans. Demographic data and hearing loss thresholds for participants 80 years and older are illustrated in Table 1.

Table 1.

Demographics and Mean Pure Tone Average, National Health and Nutrition Examination Survey (2005–2006, 2009–2010, and 2011–2012 Cycles)

	Age Brackets (Years)
Characteristic	80-84	85-89	90-941	95 and older1	Total
No. Subjects, Unweighted	393	157	57	14	621
No. Subjects, Weighted and Scaled to US Population	6,740,882	2,613,429	970,706	275,179	10,600,197
Male, No. (%)	2,620,979 (38.9)	1,051,607 (40.2)	354,558 (28.5)		4,027,144 (38.0)
Last Time Had Hearing Test, %:
Less than a year ago	17.1	20.1	29.5		18.4
1-4 years ago	25.3	27.7	30.8		26.1
5-9 years ago	10.0	9.0	11.3		9.8
>10 years ago	18.5	22.9	13.4		19.5
Never	29.0	20.3	15.0		26.1
Pure Tone Average of Better Ear (95% CI):	36.6 (35.3, 37.9)	40.9 (38.7, 43.1)	45.8 (42.1, 49.5)	50.9 (44.4, 57.4)	38.9 (37.8, 40.0)

¹Per US Census Bureau Research Data Center regulation, ages must be binned to have at least 5 subjects per table cell in order to protect subject anonymity.

Of these ≥80-year-old participants, the mean age was 84.2 years (SD=4.0, range=80-104). Approximately 38% were male. The average PTA in the better hearing ear was 38.9 dB (95% CI = 37.8, 40.0) for those 80 years and older.

Hearing thresholds were worse at each frequency with sequentially older age brackets (Figure 1 and eTable 1.) The largest worsening across age brackets was seen in the lower frequencies. The average decrease in hearing threshold was 0.96, 0.99, 0.91, 0.74, 0.51, 0.53 dB/year for 0.5, 1, 2, 4, 6, and 8 kHz frequency, respectfully.

Figure 1. Hearing Thresholds Versus Frequency by Age Brackets.

Hearing threshold was defined by the pure tone average (500, 1000, 2000, 4000 Hz) in the better hearing ear. Error bars represent 95% confidence intervals of the mean. Refer to eTable 1 for tabular values.

Pure tone averages were worse with each successive age bracket across the adult lifespan, which occurred in a non-linear fashion (Figure 2 and eTable 2). Specifically, the rate (velocity) of hearing loss increased from age 20 to approximately 100, illustrated in Figure 3, which is the slope of Figure 2. The slope of the rate, i.e., the acceleration, of hearing loss was fairly constant. The one exception was at 65-74 years, where there was a transient increase and then decrease in the acceleration of hearing loss. (Figure 3 and eTable 3).

Figure 2. Hearing Threshold Over the Adult Lifespan.

Hearing threshold was defined by the pure tone average (500, 1000, 2000, 4000 Hz) in the better hearing ear. Error bars represent 95% confidence intervals of the mean. Refer to eTable 2 for tabular values.

Figure 3. Rate of Hearing Threshold Decrease Over the Adult Lifespan.

This represents the slope of Figure 2. Hearing threshold was defined by the pure tone average (500, 1000, 2000, 4000 Hz) in the better hearing ear. Refer to eTable 3 for Tabular values.

To express the relationship between pure tone average and age with constant acceleration, linear regression with a quadratic hearing term was performed. This allows us to summarize the relationship between hearing age across the adult lifespan with a simple formula:

$$Puretoneaverageinbetterear=0.0052\times ag{e}^2$$

In other words, hearing loss accelerates at a rate of 0.0052 dB/year² (95% CI = 0.0049, 0.0055). The model precision was tested by comparing the calculated mean PTA against the observed mean PTA across the adult lifespan in both NHANES and in an unrelated cohort, HCHS. In most cases, it was within 2 dB of accuracy between ages 20 to 100 (NHANES) or 18 to 76 (HCHS). In all cases, it was within 5 dB of accuracy. Restricting the sample to only those aged 80 and above had no meaningful effect on the model coefficient.

The equation can be rearranged as:

$$Age=\sqrt{\frac{Puretoneaverageinbetterear}{0.0052}}$$

to calculate the age at which a mean pure tone average is reached. In doing this, a mean PTA of 70 dB is theoretically reached at age 116 years, 80 dB at 124 years, 90 dB at 131 years, 100 dB at 139 years, 110 dB at 145 years, and 120 dB at 152 years. This assumes that the biology of hearing loss does not dramatically change in centenarians, and that the constant acceleration of 0.0052 dB/year² continues.

For participants 80 years and older, the average PTA difference between the better and worse ear was 6.75 dB (95% CI = 5.8, 7.1). Through linear regression analysis, age over 80 was not significantly correlated with magnitude of asymmetry (coefficient = 0.07, 95% CI −0.10, 0.24). However, age was correlated with asymmetry when analyzing all participants 20 years and older, illustrating a rate of worsening asymmetry by 0.06 dB/year (coefficient = 0.06, 95% CI 0.05, 0.07, p < 0.01). (Figure 4 and eTable 4.)

Figure 4. Hearing Threshold Differences Between the Better and Worse Ear for Participants 80 Years and Older.

Hearing threshold was defined by the pure tone average (500, 1000, 2000, 4000 Hz) Error bars represent 95% confidence intervals of the mean. Refer to eTable 4 for tabular values.

DISCUSSION

This study is the first detailed characterization of hearing loss in individuals aged 80 years and older based on US population level data. While US hearing loss characteristics have been reported for adults 80 years and over as a group^1,19, true ages over 80 are typically restricted from public use, including by the US Census Bureau, in order to protect subject anonymity. This prevents a granular analysis of specific age categories or age-related trends within the older old. We overcame this limitation by obtaining access to true ages over 80 through a specialized federal security clearance process. This study builds upon a prior study that reported basic prevalence statistics.²³

The few studies that have specifically examined granular hearing loss characteristics in the older old have been limited to single sites with biased subject samples.^20,21 For example, one report noted up to 15 dB worse hearing thresholds in 80-100 year old’s than in our NHANES population.²⁰ The worse thresholds were mostly in the lower frequencies, likely because of a ceiling effect in severe high frequency loss due to near-maximal damage. However, this prior study used a convenience sample at a single tertiary care academic audiology practice. Single institution convenience samples have several critical limitations. First, they only represent one geographic area, and thus data are not generalizable to the geographic, ethnic, and socioeconomic diversity of the US population. For example, these subjects likely drew heavily from the predominately Caribbean Hispanic population around the medical center. Second, the data are biased since they come from subjects who had a significant enough problem to see an audiologist. This sampling bias would exaggerate the prevalence and severity of hearing loss. In contrast, the data from the present study is representative of the US non-institutional, civilian population, including both those who do and do not receive care for their hearing loss.

Hearing loss decreased in a non-linear fashion over the lifespan. Overall, the velocity of hearing loss increased steadily, meaning that there was constantly accelerating hearing loss. This extends prior findings of accelerating hearing loss in the older old²⁹ to the entire adult lifespan. The relationship between hearing and age is thus best expressed by linear regression using the square of age as a predictor. For every 1-year increase in age, the velocity (rate) of hearing loss increased 0.0052 dB/year. This translates into the simple equation of pure tone average = 0.0052 × age². While this number may seem too small to be meaningful, this is not the case as it represents the change in velocity (acceleration). Compounded over a lifetime, the velocity of hearing loss would increase five-fold, from 0.2 dB loss/year at age 20 to 1 dB loss/year at age 100.

This model was very accurate, with a calculated mean PTA usually within 2 dB of the observed mean PTA across the adult lifespan. In fact, restricting the model to only those age 80 and over resulted in no meaningful change.

We can anticipate that with a constant acceleration in hearing loss, a ceiling effect will be reached, in which all vulnerable hearing is lost at a certain age and no further worsening will be observed. However, this ceiling was not reached in our sample aged to approximately 100 years. If this trend continues, the mean age at which severe (70 dB) and profound (90 dB) hearing loss will be reached is 116 years and 132 years, respectively. One must use great caution, however, interpreting a model outside of the range in which it was constructed. If this constant acceleration continued in humans living well into their 100s, one would assume it would plateau once the severe or profound level (i.e., maximal hearing loss) were reached. It is also impossible to test our model at these ultra-extreme ages, given the exceedingly small number of individuals who live beyond 110 years.

Evidence of a plateau in the rate of hearing loss was seen when examining trends by individual pure tone frequencies. Between ages 80 to approximately 100 years, the rate of hearing loss at 8000 Hz was approximately half the rate of loss at 500 Hz. Biologically, this suggests that by age 80-100, near maximal damage has reached the basal hair cells, which are responsible for high frequency hearing. In contrast, the apical hair cells, which are responsible for low frequency hearing, are relatively healthy, thus allowing a greater velocity of ongoing injury.

The relentless progression of age-related hearing loss raises important questions about hearing healthcare should life expectancy continue to rise through the 21^st century. For example, many centenarians will have unaidable hearing and become cochlear implant candidates. One of the three pillars of the successful aging model is continued engagement with life.³⁰ This would be greatly challenging in a previously hearing adult who becomes deaf. In order to ensure standard-of-care hearing rehabilitation in the upper extremes of life, centenarians will need to be healthy enough to undergo cochlear implant surgery. For these elderly patients without major comorbidities, cochlear implant anesthesia risk has been shown be minimal risk.³¹

On average, participants 80 years and older had a 7 dB asymmetry between their better and worse ear (with 98% of participants having up to a 23 dB asymmetry). This asymmetry was relatively constant from age 80 to approximately 100. This corroborates a previous report in a tertiary care audiology practice that found a high prevalence of asymmetric hearing loss in patients age 95 and older²¹. In contrast, among subjects age 20 to 79, pure tone asymmetry significantly increased with age. This indicates a plateau effect of asymmetry in the older old. Because our data did not include bone conduction, we are unable to know whether these asymmetries are largely conductive or sensorineural. Assuming most are sensorineural, the naturally high prevalence of asymmetric hearing loss in the older old has implications in MRI-based screening for vestibular schwannoma, one of the few medically concerning causes of asymmetric sensorineural hearing loss.^21,32 Further studies could examine the proportion of those ≥80 years old with asymmetric hearing loss over a certain threshold, such as at 3000 Hz, which has been suggested to be predictive of vestibular schwannoma.³²

In our previous report of these data, hearing aids were under-utilized.²³ This is similar to reports in other cohorts.^2,33 Interestingly, the use of hearing tests was also very low. Most individuals age 80-84 reported having a hearing test 5 or more years ago, with nearly a third denying ever having had one. Among those aged 90 and over, nearly half reported having a hearing test 5 or more years ago.

The gross underutilization of hearing aids and hearing tests illustrates the insufficient attention given to age-related hearing loss in the American healthcare system. Strategies to improve the standard of care could include a greater emphasis on hearing health in the primary care setting as well as geriatric facilities.²⁰ Enabling regular hearing screens in the adult primary care setting, much like what is already done in pediatrics, could be an important first step towards awareness, diagnosis, and treatment. Just as primary care visits serve as an important screen for hypertension, so could they serve as a preliminary screen for age-related hearing loss.

This study has limitations. NHANES is a cross-sectional study and thus while we can make inferences based on the increasing ages of subjects, it is not a longitudinal study that truly reflects aging over time. In addition, NHANES only represents the US civilian, non-institutional population. Non-institutionalized adults over 80 will tend to be healthier than their institutionalized counterparts. As a result, our study may underestimate the severity of hearing loss in the older old. Additionally, we assume that all hearing loss is age-related and thus sensorineural. While age-related hearing loss comprises the great majority of hearing loss in adults, we cannot confirm this for all subjects without bone conduction audiometry, which is not performed in NHANES. In addition, specific factors can contribute to accelerated hearing loss in later life, including noise exposure, cardiovascular disease³⁴, diabetes³⁵, and race/ethnicity³⁶. Given the cross-sectional nature of this study, it is not possible to identify or control for individual causes of hearing loss. Finally, only 14 subjects were 95 years of age or older. The precision of estimates in this group is thus limited.

In conclusion, the rate of hearing loss constantly accelerates across the adult lifespan and does not slow down among those age 80 to approximately 100. Hearing testing use remains very low.

Supplementary Material

Supplementary Tables

eTable 1. Hearing Thresholds (dB) by age group, mean (95% CI). Hearing threshold was defined by the pure tone average (500, 1000, 2000, 4000 Hz) in the better hearing ear.

eTable 2. Hearing Thresholds Versus Frequency by Age Brackets. Hearing threshold was defined by the pure tone average (500, 1000, 2000, 4000 Hz) in the better hearing ear.

eTable 3. Rate of Hearing Threshold Decrease Over the Adult Lifespan. This represents the slope of Figure 2. Hearing threshold was defined by the pure tone average (500, 1000, 2000, 4000 Hz) in the better hearing ear.

eTable 4. Hearing Threshold Asymmetry for Participants 80 Years Old and Over. Hearing threshold was defined by the pure tone average (500, 1000, 2000, 4000 Hz) in the better hearing ear.