Population Attributable Fraction of Incident Dementia Associated With Hearing Loss

Emily Ishak; Emily A Burg; James Russell Pike; Pablo Martinez Amezcua; Kening Jiang; Danielle S Powell; Alison R Huang; Jonathan J Suen; Pamela L Lutsey; A Richey Sharrett; Josef Coresh; Nicholas S Reed; Jennifer A Deal; Jason R Smith

doi:10.1001/jamaoto.2025.0192

. 2025 Apr 17;151(6):568–575. doi: 10.1001/jamaoto.2025.0192

Population Attributable Fraction of Incident Dementia Associated With Hearing Loss

Emily Ishak ¹, Emily A Burg ², James Russell Pike ³, Pablo Martinez Amezcua ^4,⁵, Kening Jiang ^4,⁵, Danielle S Powell ⁶, Alison R Huang ^4,⁵, Jonathan J Suen ^5,⁷, Pamela L Lutsey ⁸, A Richey Sharrett ⁴, Josef Coresh ^3,⁹, Nicholas S Reed ^4,^5,¹⁰, Jennifer A Deal ^4,^5,¹⁰, Jason R Smith ^4,^5,^✉

¹Columbia University Irving Medical Center, New York, New York

²Vanderbilt Bill Wilkerson Center for Otolaryngology and Communication Sciences, Vanderbilt University Medical Center, Nashville, Tennessee

³Department of Population Health, New York University Grossman School of Medicine, New York

⁴Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland

⁵Cochlear Center for Hearing and Public Health, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland

⁶Department of Hearing and Speech Sciences, University of Maryland–College Park

⁷Johns Hopkins School of Nursing, Baltimore, Maryland

⁸Division of Epidemiology and Community Health, University of Minnesota, Minneapolis

⁹Department of Medicine, New York University Grossman School of Medicine, New York

¹⁰The Johns Hopkins Disability Health Research Center, Johns Hopkins University, Baltimore, Maryland

Accepted for Publication: January 28, 2025.

Published Online: April 17, 2025. doi:10.1001/jamaoto.2025.0192

Correction: This article was corrected on July 17, 2025, to fix an error in the Hearing Assessment section of the Methods, where it was incorrectly stated that self-reported hearing loss was ascertained using the Hearing Handicap Inventory for the Elderly. The correct source was a self-reported hearing and noise exposure form.

^✉

Corresponding Author: Jason R. Smith, ScM, Johns Hopkins Bloomberg School of Public Health, 2024 E Monument St, Ste B-1100, Baltimore, MD 21205 (jsmit491@jhu.edu).

Author Contributions: Mr Pike and Mr Smith had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Ms Ishak and Dr Burg contributed equally to this work; Dr Deal and Mr Smith contributed equally to this work.

Concept and design: Deal, Smith.

Acquisition, analysis, or interpretation of data: Burg, Ishak, Pike, Martinez-Amezcua, Jiang, Powell, Suen, Lutsey, Sharrett, Coresh, Reed, Deal, Smith.

Drafting of the manuscript: Burg, Ishak, Pike, Smith.

Critical review of the manuscript for important intellectual content: Burg, Ishak, Martinez-Amezcua, Jiang, Powell, Suen, Lutsey, Sharrett, Coresh, Reed, Deal, Smith.

Statistical analysis: Burg, Pike, Powell, Smith.

Obtained funding: Coresh.

Administrative, technical, or material support: Powell, Suen, Coresh.

Supervision: Deal.

Conflict of Interest Disclosures: Dr Suen reported grants from the National Institutes of Health (NIH) (National Institute on Aging [NIA] F31AG071353) and the Johns Hopkins Cochlear Center for Hearing and Public Health, which is supported in part by a philanthropic gift from Cochlear, Ltd. Dr Lutsey reported grants from NIH (K24HL159246) to her institution during the conduct of the study. Dr Reed reported grants from NIA/NIH K23AG065443 during the conduct of the study and being a member of the Scientific Advisory Board, Neosensory, from 2021 to 2023. Dr Deal reported grants from NIH/NIA (K01AG054693) during the conduct of the study. Dr Smith reported other from NIA Epidemiology and Biostatistics of Aging Pre-Doctoral Training Grant (NIA 5T32AG000247-27) and the Cochlear Center Epidemiology Scholarship for Sensory Loss in Aging during the conduct of the study. No other disclosures were reported.

Funding/Support: The Atherosclerosis Risk in Communities (ARIC) study has been funded in whole or in part with federal funds from the National Heart, Lung, and Blood Institute (NHLBI), NIH, Department of Health and Human Services (contract No. 75N92022D00001, 75N92022D00002, 75N92022D00003, 75N92022D00004, 75N92022D00005). The ARIC Neurocognitive Study is supported by the NIH, including the NHLBI, National Institute of Neurological Disorders and Stroke, NIA, and National Institute on Deafness and Other Communication Disorders (grant No. U01HL096812, U01HL096814, U01HL096899, U01HL096902, and U01HL096917).

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

Data Sharing Statement: See Supplement 2.

Additional Contributions: The authors thank the staff and participants of the ARIC study for their important contributions.

^✉

Corresponding author.

PMCID: PMC12006913 PMID: 40244612

Key Points

Question

What fraction of incident dementia is attributable to hearing loss in a community-based population of older adults?

Findings

In this prospective cohort study of 2946 participants, up to 32% of 8-year incident dementia could be attributable to audiometric hearing loss, and self-reported hearing loss was not associated with increased dementia risk. Population attributable fractions were larger in those 75 years and older, female individuals, and White adults.

Meaning

Treating hearing loss might delay dementia for a large number of older adults.

Abstract

Importance

Hearing loss treatment delays cognitive decline in high-risk older adults. The preventive potential of addressing hearing loss on incident dementia in a community-based population of older adults, and whether it varies by method of hearing loss measurement, is unknown.

Objective

To calculate the population attributable fraction of incident dementia associated with hearing loss in older adults and to investigate differences by age, sex, self-reported race, and method of hearing loss measurement.

Design, Setting, and Participants

This prospective cohort study was part of the Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS) and had up to 8 years of follow-up (2011-2019). The 4 ARIC field centers in the study included Jackson, Mississippi; Forsyth County, North Carolina; the Minneapolis suburbs, Minnesota; and Washington County, Maryland. Community-dwelling older adults aged 66 to 90 years without dementia at baseline who underwent a hearing assessment at ARIC-NCS visit 6 (2016-2017) were included in the analysis. Data analysis took place between June 2022 and July 2024.

Exposures

Hearing loss measured objectively (audiometric) and subjectively (self-reported).

Main Outcomes and Measures

The main outcome was incident dementia (standardized algorithmic diagnosis with expert panel review). The population attributable fractions of dementia from both audiometric and self-reported hearing loss were calculated in the same participants, which quantified the maximum proportion of dementia risk in the population that can be attributed to hearing loss.

Results

Among 2946 participants (mean [SD] age, 74.9 [4.6] years; 1751 [59.4] female; 637 Black [21.6%] and 2309 White [78.4%] individuals), 1947 participants (66.1%) had audiometric hearing loss, and 1097 (37.2%) had self-reported hearing loss. The population attributable fraction of dementia from any audiometric hearing loss was 32.0% (95% CI, 11.0%-46.5%). Population attributable fractions were similar by hearing loss severity (mild HL: 16.2% [95% CI, 4.2%-24.2%]; moderate or greater HL: 16.6% [95% CI, 3.9%-24.3%]). Self-reported hearing loss was not associated with an increased risk for dementia, so the population attributable fraction was not quantifiable. Population attributable fractions from audiometric hearing loss were larger among those who were 75 years and older (30.5% [95% CI, −5.8% to 53.1%]), female (30.8% [95% CI, 5.9%-47.1%]), and White (27.8% [95% CI, −6.0% to 49.8%]), relative to those who were younger than 75 years, male, and Black.

Conclusions and Relevance

This cohort study suggests that treating hearing loss might delay dementia for a large number of older adults. Public health interventions targeting clinically significant audiometric hearing loss might have broad benefits for dementia prevention. Future research quantifying population attributable fractions should carefully consider which measures are used to define hearing loss, as self-reporting may underestimate hearing-associated dementia risk.

This cohort study calculates the fraction of a large older adult population for whom incident dementia may have been associated with hearing loss.

Introduction

The number of people living with dementia is expected to triple worldwide over the next few decades.¹ Prevention strategies intervening on modifiable risk factors need to be tailored to populations with different risk factor profiles. Systematic reviews and meta-analyses provide converging evidence that hearing loss (HL), which is treatable and affects more than two-thirds of older adults in the US,^2,3,4 is a putative risk factor for dementia.^1,5

Recently, the ACHIEVE (Aging and Cognitive Health Evaluation in Elders) randomized clinical trial tested the effect of a hearing intervention vs health education control on cognitive decline over 3 years in cognitively healthy older adults with HL.⁶ Of the ACHIEVE participants, 24% were also participants in the Atherosclerosis Risk in Communities (ARIC) study; 76% were newly recruited from the community. In a prespecified sensitivity analysis stratifying by recruitment cohort, treating HL reduced the rate of 3-year cognitive decline by 48% among ARIC participants compared to controls. ARIC participants were followed up for up to 30 years at the time of ACHIEVE recruitment and included a greater proportion of individuals with risk factors for dementia (eg, lower educational attainment, lower baseline cognitive scores, higher proportion with diabetes).⁶ The findings provide evidence that treating HL could reduce dementia risk in some older adults. However, it is unclear what fraction of dementia risk could be prevented by treating HL over a longer follow-up period in a population of older adults.

Accounting for both the relative risk and prevalence of HL, the population attributable fraction (PAF) is the maximum proportion of dementia that can be attributed to HL in a population of both exposed and unexposed individuals. Prior PAF estimates of dementia from HL are complicated by substantial variation (2%-19% in the US).^7,8,9,10 Due in part to how HL prevalence was ascertained in prior studies, those PAFs could be underestimated. Some studies obtained HL prevalence using self-report measures,^8,9 which underestimates clinically significant HL in older adults.¹¹ No study has compared the impact of using self-report (a subjective measure) vs a clinical reference standard HL measure (pure tone audiometry) on PAFs for dementia. Importantly, no estimates currently reflect the preventive potential of reducing clinically significant HL on incident dementia in a community-based population of older adults at high dementia risk.

We quantified PAFs of 8-year incident dementia from HL within ARIC, a large, community-based cohort of older adults. This is the subsample that showed benefit from HL treatment in the ACHIEVE study. We compared objective (audiometry) vs subjective (self-report) measures of HL. We also estimate PAFs by age, sex, and self-identified race.

Methods

Study Design and Population

ARIC is a longitudinal cohort study of 15 792 adults aged 45 to 64 years from 4 US communities (Washington County, Maryland; Forsyth County, North Carolina; Jackson, Mississippi; and suburbs of Minneapolis, Minnesota) initiated from 1987 to 1989 and using up to 8 years of follow-up (2011-2019).¹² Written informed consent was obtained from all participants, or a proxy with participant assent if unable to provide consent (eg, due to dementia), at each visit with institutional review board approval from each ARIC study field center. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline was followed.

This analysis uses follow-up data from ARIC visit 5 (2011-2013) through visit 7 (2018-2019). We used hearing data collected from all participants at visit 6 (2016-2017). Given the clinically gradual onset and progression of age-related HL, we expected minimal clinically significant change in hearing thresholds between visit 5 (2011-2013) and visit 6 (2016-2017). This was confirmed given the average change in hearing level in a subsample of ARIC participants from visit 5 (the visit during which audiometric testing was piloted in a subsample of participants) to 6 was only 5 dB HL.¹³ We, therefore, set baseline for this analysis to visit 5 to maximize follow-up.

We included all participants without dementia at baseline and excluded those who did not self-identify as Black or White race and those who did not identify as White from Washington County and Minneapolis communities due to small numbers that limit representation. We excluded participants missing hearing and covariate data (eMethods in Supplement 1). Sensitivity analyses evaluated the impact of missing data on primary inference models.

Incident Dementia

Incident all-cause dementia is identified using a standardized algorithm¹⁴ that incorporates the following:

Longitudinal neuropsychological battery and cognitive data for participants attending clinic visits 5 to 7;
Supplemental cognitive data, including modified Clinical Dementia Rating Scale results (visit 5), 6-item Screener or Ascertain Dementia 8-Item Informant Questionnaire (visits 5-7), and Modified Telephone Interview for Cognitive Status (TICS-M; collected around date of visit 5) conducted outside of clinic visits; and
Hospitalizations and death certificate codes (International Classification of Diseases, Ninth Revision [ICD-9] and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision [ICD-10]) codes,^15,16 with onset date estimated as 6 months prehospitalization).¹⁷

Following the Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition) criteria,¹⁸ dementia diagnoses were corroborated by expert review.

Hearing Assessment

Pure tone air-conduction audiometry was collected at visit 6 (2016-2017). Audiometric thresholds were measured at octave frequencies from 250 Hz to 8000 Hz in decibels of hearing level (dB HL). Participants attending clinic visits were tested in a sound-treated booth using an audiometer (Interacoustics AD-629) and insert or circumaural headphones. Participants who completed home visits or resided in a long-term care facility were tested in a quiet room with ambient noise levels low enough for valid testing using a portable audiometer and circumaural headphones (SHOEBOX) that comply with American National Standards Institute standards. Prior research has demonstrated the validity of collecting hearing thresholds among individuals with dementia.¹⁹ A speech-frequency pure tone average was calculated for each participant as the mean of thresholds at 0.5, 1, 2, and 4 kHz in the better-hearing ear. Using clinically accepted classifications,²⁰ we defined audiometric HL categories: normal hearing, less than 26 dB HL; mild HL, 26 to 40 dB HL; moderate or greater HL, more than 40 dB HL; and any HL (≥26 dB HL). Hearing aid use (yes/no) in either ear during the month prior to the hearing assessment was self-reported. Self-reported HL (no: excellent or good hearing vs yes: a little trouble hearing, moderate trouble hearing, a lot of trouble hearing, or deaf) was ascertained via a hearing and noise exposure questionnaire.

Covariates

We included covariates chosen a priori based on their clinical significance. Baseline demographic covariates included age (years), self-identified sex (male, female), self-identified race (Black, White), and educational attainment (less than high school, high school or equivalent, more than high school). Other covariates included apolipoprotein E (APOE) ε4 carrier status (0 vs ≥1 alleles) ascertained using the TaqMan assay (Thermo Fisher Scientific), hypertension (systolic blood pressure of ≥140 mm Hg, diastolic blood pressure of ≥90 mm Hg, or use of a hypertensive medication), diabetes (fasting blood glucose level of ≥126 mg/dL, nonfasting blood glucose level of ≥200 mg/dL, self-reported physician’s diagnosis of diabetes, or use of diabetes medication), self-reported smoking status (never, former, current), body mass index (calculated as weight in kilograms divided by height in meters squared), and history of stroke collected using standardized methods²¹ (self-reported physician diagnosis or adjudicated stroke).

Statistical Analysis

We describe demographic and clinical characteristics of the overall study sample. We used Cox proportional hazards regression models to estimate hazard ratios (HRs) and 95% CIs of incident dementia by audiometric and self-reported HL. We verified the proportional hazards assumption by inspecting Schoenfeld residuals. We adjusted the primary inference models for age, sex, race, education, APOE ε4 carrier status, hypertension, diabetes, body mass index, smoking status, stroke history, and hearing aid use. We adjusted for hearing aid use with an interaction term between HL and hearing aid use, without a main effect term for hearing aid use (as hearing aids are only clinically beneficial for those with HL, and the relationship between HL and dementia likely varies by hearing aid use).

We computed PAFs of incident dementia from HL using the following formula adapted from Miettinen²²:

PAF = pd_i([HR_i − 1]/HR_i)

Where pd is the person-time prevalence (calculated as the sum of follow-up in those with the exposure divided by follow-up in the total sample) of the ith level of HL among dementia cases.²³ This formula that accounts for time exposed to HL is valid with the use of covariate-adjusted instantaneous relative risks.^23,24 Because we aimed to quantify the independent PAF from HL alone, we did not use the approach taken by the Lancet Commission¹ that was adapted from prior work.²⁵ Recent studies demonstrate this method can underestimate PAFs.^26,27 We also computed PAFs by age, sex, and self-identified race. Because PAFs should be estimated across sociodemographic populations, and these were specified a priori, stratified analyses are not adjusted for multiplicity.²⁸ We estimated nonparametric 95% CIs using 15 000 bootstrapped samples and the bias correction and acceleration method.²⁹

To evaluate the association of hearing aid use with dementia among those with HL (ie, those that would benefit from using a hearing aid), in secondary analyses, we quantified PAFs of dementia from non–hearing aid use among those with HL. To interpret the PAFs as the proportion of dementia attributable to non–hearing aid use among those with HL, we set the reference group for this analysis as wearing a hearing aid.

In sensitivity analyses, we used an alternative model-based approach to compute a time-varying PAF (eMethods in Supplement 1).³⁰ We examined the impact of missing data on the primary inference model by imputing missing hearing and covariate values with multiple imputation using chained equations. (eMethods in Supplement 1).³¹ As an alternative to standard adjustment for confounding via regression modeling, we used inverse probability of treatment weights for doubly robust confounding adjustment (eMethods in Supplement 1). We repeated the primary analysis using recent World Health Organization thresholds for HL (normal hearing, <20 dB; mild, 20-34 dB HL; moderate or greater, ≥35 dB HL).³²

We performed all analyses using R statistical software, version 4.2.3 (R Project for Statistical Computing), using the survival package for estimating Cox regression models and the boot package to obtain bootstrapped CIs. Statistical analysis took place between June 2022 and July 2024.

Results

Participant Selection

A total of 4003 ARIC Neurocognitive Study (ARIC-NCS) participants at visit 6 (2016-2017) were eligible. Participants with dementia at baseline (n = 74) were excluded, and participants who did not self-identify as Black or White race and those who did not identify as White race from Washington County and Minneapolis communities (n = 22) were excluded due to small numbers that would limit representation. A total of 342 participants with missing hearing data were excluded, and 684 with missing covariate data were excluded (eMethods in Supplement 1). The final analytic sample consisted of 2946 individuals.

Participant Characteristics

Among 2946 participants (mean [SD] age, 74.9 [4.6] years; 1751 [59.4] female; 637 Black [21.6%] and 2309 White [78.4%] individuals), 1947 participants (66.1%) had clinically significant (audiometric) HL (mild HL: 1151 participants [39%]; moderate or greater HL: 796 participants [27%]), and 1097 participants (37.2%) self-reported HL (Table 1). Of those with objectively measured moderate or greater HL, 445 (55.9%) used hearing aids, which accounted for the majority of hearing aid use in this sample. Relative to participants with normal hearing or mild HL, on average, those with moderate or greater HL were slightly older and had lower educational attainment. Among those with audiometric HL, a higher proportion of those who did vs did not self-report HL were female and Black race. The median (IQR) follow-up time was 6.6 (6.0-6.9) years, during which 239 participants (8%) developed incident dementia.

Table 1. Baseline Participant Characteristics by Hearing Loss Status, ARIC-NCS (2011-2013).

Characteristic	No. (%)
	Overall (N = 2946)	Normal hearing (n = 999)	Hearing loss
	Overall (N = 2946)	Normal hearing (n = 999)	Mild (n = 1151)^a	Moderate or greater (n = 796)^b	Any (n = 1947)^c
Better-ear PTA, mean (SD), dB	32.8 (13.6)	18.8 (4.6)	32.5 (4.3)	50.7 (8.3)	40.0 (10.9)
Self-reported hearing loss	1097 (37.2)	102 (10.2)	365 (31.7)	630 (79.1)	995 (51.1)
Hearing aid use	601 (20.4)	14 (1.4)	142 (12.3)	445 (55.9)	587 (30.1)
Age, mean (SD), y	74.9 (4.6)	73.2 (3.7)	74.8 (4.5)	77.0 (4.9)	75.7 (4.8)
Sex
Female	1751 (59.4)	724 (72.5)	670 (58.2)	357 (44.8)	1027 (52.7)
Male	1195 (40.6)	275 (27.5)	481 (41.8)	439 (55.2)	920 (47.3)
Race^d
Black	637 (21.6)	326 (32.6)	223 (19.4)	88 (11.1)	311 (16.0)
White	2309 (78.4)	673 (67.3)	928 (80.6)	708 (88.9)	1636 (84.0)
Education
<High school	317 (10.8)	101 (10.1)	117 (10.2)	99 (12.4)	216 (11.1)
High school	1222 (41.5)	383 (38.3)	473 (41.1)	366 (46.0)	839 (43.1)
>High school	1407 (47.8)	515 (51.6)	561 (48.7)	331 (41.6)	892 (45.8)
APOE ε4 carrier	829 (28.1)	302 (30.2)	320 (27.8)	207 (26.0)	527 (27.1)
Hypertension	2130 (72.3)	719 (72.0)	818 (71.1)	593 (74.5)	1411 (72.5)
Diabetes	857 (29.1)	297 (29.7)	333 (28.9)	227 (28.5)	560 (28.8)
Smoking status
Never	1235 (41.9)	452 (45.2)	456 (39.6)	327 (41.1)	783 (40.2)
Former	1549 (52.6)	491 (49.1)	624 (54.2)	434 (54.5)	1058 (54.3)
Current	162 (5.5)	56 (5.6)	71 (6.2)	35 (4.4)	106 (5.4)
BMI,^e mean (SD)	28.9 (5.5)	28.9 (5.4)	29.1 (5.8)	28.5 (5.0)	28.8 (5.5)
History of stroke	79 (2.7)	22 (2.2)	28 (2.4)	29 (3.6)	57 (2.9)
Follow-up, median (IQR), y	6.6 (6.0-6.9)	6.6 (6.0-6.9)	6.5 (6.0-7.0)	6.5 (5.9-6.9)	6.5 (5.9-7.0)

Open in a new tab

Abbreviations: APOE, apolipoprotein E; ARIC-NCS, Atherosclerosis Risk in Communities Neurocognitive Study; BMI, body mass index; PTA, pure tone average.

^{^a}

Defined as 26 dB to 40 dB.

^{^b}

Defined as more than 40 dB.

^{^c}

Defined as 26 dB or more.

^{^d}

ARIC-NCS participants who did not self-identify as Black or White race and those who did not identify as White race from Washington County, Maryland, and Minneapolis, Minnesota, communities were excluded due to small numbers that would limit representation (n = 22).

^{^e}

Calculated as weight in kilograms divided by height in meters squared.

Among 1097 participants that self-reported HL, 995 (90.7%) had audiometric HL (mild HL: 365 participants [33.3%]; moderate or greater HL: 630 participants [57.4%]) (eTable 1 in Supplement 1). Because most participants self-reporting HL had moderate or greater audiometric HL, the mean (SD) pure tone average in this group was 42.9 (13.0) dB.

PAF of Incident Dementia From Hearing Loss

The PAF of 8-year incident dementia from any degree of audiometric HL (mild or greater HL) was 32.0% (95% CI, 11.0%-46.5%) (Table 2). PAFs from mild (16.2% [95% CI, 4.2%-24.2%]) and moderate or greater (16.6% [95% CI, 3.9%-24.3%]) HL were similar. Self-reported HL was not associated with an increased risk for dementia; the HR was less than 1, so a PAF was not quantifiable.

Table 2. Population Attributable Fraction of 8-Year Incident Dementia Associated With Hearing Loss, ARIC-NCS (2011-2019).

Hearing loss variable	No. of dementia cases/total exposed	Risk factor prevalence (95% CI), %^a	HR (95% CI) for dementia^b	PAF of dementia (95% CI), %^c
Audiometric hearing loss
Normal	47/999	20.4 (18.3-22.8)	1 [Reference]	NA
Mild^d	102/1151	41.8 (39.1-44.6)	1.63 (1.13-2.35)	16.2 (4.2-24.2)
Moderate or greater^e	90/796	37.7 (35.1-40.5)	1.78 (1.16-2.74)	16.6 (3.9-24.3)
Any^f	192/1947	79.6 (77.2-81.7)	1.67 (1.18-2.38)	32.0 (11.0-46.5)
Self-reported hearing loss
No	146/1849	59.8 (57.0-62.5)	1 [Reference]	NA
Yes	93/1097	40.2 (37.5-43.0)	0.88 (0.63-1.22)	NA^f

Open in a new tab

Abbreviations: ARIC-NCS, Atherosclerosis Risk in Communities Study Neurocognitive study; HR, hazard ratio; PAF, population attributable fraction; NA, not applicable.

^{^a}

Conditional person-time prevalence of hearing loss category in dementia cases.

^{^b}

Cox regression models adjusted for age, sex, race, education, apolipoprotein E4 carrier status, hypertension, diabetes, smoking status, body mass index, prevalent stroke, and interaction term between hearing loss and hearing aid use (without a main effect term for hearing aid use).

^{^c}

PAF = conditional person time prevalence × ([HR − 1]/HR)

^{^d}

Defined as 26-40 dB hearing loss.

^{^e}

Defined as >40 dB hearing loss.

^{^f}

No dementia cases (0%) were attributable to self-reported hearing loss due to an HR of less than 1.

PAFs indicated a greater proportion of dementia was attributed to any HL measured in those 75 years and older (30.5% [95% CI, −5.8% to 53.1%]) compared to those younger than 75 years (22.0% [95% CI, −7.8% to 39.8%]); this was due to larger PAFs from moderate or greater HL in those 75 years and older (Figure 1; eTable 2 in Supplement 1). The PAF for any HL was slightly higher in female (30.8% [95% CI, 5.9%-47.1%]) relative to male (24.0% [95% CI, −24.1% to 52.4%]) participants, also driven by the difference in the PAF among those with moderate or greater HL (Figure 2; eTable 3 in Supplement 1). PAFs from any HL were also slightly larger among White (27.8% [95% CI, −6.0% to 49.8%]) relative to Black (22.9% [95% CI, −11.7% to 41.1%]) participants (Figure 3; eTable 4 in Supplement 1). Mild HL was associated with similar PAFs across populations; however, moderate or greater HL was associated with larger PAFs among White participants (14.3% [95% CI, −6.3% to 27.2%]) compared to Black participants (6.6% [95% CI, −12.3% to 14.5%]).

Figure 1. — A, The conditional person-time prevalence of each audiometric hearing loss category among those with incident dementia is shown. B, The corresponding PAF of incident dementia from audiometric hearing loss is shown. Error bars indicate bias-corrected nonparametric 95% CIs.

Figure 2. — A, The conditional person-time prevalence of each audiometric hearing loss category among those with incident dementia is shown. B, The corresponding PAF of incident dementia from audiometric hearing loss is shown. Error bars indicate bias-corrected nonparametric 95% CIs.

Figure 3. — A, The conditional person-time prevalence of each audiometric hearing loss category among those with incident dementia is shown. B, The corresponding PAF of incident dementia from audiometric hearing loss is shown. Error bars indicate nonparametric 95% CIs.

PAFs from non–hearing aid use among those with HL use were more modest (eTable 5 in Supplement 1). For example, among those with any HL, the PAF from no hearing aid use was 12.9% (95% CI, −13.4% to 31.8%).

Sensitivity Analysis

PAFs from any HL computed using the model-based approach were similar in magnitude to the primary analysis (26.8%-32.0%; eTable 6 in Supplement 1). Associations for any HL after imputing missing data (HR, 1.60 [95% CI, 1.18-2.16]), and after adjustment for confounding with inverse probability of treatment weights (HR, 1.56 [95% CI, 1.02-2.38]), were similar in magnitude to the primary analysis (HR, 1.67 [95% CI, 1.18-2.38]) (eTables 7-9 in Supplement 1). PAFs from HL (eg, PAF from any HL, 38.8% [95% CI, 3.8%-61.2%]) defined using current World Health Organization criteria were larger than the primary analysis (eTable 10 in Supplement 1).

Discussion

In this cohort study of a large, community-based sample of Black and White adults 65 years and older in the US, up to 32% of incident dementia over 8 years could be attributed to clinically significant audiometric HL through calculating the population attributable fraction. Self-reported HL underestimated clinically significant HL and did not capture the preventive potential of addressing HL on dementia risk. Relative to those younger than 75 years, male, and Black populations, PAFs were generally larger among those 75 years and older, female, and White populations.

The PAF from any HL (32%) is substantially larger than previous US estimates (2%-19%).^7,8,9,10 There are several potential explanations. First, using self-report measures^8,9 results in substantially lower prevalence estimates than what would be expected for older adults in the US.³ Second, some studies^8,9 used a PAF formula that is conservatively biased if there is positive confounding, which would lower estimates.^33,34 And third, mild HL defined using pure tone audiometry was weakly associated with increased prevalent dementia in prior work (PAF, 4%),^10,35 meaning it contributed less to a PAF from any degree of HL relative to this study. Our estimate is also higher than that reported in the 2020 Lancet Commission on Dementia Prevention (8% for objectively measured HL),¹ which accounted for nonindependence between several risk factors.²⁵ Prior work indicates this approach can underestimate PAFs, for example, when risk factors cluster together and contribute additively to dementia risk instead of multiplicatively.^26,27 This ignores potential synergistic interactions, which could be untenable for risk factors that cluster and act on dementia through similar pathways (eg, hearing and vision loss; hypertension and diabetes) and should be investigated in future research to estimate the summary contribution of multiple sensory risk factors to risk of dementia. The Commission’s unadjusted PAF (22%) is smaller, owing to a lower estimated prevalence of HL.

We found that PAFs from HL were generally larger for older, female, and White populations relative to younger, male, and Black populations. These differences were primarily related to the higher prevalence of HL in these populations, aside from sex-specific estimates, which accords with prior nationally representative estimates of HL prevalence in the US.² Although these data suggest a greater level of attribution in these populations, estimates should be interpreted cautiously due to potentially limited sample size and power in estimating the associated relative risks for the smaller subgroups.

Our study extends the literature by directly comparing the impact of using objective and subjective measures of HL when quantifying PAFs of dementia. Self-reported HL underestimated clinically significant HL prevalence in this sample. Misclassification using self-report HL is differential by age and increases substantially in older adults (20% misclassified in midlife and at least 70% in those 70 years and older)¹¹ who are at highest risk for dementia, which has important implications for researchers and policymakers. Although self-report is easier to capture on a larger scale, this likely measures a construct of hearing distinct from peripheral hearing acuity (eg, a perceived or functional impact of HL). Future work quantifying PAFs in older adults should prioritize objective measures of HL for estimating HL prevalence. Furthermore, misclassification of HL by self-report will lower relative risk estimates of dementia due to HL being incorrectly classified as normal hearing. Our findings demonstrate how dramatically this misclassification can affect inference, as self-reported HL showed no association with dementia risk in our sample. In future PAF studies, if relative risks for HL are obtained from systematic reviews or meta-analyses, priority should be given to studies that measured HL objectively.

Given the presumed underlying causal relationship between HL and dementia (ie, peripheral HL increasing cognitive load, inducing structural and functional brain changes, and decreasing social engagement),^36,37,38 our findings reinforce the importance of investigating the preventive potential of HL intervention to prevent dementia. In the ACHIEVE randomized clinical trial participants recruited from ARIC, hearing intervention vs control decreased the rate of 3-year cognitive decline by 48%; no effect on risk of incident cognitive impairment was observed, although the trial was not powered for this outcome.⁶ Our analysis showed only a modest association between hearing aid use and decreased risk of dementia, which could be indicative of a longer follow-up needed to detect a benefit to overt cognitive impairment. However, a recent meta-analysis found that hearing aid use in adults with audiometric HL was associated with a 29% decrease in risk for any cognitive decline relative to those not using hearing aids.³⁹

These findings underscore recommendations that efforts to delay and prevent dementia should continue in late life.¹ Relative to other potentially modifiable risk factors earlier in life (PAFs: midlife obesity, 18%; physical inactivity, 12%; low education, 12%),⁸ the preventive potential from addressing HL in late life in the US could be sizeable. The ability to control important upstream risk factors for HL (noise exposure, infection, ototoxic medications), as well as the capacity for hearing treatment and rehabilitation through technological interventions, such as hearing aids and cochlear implants, provides a compelling public health message about the importance of accessing affordable and acceptable hearing health care for maintaining auditory health across the course of life.

Strengths and Limitations

Our study has several strengths. In contrast to prior work,^8,9 this is the first study to provide data on longitudinal PAFs of incident dementia from HL for a large-scale community-based population of older adults with 8 years of follow-up. Second, ARIC-NCS has both objective and subjective measures of hearing, enabling direct comparisons. Third, dementia ascertainment in ARIC-NCS is comprehensive,¹⁷ with nearly complete ascertainment due to the combination of data sources used.¹⁴ And fourth, our sample was derived from the cohort in which the ACHIEVE trial demonstrated efficacy of an HL intervention on cognitive decline.

We also recognize several limitations to our study. First, because this is a community-based cohort of self-identified Black and White adults, there could be limited generalizability of prevalence estimates to the wider US population. However, our prevalence estimates are consistent with prior nationally representative data.^2,4 Second, classification of dementia using hospitalizations and death codes might not be sensitive in capturing those with milder dementia. Nevertheless, prior work has shown minimal variation in estimates with this limitation.⁴⁰ Additionally, community surveillance minimizes dropout potentially associated with more advanced dementia. Third, there could be potential survivor bias, as those with HL at visit 5 could have been more likely to develop dementia and not return for the hearing assessment at visit 6. Fourth, although by design, ARIC oversampled adults that self-identified as Black for greater representation, our study likely lacks power for analyses stratified by race. Similarly, we excluded the small number of individuals who self-identified as neither Black nor White. Because dementia risk profiles vary across racial and ethnic populations, work quantifying PAFs in different and diverse populations is needed to generate an evidence base to confront health inequities.⁴¹ Fifth, among participants using hearing aids, we lack information on appropriate hearing aid use, fit, and compliance. Finally, we were unable to account for potential cumulative effects of HL on risk of dementia.

Conclusions

In a large cohort of community-dwelling older adults with a mean age of 75 years, nearly 1 in 3 incident dementia cases could be attributed to clinically significant HL. Self-report substantially underestimated HL prevalence and was not associated with any dementia incidence. Interventions for sensory health in late life might be associated with a broad benefit for cognitive health. Future studies should prioritize objective measures of hearing over subjective measures to quantify its preventative potential on dementia risk, especially among groups known to underestimate their audiometric HL with self-report.

Supplement 1.

eMethods

eTable 1. Baseline characteristics by self-reported hearing loss, ARIC-NCS (2011-2019)

eTable 2. Impact of hearing loss on incident dementia, stratified by age, ARIC-NCS (20112019)

eTable 3. Impact of hearing loss on incident dementia, stratified by sex, ARIC-NCS (2011-2019)

eTable 4. Impact of hearing loss on incident dementia, stratified by self-identified race, ARIC-NCS (2011-2019)

eTable 5. Impact of no hearing aid use on incident dementia among those with hearing loss, ARIC-NCS (2011-2019)

eTable 6. Alternative model-based estimation of population attributable fraction of incident dementia from hearing loss, ARIC-NCS (2011-2019)

eTable 7. Comparison of characteristics of participants included versus excluded based on missing exposure and covariate data, ARIC-NCS (2011-2013)

eTable 8. Association of hearing loss with incident dementia after imputing missing exposure and covariate data, ARIC-NCS (2011-2019)

eTable 9. Association of hearing loss with incident dementia after adjustment with inverse probability of treatment weights, ARIC-NCS (2011-2019)

eTable 10. Impact of hearing loss on incident dementia, using 2020 World Health Organization hearing loss threshold, ARIC-NCS (2011-2020)

eReferences

jamaotolaryngolheadnecksurg-e250192-s001.pdf^{(348KB, pdf)}

Supplement 2.

Data sharing statement

jamaotolaryngolheadnecksurg-e250192-s002.pdf^{(17.4KB, pdf)}

References

1.Livingston G, Huntley J, Sommerlad A, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. Lancet. 2020;396(10248):413-446. doi: 10.1016/S0140-6736(20)30367-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Lin FR, Niparko JK, Ferrucci L. Hearing loss prevalence in the United States. Arch Intern Med. 2011;171(20):1851-1852. doi: 10.1001/archinternmed.2011.506 [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Lin FR, Thorpe R, Gordon-Salant S, Ferrucci L. Hearing loss prevalence and risk factors among older adults in the United States. J Gerontol A Biol Sci Med Sci. 2011;66(5):582-590. doi: 10.1093/gerona/glr002 [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Reed NS, Garcia-Morales EE, Myers C, et al. Prevalence of hearing loss and hearing aid use among US Medicare beneficiaries aged 71 years and older. JAMA Netw Open. 2023;6(7):e2326320. doi: 10.1001/jamanetworkopen.2023.26320 [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Loughrey DG, Kelly ME, Kelley GA, Brennan S, Lawlor BA. Association of age-related hearing loss with cognitive function, cognitive impairment, and dementia: a systematic review and meta-analysis. JAMA Otolaryngol Head Neck Surg. 2018;144(2):115-126. doi: 10.1001/jamaoto.2017.2513 [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Lin FR, Pike JR, Albert MS, et al. ; ACHIEVE Collaborative Research Group . Hearing intervention versus health education control to reduce cognitive decline in older adults with hearing loss in the USA (ACHIEVE): a multicentre, randomised controlled trial. Lancet. 2023;402(10404):786-797. doi: 10.1016/S0140-6736(23)01406-X [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Lee M, Whitsel E, Avery C, et al. Variation in population attributable fraction of dementia associated with potentially modifiable risk factors by race and ethnicity in the US. JAMA Netw Open. 2022;5(7):e2219672. doi: 10.1001/jamanetworkopen.2022.19672 [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Nianogo RA, Rosenwohl-Mack A, Yaffe K, Carrasco A, Hoffmann CM, Barnes DE. Risk factors associated with Alzheimer disease and related dementias by sex and race and ethnicity in the US. JAMA Neurol. 2022;79(6):584-591. doi: 10.1001/jamaneurol.2022.0976 [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Ehrlich JR, Goldstein J, Swenor BK, Whitson H, Langa KM, Veliz P. Addition of vision impairment to a life-course model of potentially modifiable dementia risk factors in the US. JAMA Neurol. 2022;79(6):623-626. doi: 10.1001/jamaneurol.2022.0723 [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Smith JR, Huang AR, Lin FR, Reed NS, Deal JA. The population attributable fraction of dementia from audiometric hearing loss among a nationally representative sample of community-dwelling older adults. J Gerontol A Biol Sci Med Sci. 2023;78(7):1300-1306. doi: 10.1093/gerona/glad117 [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Kamil RJ, Genther DJ, Lin FR. Factors associated with the accuracy of subjective assessments of hearing impairment. Ear Hear. 2015;36(1):164-167. doi: 10.1097/AUD.0000000000000075 [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Wright JD, Folsom AR, Coresh J, et al. The ARIC (Atherosclerosis Risk In Communities) study: JACC focus seminar 3/8. J Am Coll Cardiol. 2021;77(23):2939-2959. doi: 10.1016/j.jacc.2021.04.035 [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Deal JA, Sharrett AR, Albert MS, et al. Hearing impairment and cognitive decline: a pilot study conducted within the atherosclerosis risk in communities neurocognitive study. Am J Epidemiol. 2015;181(9):680-690. doi: 10.1093/aje/kwu333 [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Knopman DS, Gottesman RF, Sharrett AR, et al. Mild cognitive impairment and dementia prevalence: the Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS). Alzheimers Dement (Amst). 2016;2:1-11. doi: 10.1016/j.dadm.2015.12.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
15.World Health Organization . International Classification of Diseases, Ninth Revision (ICD-9). World Health Organization; 1977. [Google Scholar]
16.World Health Organization . International Statistical Classification of Diseases, Tenth Revision (ICD-10). World Health Organization; 1992. [Google Scholar]
17.Knopman DS, Pike JR, Gottesman RF, et al. Patterns of cognitive domain abnormalities enhance discrimination of dementia risk prediction: the ARIC study. Alzheimers Dement. 2024;20(7):4559-4571. doi: 10.1002/alz.13876 [DOI] [PMC free article] [PubMed] [Google Scholar]
18.American Psychiatric Association . Diagnostic and Statistical Manual of Mental Disorders. 5th ed. American Psychiatric Association; 2013. [Google Scholar]
19.McClannahan KS, Chiu YF, Sommers MS, Peelle JE. Test-retest reliability of audiometric assessment in individuals with mild dementia. JAMA Otolaryngol Head Neck Surg. 2021;147(5):442-449. doi: 10.1001/jamaoto.2021.0012 [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Clark JG. Uses and abuses of hearing loss classification. ASHA. 1981;23(7):493-500. [PubMed] [Google Scholar]
21.Koton S, Schneider AL, Rosamond WD, et al. Stroke incidence and mortality trends in US communities, 1987 to 2011. JAMA. 2014;312(3):259-268. doi: 10.1001/jama.2014.7692 [DOI] [PubMed] [Google Scholar]
22.Miettinen OS. Proportion of disease caused or prevented by a given exposure, trait or intervention. Am J Epidemiol. 1974;99(5):325-332. doi: 10.1093/oxfordjournals.aje.a121617 [DOI] [PubMed] [Google Scholar]
23.Gassama M, Bénichou J, Dartois L, Thiébaut AC. Comparison of methods for estimating the attributable risk in the context of survival analysis. BMC Med Res Methodol. 2017;17(1):10. doi: 10.1186/s12874-016-0285-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Benichou J. A review of adjusted estimators of attributable risk. Stat Methods Med Res. 2001;10(3):195-216. doi: 10.1177/096228020101000303 [DOI] [PubMed] [Google Scholar]
25.Norton S, Matthews FE, Barnes DE, Yaffe K, Brayne C. Potential for primary prevention of Alzheimer’s disease: an analysis of population-based data. Lancet Neurol. 2014;13(8):788-794. doi: 10.1016/S1474-4422(14)70136-X [DOI] [PubMed] [Google Scholar]
26.Welberry HJ, Tisdell CC, Huque MH, Jorm LR. Have we been underestimating modifiable dementia risk? An alternative approach for calculating the combined population attributable fraction (PAF) for modifiable dementia risk factors. Am J Epidemiol. 2023;192(10):1763-1771. doi: 10.1093/aje/kwad138 [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Smith JR, Deal JA. The need for a more holistic approach to dementia prevention. Lancet Healthy Longev. 2024;5(6):e382-e383. doi: 10.1016/S2666-7568(24)00071-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Rothman KJ. No adjustments are needed for multiple comparisons. Epidemiology. 1990;1(1):43-46. doi: 10.1097/00001648-199001000-00010 [DOI] [PubMed] [Google Scholar]
29.Thomas JD, Bradley E. Bootstrap confidence intervals. Stat Sci. 1996;11(3):189-228. [Google Scholar]
30.Sjölander A, Vansteelandt S. Doubly robust estimation of attributable fractions in survival analysis. Stat Methods Med Res. 2017;26(2):948-969. doi: 10.1177/0962280214564003 [DOI] [PubMed] [Google Scholar]
31.White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Stat Med. 2011;30(4):377-399. doi: 10.1002/sim.4067 [DOI] [PubMed] [Google Scholar]
32.World Report on Hearing. World Health Organization; 2021. [Google Scholar]
33.Darrow LA, Steenland NK. Confounding and bias in the attributable fraction. Epidemiology. 2011;22(1):53-58. doi: 10.1097/EDE.0b013e3181fce49b [DOI] [PubMed] [Google Scholar]
34.Ferguson J, Alvarez A, Mulligan M, Judge C, O’Donnell M. Bias assessment and correction for Levin’s population attributable fraction in the presence of confounding. Eur J Epidemiol. 2024;39(2):111-119. doi: 10.1007/s10654-023-01063-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Huang AR, Jiang K, Lin FR, Deal JA, Reed NS. Hearing loss and dementia prevalence in older adults in the US. JAMA. 2023;329(2):171-173. doi: 10.1001/jama.2022.20954 [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Lin FR, Albert M. Hearing loss and dementia—who is listening? Aging Ment Health. 2014;18(6):671-673. doi: 10.1080/13607863.2014.915924 [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Whitson HE, Cronin-Golomb A, Cruickshanks KJ, et al. American Geriatrics Society and National Institute on Aging Bench-to-Bedside Conference: sensory impairment and cognitive decline in older adults. J Am Geriatr Soc. 2018;66(11):2052-2058. doi: 10.1111/jgs.15506 [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Powell DS, Oh ES, Lin FR, Deal JA. Hearing impairment and cognition in an aging world. J Assoc Res Otolaryngol. 2021;22(4):387-403. doi: 10.1007/s10162-021-00799-y [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Yeo BSY, Song HJJMD, Toh EMS, et al. Association of hearing aids and cochlear implants with cognitive decline and dementia: a systematic review and meta-analysis. JAMA Neurol. 2023;80(2):134-141. doi: 10.1001/jamaneurol.2022.4427 [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Deal JA, Power MC, Palta P, et al. Relationship of cigarette smoking and time of quitting with incident dementia and cognitive decline. J Am Geriatr Soc. 2020;68(2):337-345. doi: 10.1111/jgs.16228 [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Adkins-Jackson PB, George KM, Besser LM, et al. The structural and social determinants of Alzheimer’s disease related dementias. Alzheimers Dement. 2023;19(7):3171-3185. doi: 10.1002/alz.13027 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials