Estimating the United States Patient Population Size Meeting Audiologic Candidacy for Cochlear Implantation

Abstract

Objective:

This study aimed to estimate the size of the United States candidacy pool meeting expanded Center for Medicare Services criteria for cochlear implantation.

Study Design:

Retrospective cross-sectional

Setting:

Tertiary care center

Methods:

Preimplantation audiometric data from 486 patients seen at a single academic medical center were collected retrospectively and used to generate a predictive model of AzBio score based on audiometric pure tone thresholds. This model was then used to estimate nationally representative CI-candidacy using pure tone averages included in the National Health and Nutrition Examination Survey. Qualitative and quantitative analyses were performed.

Results:

We find that the estimated prevalence of CI candidacy in individuals 65 years of age or older is expected to more than double with a change in the CI candidacy criteria from ≤40% to ≤60% (from 1.42%, 95% confidence interval [1.33, 1.63] to 3.73% [2.71, 6.56]) on speech testing. We also found the greatest absolute increase in candidacy in the 80+ age group, increasing from 4.14% [3.72, 5.1] of the population meeting the ≤40% criteria to 12.12% [9.19, 18.35] meeting the ≤60% criteria.

Conclusions:

The United States population size meeting expanded CMS audiologic criteria for cochlear implantation is estimated to be 2.5 million adults and 2.1 million age 65 or older. Changing the CI candidacy criteria from ≤40% to ≤60% on CI testing has the greatest effect on the eligible patient population in the > 65-year-old age group. The determination of utilization rates in newly eligible patients will require further study.

Introduction

Hearing loss affects up to 20% of Americans aged 12 years or older and is associated with reduced cognitive performance, social isolation, and dementia.^1-4 Economically, hearing loss is associated with less employment and lower compensation, with an estimated annual lost income of over 176.3 billion dollars nationwide.^5,6 For individuals with severe to profound sensorineural hearing loss no longer benefiting from amplification, cochlear implantation (CI) is an effective intervention.^7,8 In 1986, Center for Medicare Services (CMS) began covering CI in post-lingual deaf adults and later expanded coverage to all people meeting CI testing criteria. From 2005 until 2022, CMS set the criteria to qualify for CI at a score of ≤40% correct in the best-aided listening condition on speech testing.⁹

Recent evidence suggests that a 40% threshold may not capture all patients that could benefit from CI, particularly older individuals.^10-12 In 2012, Lin et. al showed that individuals older than 60 improved by an average of 10% more on the Hearing in Noise Test (HINT) if pre-operative CI testing scores were between 40% and 60% rather than under 40%.¹⁰ In 2021, Perkins et. al showed that, of candidates with pre-operative CI testing scores up to 60%, a majority improved post-operatively and none had significant decrements in their hearing.¹¹ Finally, Zwolan et. al showed, in a non-randomized prospective clinical trial, that cochlear implantation in adults 65 or older with pre-operative AzBio scores from 41% to 60% resulted in a median improvement of over 30%.¹² In response, CMS recently expanded its criteria for CI candidacy from ≤40% to ≤60%, best aided, sentence recognition score.¹³ However, the number of patients who qualify under this new threshold is uncertain, and nationally representative data on CI testing is not readily available. Reliable estimation of the patient population allows for accurate estimation of utilization statistics and can help drive relevant changes to practice and policy to keep pace with the growing hearing-loss patient population.^14,15

To achieve this, we leveraged audiometry data included in the National Health and Nutrition Examination Survey (NHANES), which was recently used to estimate the prevalence of hearing loss in the US population by Carlson et. al.^15-17 While NHANES includes pure tone thresholds (PTTs), it does not include speech understanding testing, which is the basis of cochlear implantation criteria. Studies have shown that PTTs alone may not accurately estimate CI candidacy.^18,19 We therefore used patient-level data from CI candidacy evaluations at a single academic institution to develop a model to predict speech understanding scores from pure tone averages (PTAs). In this retrospective cross-sectional study, we used nationally-representative PTAs, as reported by NHANES, with predictive modeling to estimate the number of individuals in the United States expected to meet current CMS criteria for CI candidacy.

Methods

Approval was obtained from the Johns Hopkins Medicine Institutional Review Board (IRB00188251). This study was conducted in 2 parts, termed “development sample” and “national sample”. In the development sample, data from CI candidacy evaluations at a tertiary academic medical center were used for non-linear regression to generate a model to predict AzBio scores from PTAs. In the national sample, the predictive model was applied to PTAs from NHANES to predict AzBio scores, which were used to calculate the prevalence of CI candidacy.

Study Population

Development sample:

Data were retrospectively collected from patients aged 21 to 91 and seen between January 2021 and September 2022 at the Johns Hopkins Hospital for CI consultation with initial pre-operative evaluations ranging between 2011 and September 2022 (Supplemental Figure 1). This totaled 486 patients. 299 patients were excluded due to single-sided deafness, no pre-operative testing available, or lack of PTTs, AzBio sentences scores, or word recognition scores on preoperative assessment. 5 more pediatric patients were excluded as we restricted our study to adults (aged 18 or older).

National sample:

The Continuous NHANES 2015-2016 and 2017-2018 releases both included audiometry as half samples. In the 2015-2016 sample, adults aged 20-69 years were eligible. In the 2017-2018 sample, people aged 6-19 and 70+ years were eligible. Patients who were unable to remove their hearing aids or who had ear pain precluding testing with headphones were excluded from NHANES.²⁰ We restricted our study to adults (18 or older).

Audiometric Testing

Development sample:

Pre-operative testing included unaided pure-tone audiometry, and some combination of aided HINT sentences, CNC (Consonant-Nucleus-Consonant) word lists, or AzBio sentence scores. For this study, we exclusively used AzBio sentences tests done in quiet to maintain consistent associations between PTAs and CI testing scores and to maximize the pool of subjects with uniform speech testing data. Speech testing was conducted on a per-ear basis in the soundfield with stimuli presented at 60dbSPL in quiet.

National sample:

As described in the NHANES documentation, audiometry exam sections were performed by trained examiners in a sound-isolating room in the mobile examination center. Hearing thresholds were collected for the following frequencies: 500Hz, 1kHz, 2kHz, 3kHz, 4kHz, 6kHz, and 8kHz. Testing was conducted with a modified Hughson Westlake procedure using the automated testing mode of the audiometer. Observed values for hearing thresholds varied between −10 and 110dB.²⁰ PTTs that had ‘no response’ were coded as 120db HL and ‘could not obtain’ were coded as missing data. Data for PTAs are only collected for all ages above 5. Population estimates were generated using NHANES prevalence estimates and U.S. Census data from the 2019 release of the current population survey.²¹

Statistical Analysis

2-parameter log-logistic curves were fitted using the statistical package ‘drc’. Prevalence estimates were calculated using the statistical package ‘survey’ and reported as point estimates with 95% confidence intervals representing variance in model predictions. All statistical analyses were conducted in R4.2.2 and followed NHANES guidelines for weighting and data analysis.²²

Model generation and prevalence estimation

Institutional data were collected from preoperative CI evaluations based on inclusion and exclusion criteria listed above. This dataset contained ear-specific PTAs and corresponding AzBio Sentence scores. Each patient ear was treated separately for model regression. PTAs were calculated as an average of the PTTs at 500Hz, 1kHz, and 2kHz. A 2-parameter log-logistic model was created that predicted CI test scores using PTAs:

$$\frac{AzBioScore}{100}=\frac{1}{1+e^{\left(b\right(\log \left(PTA\right)-\log \left(m\right))}}$$

The parameters b and m represent the steepness and midpoint of the model and are found using regression (Figure 2). This model was then applied to the population PTAs obtained from NHANES to predict the proportion of individuals that would have AzBio Sentences scores below the defined threshold for CI candidacy (either ≤40% or ≤60%). While NHANES PTAs are also ear-specific, CI candidacy in this study was based on meeting threshold in both ears to reflect bilateral hearing loss.²¹

Sensitivity analysis

We tested the robustness of our results in two ways. First, due to the sigmoidal shape of the modeling fit, we tested the sensitivity of the predictive model to the midpoint (m) of our data using the same log-logistic model as above. Second, we also tested the assumption that our data were best fit with a log-logistic model by comparing it to a logistic model generated using the same data, with parameters b and m representing steepness and midpoint, respectively:

$$\frac{AzBioScore}{100}=\frac{1}{1+e^{b(\text{PTA}-m)}}$$

Results

Model Parameters

In the development sample, the average age was 68.2 ± 16.1 years old, ranging from 21 to 91. The average PTA was 73.1 db HL, ranging from 8.33 db HL to 118 db HL. The average AzBio score was 32.9% ± 31.2%. There were 117 scores below 40%, 26 scores in the 40%-60% range, and 49 scores above 60%. Demographics are summarized in Table 1. We modeled the PTA-AzBio relationship using a log-logistic model (Methods), which is shown in Figure 1 with a 95% confidence interval. We found that an AzBio score of 50% corresponded to a PTA of 57.5 ± 1.5 db HL. A PTA of 52.0 db HL corresponded to a score of 60 ± 6% and a PTA of 63.6 db HL corresponded to a score of 40 ± 4%.

Table 1.

Demographics

	Overall (N=182)
Age
Mean (SD)	68.2 (16.1)
Median [Min, Max]	70.0 [21.0, 91.0]
PTA (dB)
Mean (SD)	73.1 (19.4)
Median [Min, Max]	71.7 [8.33, 118]
AzBio Score
Mean (SD)	32.9 (31.2)
Median [Min, Max]	27.0 [0, 100]

Figure 1.

Predictive mode

Population Estimates

We found that 0.45% [0.43, 0.5] of the adult (≥18) population in NHANES met the threshold of ≤40% correct while 0.96% [0.74, 1.61] of the U.S. population met the expanded criteria threshold of ≤60% correct on CI testing, representing 1,162,473 and 2,479,942 individuals, respectively. Below the age of 65, small numbers of NHANES participants qualified for CI and prediction was less reliable. In participants aged 18 to 64 years old, 0.21% [0.21, 0.22] of the population (425,207 individuals) met the criteria for ≤40% correct while 0.27% [0.25, 0.37] met the criteria for ≤60% correct on CI testing (546,694 individuals). Over the age of 65, 1.42% [1.33, 1.63] met the criteria for ≤40% correct while 3.73 % [2.71, 6.56] met the criteria for ≤60% correct on CI testing, totaling 793,041 and 2,083,129 individuals, respectively. Prevalence estimates are summarized in Table 2.

Table 2.

Prevalence

Age Group (years)	Prevalence of Speech Perception Score ≤40%	95% CI	Population with Speech Perception Score ≤40%	95% CI	Prevalence of Speech Perception Score ≤60%	95% CI	Population with Speech Perception Score≤60%	95% CI	Ratio
Total (>=18)	0.45%	[0.43, 0.5]	1,162,473	[1,110,807, 1,291,637]	0.96%	[0.74, 1.61]	2,479,942	[1,911,622, 4,159,070]	213%
18-64	0.21%	[0.21, 0.22]	425,207	[425,207, 445,455]	0.27%	[0.25, 0.37]	546,694	[506,198, 749,174]	129%
65-69	1.1%	[1.1, 1.1]	202,338	[202,338, 202,338]	1.45%	[1.1, 3.16]	266,718	[202,338, 581,261]	132%
70-74	0.75%	[0.75, 0.83]	114,539	[114,539, 126,756]	0.89%	[0.83, 1.61]	135,919	[126,756, 245,876]	119%
75-79	0%	[0, 0]	0	[0, 0]	3.21%	[1.51, 7.59]	317,943	[149,562, 751,772]	--
80+	4.14%	[3.72, 5.1]	508,270	[456,707, 626,130]	12.12%	[9.19, 18.35]	1,487,980	[1,128,262, 2,252,841]	293%
CMS (>=65)	1.42%	[1.33, 1.63]	793,041	[742,778, 910,322]	3.73%	[2.71, 6.56]	2,083,129	[1,513,480, 3,663,626]	263%

We performed sensitivity analysis of the model estimate by shifting the midpoint (m) of our predictive model within the observed variance of PTAs (Supplemental Figure 2). When testing the effects of a shifting midpoint, our data showed that the ratio of patients meeting the ≤40% compared to the ≤60% changes by a maximum of 39%, ranging between 1.2 to 3.6 million individuals age 65 or older meeting CI candidacy (Table 3). In a separate sensitivity analysis, we also evaluated the effects of a log-logistic compared to previously published logistic models (Supplemental Table 1). ²³

Table 3.

Sensitivity to midpoint shift. Bounds are baseline +/− 3*SE

Prevalence	Baseline (57.5dB)	Upper Bound (51.7dB)	Lower Bound (63.3dB)
Prevalence of CI Score < 40% age 65 or older	1.42%	2.62%	1.04%
Prevalence of CI Score < 60% age 65 or older	3.73%	6.56%	2.62%
Population with CI Score < 40% age 65 or older	1,162,473	1,463,216	580,819
Population with CI Score < 60% age 65 or older	2,479,942	3,663,626	1,463,216
Ratio	213%	250%	252%

Discussion

Although nationally representative data estimating hearing loss in the US are widely available, they routinely rely on pure tone thresholds alone, even though CI candidacy is based on speech perception scores.^15,17,24 In this study, we develop a method to account for speech perception scores in addition to pure tone thresholds and present an estimate of the U.S. population size for CI candidacy at both 40% and 60% speech understanding thresholds, reflecting recent changes in CMS guidelines.⁹ We estimate that approximately 2.1 million adults 65 or older meet an expanded candidacy threshold of ≤60% on CI testing and that the impact scales with age. For instance, we found that expanding audiologic criteria from 40% to 60% on speech testing resulted in a minimal increase in the candidate pool for those under 65 years of age, from 0.21% to 0.27% of the population, totaling approximately 100,000 new qualifying individuals. In contrast, for individuals over the age of 80, the number of candidate individuals more than doubles (293%) to approximately 1.5 million qualifying individuals. These prevalence estimates further highlight the need of understanding CI-related outcomes, device use, quality of life, and barriers to care specifically in the elderly.

Our model estimates are consistent when contextualized with previous estimates at the 40% audiologic threshold. In 2013, Sorkin et al. estimated the total number of eligible CI candidates in the US to be 1.2 million people.²⁵ By 2015, Nassiri and colleagues estimated that up to 1.3 million people qualified for CI.^{17 15}Goman et. al used audiometric pure tone averages alone to yield a prevalence of 2.47% in the population over 60, representing 1.92 million candidates.²⁴ Using both speech perception scores on CI evaluation and pure tone thresholds from a nationally-representative sample, our model’s findings of 1.1 million people at the 40% threshold concords with these findings.

Our method is likely an underestimate for several reasons. First, we use only 500Hz, 1kHz, and 2kHz PTTs, which does not capture high frequency hearing loss with relatively preserved low frequency hearing that would render an individual a candidate for hybrid CI.^24,26 While the inclusion of higher frequency PTTs may increase the estimated patient population, it risks overestimation of the population qualifying under revised criteria. Second, our qualification criteria for CI require hearing loss under the 40% or 60% threshold in both ears, which may differ from real-world conditions.²⁷ Third, our predictive model was generated using AzBio scores in quiet to achieve necessary statistical power and avoid over-estimation. While scores obtained in noise would likely result in a larger candidate pool, prior studies have shown that testing in +10 dbSNR leads to a marginal increase in the candidate pool size by approximately 10%.²⁸ Lastly, our study focused on individuals with bilateral hearing loss, the broader CI population including asymmetric hearing loss and single-sided deafness has been estimated to be nearly 8 million candidates in total.^17,24

While our estimates quantify the population that meets audiologic CI candidacy criteria, this estimate does not account for medical (contra)indications, access to CI centers²⁹, and the highly personal decision in obtaining a CI. Previous studies have estimated CI utilization to be as low as 12.7% of individuals that meet traditional audiometric criteria.^15,17 To our knowledge, there are no data commenting on the differences in CI utilization based on the degree of hearing loss. However, it is reasonable to expect that CI candidates with lesser degrees of hearing loss may present for evaluation of their hearing loss at a lower rate than their counterparts with greater degrees of hearing loss. Previous studies into barriers to CI utilization suggest that issues outside of candidacy, such as socioeconomics, insurance accessibility, awareness, and restrictive binaural requirements represent significant barriers to greater CI uptake.^14,30,31 Although expansion of existing criteria by CMS will have an effect on the qualifying population, further efforts to remove barriers for patients desiring CI are necessary to improve utilization and quality of life outcomes for hearing loss patients.

While true validation is not possible, uncertainty in extrapolating the predictive model to the national population was addressed using a sensitivity analysis (Table 3). By shifting the midpoints in our model based on the observed variance of test scores in our clinical population, the size of the candidate pool age 65 or older and meeting <60% on CI testing ranged between 1.5 to 3.6 million individuals. We found that, while absolute populations changed, the ratio of new candidate population to previous candidates remained relatively stable and changed by at most 39% (Table 3), suggesting that our estimates of the relative increase in CI candidates are resistant to perturbations in our model. Further validation of our model would ideally include data from multiple centers.

Previously, multiple groups have shown that PTTs are effective predictors of speech recognition scores and word discrimination scores.^19,23 In 2015, Hoppe et. al demonstrated an estimation method for predicting aided Freiburg monosyllabic words speech recognition scores using a four-frequency hearing threshold average (composed of an average of 0.5Hz, 1kHz, 2kHz, and 4kHz PTTs).²³ Similar to our approach, they used a sigmoid function and found that 60.2db HL corresponded to a 50% score.²³ In 2016, Gubbels et. al correlated AzBio scores in quiet to individual PTTs, finding that a score of 40% on AzBio in quiet correlated to a 62.5db deficit at 500Hz.¹⁹ Compared to both previously described associations, our model has similar characteristics with a midpoint of 57.5db HL and a PTA of 63.6db HL corresponding to an AzBio in quiet score of 40 ± 4%. Additionally, the relatively wide distribution of data observed in our data (Figure 1) is seen in both Hoppe and Gubbels’ observations, suggesting that there are likely other factors at play in determining CI testing score, as enumerated in previous publications.¹⁸ We expand upon this work by Hoppe et. al and Gubbels et. al by estimating the total eligible population accounting for the new Medicare CI candidacy criteria outlined in 2022.^9,13

We chose a log-logistic model instead of the previously²³ used sigmoid model as the data collected from patients seen at our hospital deviated from the expected normal distribution (Supplemental Figure 2). As described above, our model parameters remained close to previously published models (midpoint = 57.5db HL vs 60.2db HL).²³ Additionally, when comparing both log-logistic and logistic models, we found that our data was better explained by a log-logistic (AIC = −24.2, BIC= −14.6) rather than a logistic model (AIC = −23.1, BIC= − 10.3). Finally, in sensitivity analyses, we found that using a logistic model did not change point estimates compared to a log-logistic model but expanded the 95% confidence intervals by up to 1.37% in one case.

While our study estimates align with previous estimates of the population qualifying for CI, several limitations exist. First, our data show an unexpected dip in the prevalence of CI score ≤40% from the 70-74 age group to the 75-79 age group. This phenomenon disappears in the estimate for the prevalence of CI score ≤60%, which suggests that this may represent an artifact of our modeling methodology. Second, our predictive model was generated using data from a single tertiary care center, limiting the generalizability of our model. Given the inaccessibility of data for external validation, we performed sensitivity analyses of our model estimates in an attempt to account for biases in our patient population. Lastly, this study focused solely on audiometric criteria for cochlear implantation. Medical and personal factors have important impact on the decision to undergo CI, and consequent population-level utilization rates, which are not addressed in this study.

Despite these limitations, we present a refined methodology for estimating CI candidacy pool based on speech perception testing rather than pure tone thresholds alone. Using this methodology, we present new estimates of potential CI population size based on newly approved CMS candidacy criteria and show that the largest impact to candidacy change is expected to occur in the over 80 age group, highlighting the need for better understanding CI use and outcomes in the elderly and the role of CI in aging health.

Conclusion

In this retrospective cross-sectional study, we estimate that 2.5 million adults in the United States age 18 or older meet newly-approved audiologic candidacy criteria for CI. We estimate that current CMS CI candidacy criteria will result in a candidate pool of approximately 2.1 million adults age 65 or older in the United States and the relative increase in size of the candidacy pool is associated with age. Despite this change in candidacy criteria, existing barriers to utilization of CI technology remain and further work is still needed to characterize CI utilization characteristics across both newly and previously qualified populations.