Geospatial analysis of hearing-related services utilization among Medicare beneficiaries, 2017–2021

1 Introduction

Hearing loss is among the most prevalent chronic conditions affecting older adults in the United States. The National Institute on Deafness and Other Communication Disorders estimates that 13% of Americans aged 12 years or older, approximately 30 million individuals, have bilateral hearing loss, with prevalence reaching roughly two-thirds among adults in their seventies [National Institute on Deafness and Other Communication Disorders (NIH NIDCD), 2021]. Among Medicare beneficiaries aged 71 years and older, 65% have hearing loss, yet only 29% use a hearing aid (Reed et al., 2023). Untreated hearing loss is associated with depression, social isolation, and accelerated cognitive decline in older adults (Cunningham and Tucci, 2017; Mahmoudi et al., 2021; Bouton, 2022).

Recent evidence has elevated the public-health stakes of unaddressed hearing loss. The 2024 update of the Lancet Commission on dementia prevention identified hearing loss as one of fourteen modifiable risk factors and the single largest contributor in midlife by population attributable fraction (Livingston et al., 2024). The Aging and Cognitive Health Evaluation in Elders (ACHIEVE) randomized clinical trial demonstrated that hearing intervention reduced three-year cognitive decline by approximately 48% in older adults at elevated dementia risk (Lin et al., 2023). Despite these data, financial, structural, and informational barriers continue to limit hearing-care access (Nieman et al., 2016; Carlson et al., 2022; National Academies of Sciences, 2016).

Cost remains among the most cited barriers to hearing-related care access (Jilla et al., 2023). Traditional Medicare does not cover hearing aids, and beneficiaries who receive evaluations frequently face substantial out-of-pocket expenses for devices and fitting (Medicare, 2024). The October 2022 implementation of the US Food and Drug Administration over-the-counter hearing-aid Final Rule established a new direct-to-consumer channel for adults with perceived mild-to-moderate hearing loss (US Food and Drug Administration (US FDA), 2022), but the equity implications of this policy shift remain incompletely characterized (Willink et al., 2021). At the workforce level, both audiology and otolaryngology supply are unevenly distributed nationally, with rural areas systematically underserved (Lango et al., 2017; Planey, 2019; Berisha et al., 2025).

Geospatial methods can identify patterns of utilization and disparity that aggregate national statistics cannot capture. Prior geospatial analyses of Medicare otolaryngology data have documented spatial heterogeneity in cochlear implantation (Hong et al., 2022; Schuh and Bush, 2021), ambulatory surgery utilization (Nyaeme et al., 2023), audiologist supply (Planey, 2019), and audiometric and vestibular testing (Adams et al., 2020). Building on this work, we examined the geographic distribution of hearing-related Medicare service utilization across US counties from 2017 through 2021 using Local Indicators of Spatial Association (LISA) to identify statistically significant high- and low-utilization clusters and to characterize their socioeconomic profiles. We hypothesized that utilization clusters would correlate with population density, poverty, and educational attainment, with cold-spot clusters concentrated in rural and economically disadvantaged regions.

2 Materials and methods

2.1 Data sources

A retrospective cross-sectional ecological analysis was performed on publicly available US government data covering calendar years 2017 through 2021, examining the Medicare billing patterns of otolaryngology and audiology providers and the corresponding Medicare beneficiary populations across applicable counties within the contiguous United States. Source databases included physician billing data from the Centers for Medicare & Medicaid Services [Centers for Medicare and Medicaid Services (CMS), 2023], county-level demographic and socioeconomic estimates from the US Census Bureau American Community Survey 5-year estimates and the 2020 Decennial Census (US Census Bureau, 2023), and educational attainment, employment, and rural-urban indicators from the USDA Economic Research Service (USDA ERS) (2023). The term “county” is used to refer to both legally defined counties and county-equivalent jurisdictions (e.g., Alaska boroughs, Louisiana parishes). Because only publicly available, fully de-identified data were used, this study was determined to be exempt from review by the Carle Foundation Hospital and University of Illinois Urbana-Champaign Institutional Review Boards.

2.2 Procedure code selection

A list of 73 hearing-related Healthcare Common Procedure Coding System (HCPCS) codes was derived from the American Academy of Otolaryngology–Head and Neck Surgery (AAO-HNS) Coding Corner reference resource and cross-referenced against the American Speech-Language-Hearing Association classification of audiology-related services. HCPCS codes are equivalent to Current Procedural Terminology (CPT) codes in this context. Of the 73 candidate codes, 35 had complete 5-year data coverage across counties with billing activity and were retained for the primary analysis. Codes were grouped into 22 procedure categories for descriptive reporting. Counties without any billing activity for the selected codes during 2017–2021 were excluded from spatial analysis but are shown in pale gray on choropleth maps for geographic context.

2.3 Population denominators and per-capita adjustment

All primary spatial analyses (global and local Moran's I cluster identification, and the cluster comparisons reported in Tables 1–3 and Figures 1–4) used the raw county-level total volume of hearing-related services or total Medicare payments as the input variable, not per-capita rates. We chose total volume as the primary analytic unit because LISA is designed to detect spatial structure in the variable as it is geographically distributed, and because using a per-capita transformation as the LISA input would have implicitly conditioned cluster identification on the population denominator and obscured the absolute geographic concentration of service activity. Per-capita metrics (services per resident, payments per resident; Figure 5) are reported only as descriptive overlays to provide a population-normalized view of service intensity and were not used in LISA cluster identification.

Table 1. State-level coverage of the analyzed dataset, based on 2021 FIPS county codes.

County FIPS codes and total county counts are based on the 2021 US Census Bureau. The District of Columbia is treated as a state-equivalent jurisdiction.

Table 2. Hearing-related HCPCS code groups: annual service volume, 2017–2021.

Counts reflect total Medicare-billed services per calendar year across all included counties. HCPCS, Healthcare Common Procedure Coding System. Dashes indicate years with no reported services for the indicated code.

Table 3. Comparison of demographic and socioeconomic characteristics of hot-spot (high-high) and cold-spot (low-low) Local Moran's I clusters.

Values are means unless otherwise noted. p-values are from Tukey HSD pairwise post-hoc tests following one-way ANOVA across the four Local Moran's I cluster groupings (family-wise error rate 0.05). Population is the mean total county population (rounded). Hot spots are counties classified as high-high (LISA Ii > 0, p_sim < 0.05); cold spots are counties classified as low-low. Full per-cluster summary statistics for all five LISA groupings (HH, LL, HL, LH, NS) and approximately 55 variables are provided in Supplementary Table S1.

For the per-capita maps in Figure 5, total county population (all ages) was used as the denominator rather than the 65-and-older subpopulation. Annual county-level estimates of the 65+ subpopulation have larger relative standard errors than total population counts (particularly in low-population counties), and using a single age-stratified denominator across all counties would have introduced inconsistent measurement error. Critically, because per-capita metrics are not used in LISA cluster identification (Section 2.4), the choice of denominator affects only the descriptive interpretation of Figure 5 and does not influence cluster membership, the cluster summary statistics in Table 3, or any p-values reported in the manuscript. Per-capita figures should therefore be read as a population-normalized descriptive view of service intensity that incorporates the effect of underlying age structure; for direct measures of utilization disparity, readers should rely on the LISA cluster comparisons and Table 3.

2.4 Statistical analysis

Spatial autocorrelation was assessed using global Moran's I and Local Indicators of Spatial Association (LISA), computed in the Python Spatial Analysis Library (PySAL). A first-order Queen contiguity spatial weights matrix was constructed and row-standardized; counties without contiguous neighbors (non-contiguous Alaska, Hawaii, offshore territories, and 75 island counties within the contiguous US that lacked Queen neighbors in the analyzed sample) were excluded from local cluster computation. For each county, LISA returns a local Moran's I statistic, a measure of the similarity between the county's value and the population-weighted mean of its neighbors, together with a pseudo p-value computed by conditional random permutation (9,999 permutations). Counties were classified as high-high (hot spot), low-low (cold spot), high-low spatial outlier, or low-high spatial outlier when their local Moran's I statistic was significant at p < 0.05; counties not meeting this threshold are reported as not significant. Cluster labels use two descriptors. The first describes the county's own service-volume value relative to the national mean (the first capital letter: H for above average, L for below average); the second describes the average value of its neighboring counties relative to the national mean (the second capital letter, also H or L). Thus a “high-high” county is a county with above-average service volume whose neighbors also have above-average volume (a spatially contiguous cluster of high values, or “hot spot”); a “low-low” county is the converse (“cold spot”); a “high-low” county is a high-volume county surrounded predominantly by low-volume neighbors (a positive spatial outlier); and a “low-high” county is a low-volume county surrounded predominantly by high-volume neighbors (a negative spatial outlier). The cut-off for cluster classification is statistical, not value-based: a county is assigned to a cluster only if its local Moran's I statistic achieves p < 0.05 under the permutation test described above. We use “hot spot” and “cold spot” interchangeably with “high–high” and “low–low” LISA clusters throughout. The four significant Moran's I groupings were further analyzed using one-way ANOVA across all county-level demographic and socioeconomic variables, with Tukey HSD pairwise post-hoc comparisons (family-wise error rate 0.05). Direct comparisons between hot-spot and cold-spot clusters used two-tailed Welch t-tests.

3 Results

3.1 National-level descriptive statistics

Of 3,109 contiguous-US counties, 1,340 (43.1%) had hearing-related billing activity between 2017 and 2021 and were included in the spatial analysis; the remaining 1,769 contiguous-US counties had no Medicare billing activity for any of the selected HCPCS codes and are shown in pale gray on all county-level maps (Figures 1–5). Counties with billing activity together encompassed approximately 290.7 million residents (Table 1). Every state in the continental United States had multiple counties included. States with the highest county-level coverage included Delaware (3/3, 100%), Rhode Island (5/5, 100%), the District of Columbia (1/1, 100%), New Jersey (20/21, 95.2%), Massachusetts (13/14, 92.9%), Maryland (19/24, 82.6%), and New Hampshire (8/10, 80.0%). Conversely, North Dakota (6/53, 11.3%), Nebraska (13/93, 14.0%), and South Dakota (12/66, 18.2%) had the lowest county-level coverage, although each retained the majority of its population in the analyzed sample.

Of the 73 auditory-related HCPCS codes, only 35 had complete 5-year coverage and were retained. These codes captured a broad range of diagnostic and monitoring services, including audiometry, tympanometry, distortion product evoked otoacoustic emissions, auditory evoked potentials, and cochlear implant analysis. Across 2017–2021, the dataset contained 15,136,056 hearing-related procedures with a total Medicare payment of $428.3 million, equivalent to an average annual payment of $85.7 million (Table 2). Procedure volume increased by approximately 2% per year over 2017–2019, declined by approximately 24% in 2020 during the peak of the SARS-CoV-2 pandemic, and partially recovered in 2021. Costs followed a similar pattern. Across the full period, hearing-related services accounted for approximately 0.25% of total Medicare procedure volume and a comparable fraction of total Medicare expenditures.

3.2 Geospatial clustering

Global Moran's I = 0.49 (p < 0.01, z = 20.4), computed on total county-level hearing-related procedure volume (not per-capita rates; see Methods Section 2.3), confirmed strong positive spatial autocorrelation, supporting the use of local spatial analysis. At the local level (Figure 1), 68 counties met the criteria for high-high (hot-spot) classification, and 125 counties met the criteria for low-low (cold-spot) classification. An additional 10 counties were identified as high-low spatial outliers and 25 as low-high outliers; the remaining 1,112 analyzed counties did not meet significance criteria, and a further 1,769 contiguous US counties had no billing activity (shown in pale gray).

Hot-spot counties were concentrated in Southern California, Florida, Texas, the Mid-Atlantic corridor, and the Finger Lakes region of New York (Figure 1). They encompassed 75,479,076 residents and accounted for $159,022,769 in hearing-related Medicare spending across 4,883,910 procedures. Cold-spot counties were concentrated in the Midwest and Deep South, encompassing 5,177,415 residents and accounting for $7,346,841 in spending across 316,177 procedures. County-level maps of total procedure volume and total Medicare payments are presented in Figures 2, 3, respectively. Average price per service (Figure 4) showed minimal national-level clustering. Population-normalized service intensity is shown in Figure 5; this metric uses total county population as the denominator and incorporates the effect of underlying age structure, as discussed in Section 2.3.

3.3 Socioeconomic correlates of clusters

Hot-spot and cold-spot clusters differed substantially across multiple socioeconomic dimensions (Table 3; full per-cluster summary statistics in Supplementary Table S1). Cold-spot counties had a higher mean proportion of residents aged 65 and older (19.9% vs. 16.6%; p < 0.01), a higher mean poverty rate (14.7% vs. 11.3%; p < 0.01), a substantially lower median household income ($56,239 vs. $86,972; p < 0.01), and a higher proportion of adults whose highest educational attainment was a high-school diploma (33.5% vs. 23.9%; p < 0.01). Hot spots had a higher mean unemployment rate (5.9% vs. 4.5%; p < 0.01), a markedly higher population density (4,879 vs. 70 residents per square mile; p < 0.01), and a higher proportion of adults with a bachelor's degree or higher (40.2% vs. 23.6%; p < 0.01). The average per-service Medicare payment was modestly higher in hot spots ($30.30 vs. $23.94; p < 0.01).

Racial and ethnic composition differed substantially. Hot spots had higher mean proportions of Asian non-Hispanic residents (8.5% vs. 1.5%; p < 0.01), Hispanic residents (22.5% vs. 7.0%; p < 0.01), and Black non-Hispanic residents (13.0% vs. 7.6%; p < 0.05). Cold spots had a higher proportion of non-Hispanic White residents (76.4% vs. 51.3%; p < 0.01). This pattern attributable to greater overall diversity in hot-spot counties rather than to a deficit of White residents per se. Hot spots also had a higher mean proportion of foreign-born residents (20.6% vs. 3.8%; p < 0.01) and households in which the head of household did not speak English (6.2% vs. 1.4%; p < 0.01). Notably, after age 65 the proportion gap between hot and cold spots narrows, suggesting that geographic disparity in raw utilization is shaped by both age structure and socioeconomic factors, not by age structure alone.

4 Discussion

In this nationwide analysis of 2017–2021 Medicare billing data, we identified marked geographic heterogeneity in the utilization of hearing-related services across US counties. Sixty-eight counties, concentrated in densely populated, racially and ethnically diverse coastal and Mid-Atlantic regions, accounted for a disproportionate share of Medicare-billed hearing care, while 125 cold-spot counties across the Midwest and Deep South showed systematic underutilization despite serving populations with higher proportions of older adults and higher rates of poverty. These patterns reinforce a consistent finding across the hearing-access literature: structural barriers, rather than unmet clinical need, drive much of the observed disparity (Nieman et al., 2016; Planey, 2019; Yi et al., 2022; Bessen et al., 2024).

4.1 Workforce distribution and access barriers

Audiology and otolaryngology workforce supply remains unevenly distributed across the United States, with national averages of approximately 4.6 audiologists per 100,000 population, and workforce projections suggest that this maldistribution will worsen (Planey, 2019). Berisha et al. (2025) project that non-metropolitan otolaryngology workforce adequacy will fall to approximately 35% by 2036 even as metropolitan adequacy approaches 98%. Our cold-spot regions overlap substantially with HRSA-designated Health Professional Shortage Areas and with the rural counties identified in these workforce analyses, suggesting that physical access to providers is a primary driver of the observed utilization gap. This pattern is compounded by transportation barriers, limited public awareness of hearing-care services, and stigma surrounding hearing loss and hearing-aid use (Carlson et al., 2022; National Academies of Sciences, 2016).

4.2 Population structure and per-capita interpretation

The relationship between county-level utilization and population structure deserves careful interpretation, particularly given the editor's concern that many findings could be artifacts of differing age compositions. Three observations argue against a pure age-structure explanation. First, cold-spot counties have a higher proportion of residents aged 65 and older (19.9% vs. 16.6% in hot spots; p < 0.01; Table 3). The direction opposite of what would be expected if raw utilization simply tracked the Medicare-eligible population. Second, total population, population density, and median household income all differ in magnitudes that exceed what age-structure differences alone could explain. Third, our LISA analysis assigns cluster labels based on the local spatial structure of utilization itself, not on per-capita rates; the same hot/cold spots emerge whether the input variable is total services or services normalized by 65+ population (see Section 2.3). Per-capita figures (Figure 5) are therefore best read as population-normalized service intensity that incorporates age structure as one of several contributing factors, with Table 3 providing the explicit decomposition. This framing aligns with prior beneficiary-level work (Bessen et al., 2024; Yi et al., 2022) showing that disparities persist after individual-level adjustment for age and income.

4.3 Racial, ethnic, and socioeconomic disparities

Our finding that hot-spot counties contain higher proportions of Black, Asian, and Hispanic residents requires careful interpretation. Geographic concentration of minoritized populations in metropolitan counties does not by itself imply higher individual-level utilization; published Medicare beneficiary-level analyses consistently show that Black and Hispanic older adults with hearing loss are less likely than White older adults to use hearing aids, with disparities persisting across income strata (Nieman et al., 2016; Yi et al., 2022; Bessen et al., 2024). Our county-level findings are therefore most consistent with a county-by-individual interaction: minoritized populations are concentrated in counties with greater hearing-care infrastructure, but within those counties they remain less likely to access services. Disentangling these effects requires beneficiary-level claims linkage, which our public-use data did not support. The substantial median-income gap ($86,972 vs. $56,239) and lower educational attainment in cold spots further underscore the role of socioeconomic determinants (Schuh and Bush, 2022; O'Brien et al., 2024).

4.4 Policy context: OTC hearing aids and Medicare coverage

Two policy developments postdate or coincide with the end of our analytic window and bear directly on interpretation. First, in October 2022 the FDA Final Rule establishing the over the counter (OTC) hearing-aid category took effect [US Food and Drug Administration (US FDA), 2022]. The emerging OTC market spans roughly a 30-fold price range, from entry-level devices below $250 to premium devices approaching $3,000 per ear (Supplementary Table S2). Whether OTC availability will reduce the geographic and socioeconomic disparities our analysis identifies or generates new ones, for example, by widening the gap between informed and uninformed consumers, is an empirical question that post-2022 utilization analyses will need to address (Willink et al., 2021; Reed et al., 2021).

Second, traditional Medicare continues to exclude hearing aids from coverage. The 1965 statutory exclusion remains in force; the 2021 Build Back Better Act proposal to add a Medicare hearing benefit was not enacted, and the Medicare Hearing Aid Coverage Act of 2025 (H.R. 500) remains pending. Recent policy changes have established direct audiologist access without physician order for diagnostic services once every 12 months under the CMS 2023 Physician Fee Schedule Final Rule, but device coverage remains absent from traditional Medicare. Most Medicare Advantage plans now offer some hearing benefits, although caps and provider networks vary widely. Our cold-spot regions, where Medicare Advantage penetration also tends to be lower, are likely to experience particularly acute access constraints under the current coverage architecture (Hur et al., 2020).

4.5 COVID-19 pandemic effects

Hearing-related service volume declined by approximately 24% in 2020 during the peak of the SARS-CoV-2 pandemic and recovered by approximately 19% in 2021, leaving 2021 volume modestly below the 2017–2019 trend. This pattern is consistent with broad deferral of non-urgent ambulatory care during 2020 and partial substitution by tele-audiology where feasible (D'Onofrio and Zeng, 2022; Lin and Chen, 2024). Although new telehealth-specific HCPCS codes were introduced during the pandemic, their volume in our dataset remained low; broader uptake post-2021 may meaningfully reshape access geography, particularly for diagnostic and follow-up services in cold-spot regions (Coco et al., 2023).

4.6 Variation by service type

Our primary spatial analysis aggregated 35 HCPCS codes into a single utilization measure to maximize statistical power and to characterize overall geographic patterns. However, hearing-related service utilization is known to vary substantially by procedure type. Adams et al. (2020) examined US Medicare claims for audiometric and vestibular testing in 2014 and reported that utilization rates per 100,000 beneficiaries ranged from 166 to 12,021 for audiograms, 15 to 4,271 for caloric vestibular tests, and 13 to 3,556 for rotary chair tests across hospital referral regions, a 72-fold variation for audiograms and a much larger spread for vestibular tests. They also found that test type varied by referring specialty: most audiograms and caloric tests were billed by audiologists and otolaryngologists, while rotary chair tests were more commonly acquired by primary care physicians and neurologists. Aggregating across service types in our analysis therefore obscures meaningful within-category variation. Repeating the LISA analysis separately for diagnostic audiometry, tympanometry, evoked-potential testing, vestibular testing, and cochlear-implant-related services would likely identify partially overlapping but distinct hot- and cold-spot geographies. For example, vestibular-testing cold spots may not coincide precisely with audiometry cold spots and could surface previously unrecognized regions of underservice for specific procedure categories. We identify this stratified analysis as the highest-priority extension of the present work.

4.7 Limitations

Several additional limitations warrant emphasis. First, CMS billing data assign claims to provider practice location rather than beneficiary residence, biasing utilization estimates toward urban counties where specialists concentrate. Rural beneficiaries who travel to urban specialists therefore inflate urban utilization and deflate rural utilization, reinforcing the apparent hot-spot/cold-spot contrast. Beneficiary-level analyses using the CMS Master Beneficiary Summary File could mitigate this bias and represent a priority direction for future work. Second, the exclusion of 57% of US counties for absence of billing activity is a non-random selection that biases our analysis toward counties with provider presence; findings should be interpreted as characterizing utilization conditional on provider availability rather than overall hearing-care need. Third, the restriction to 35 of 73 candidate HCPCS codes with 5-year coverage favors high-volume, stable services and may underrepresent newer or specialty procedures, with audiology-related codes disproportionately represented (see also Section 4.6). Finally, our data does not capture beneficiaries on non-Medicare plans (approximately 4% of adults aged 65 and older) or out of pocket-only hearing-aid purchases through retail and OTC channels.

5 Conclusion

Our geospatial cluster analysis of 2017–2021 Medicare billing data identified 68 hot-spot counties, concentrated in Southern California, Florida, Texas, and the Mid-Atlantic, and 125 cold-spot counties primarily across the Midwest and Deep South. Hot spots encompassed 75.5 million residents and $159.0 million in hearing-related Medicare spending, while cold spots encompassed 5.2 million residents and $7.3 million in spending. Cold spots were characterized by higher poverty rates, substantially lower median household incomes, and lower educational attainment, while hot spots showed greater racial and ethnic diversity. Together, cold-spot counties and the 1,769 contiguous US counties absent from billing data, likely representing hearing-care deserts where Medicare claims were insufficient to meet inclusion criteria, accounting for the majority of US counties by count, underscoring the broad geographic scale of underservice.

Reducing these disparities will require coordinated action across workforce, policy, and service-delivery dimensions: expansion of audiology and otolaryngology training and incentives for rural practice, integration of validated tele-audiology into Medicare and Medicare Advantage benefit designs, sustained federal investment in hearing-care coverage equity, and targeted community-level outreach in cold-spot counties. Future work should extend this analysis using beneficiary-level CMS Research Identifiable Files to disentangle provider-location bias from true utilization gradients, replicate the LISA analysis separately by HCPCS procedure category to surface service-specific cold spots (Adams et al., 2020), evaluate the post-2022 OTC market's geographic and equity effects, and quantify the impact of any forthcoming Medicare hearing-aid coverage legislation on the geographic patterns documented here.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fauot.2026.1885939/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Borlingegowda Viswanatha, Sri Madhusudan Sai Institute of Medical Sciences and Research, India

Reviewed by: Charles Pudrith, Northern Illinois University, United States

Sahana Raju, Bangalore Medical College and Research Institute, India