Abstract
Purpose of Review
The purpose of this review is to summarize candidacy criteria and commonly used referral guidelines for adult cochlear implant (CI) patients. This review describes how machine learning can be used to predict CI candidacy and the potential impact of an automated referral guideline.
Recent Findings
Less than 2% of eligible adults are receiving CIs under traditional and expanded candidacy criteria. Lack of understanding of referral criteria, both among providers and patients, significantly contributes to the underutilization of CIs. Recently, a novel machine learning-based CI referral algorithm has been developed that shows high sensitivity, specificity, and accuracy in predicting CI candidacy among adults.
Summary
An automated, machine learning-based referral guideline can mitigate the lack of clarity regarding when to refer a patient and help bridge the large gap in CI care delivery that currently exists. Future research needs to externally validate such an algorithm and evaluate its uptake in routine clinical settings.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
Introduction
Impacting nearly half a billion people globally, hearing loss can have devastating consequences on social interactions, quality of life, and cognitive function [1, 2]. Patients who have moderate-to-profound sensorineural hearing loss and get limited or no benefit from their hearing aids can benefit from cochlear implants (CIs). Given their success in improving speech perception, quality of life, and cognition, CIs have quickly become the standard of care [3,4,5,6]. However, this technology remains largely underutilized, partly due to continually evolving candidacy criteria and lack of clarity regarding when patients should be referred. Multiple screening tools have been developed to assist providers with their clinical decision-making [7•, 8,9,10,11,12]. More recently, a machine learning-based CI referral algorithm was developed to predict candidacy with high sensitivity, specificity, and accuracy [13•]. In this review, we summarize the recent literature on evolving candidacy criteria and commonly used referral guidelines for adult CI patients. Additionally, we describe how machine learning can be used to address gaps in the field.
Evolution of Adult Cochlear Implant Candidacy Criteria
The United States Food and Drug Administration (FDA) first approved CIs in 1987 for adults who had bilateral profound sensorineural hearing loss and no open-set speech recognition. In 1998, CIs were approved for adults who scored up to 40% on sentence testing. Today, to qualify under traditional CI criteria for most insurance programs in the United States including Medicare, adults must have bilateral moderate-to-severe sensorineural hearing loss. On sentence testing, they must score up to 50% correct in the ear to the implanted. Furthermore, they must demonstrate no more than 60% sentence recognition when both ears are optimally fit with hearing aids [14, 15].
With respect to expanded criteria, the FDA granted landmark approvals for hybrid/EAS devices in 2014. Patients with low-frequency thresholds in the normal range could now qualify for a CI, and word scores for each ear, rather than bilateral sentence scores, determined candidacy. In 2019, indications for CIs continued to expand, with approvals for select cases of single-sided deafness (SSD) and asymmetric hearing loss (AHL).
Cochlear Implant Utilization Rates Among Adults
Despite their known benefits, CIs are being used by less than 10% of eligible adults under traditional candidacy criteria. Utilization rates drop to 2% when further accounting for expanded criteria (e.g. patients with normal or near-normal low-frequency hearing, SSD, AHL) in the United States (Fig. 1) [16]. Given that over 7 million individuals with debilitating hearing loss are currently estimated to be untreated, significant gaps in CI care delivery are being identified and addressed within the field of otology/neurotology.
Estimated adult CI utilization rate of 2% under traditional and expanded candidacy criteria (data from Nassiri et al. 2022) [16]
Barriers to Care
A key barrier to care is limited knowledge of referral criteria among providers. At multiple centers, the majority of patients who present for CI evaluations have been found to have at least severe hearing loss and score far below the upper range of current candidacy criteria [17,18,19]. As the United States Preventive Services Task force does not recommend routine hearing screenings for adults, only 14.6% of primary care physicians are conducting any form of hearing screening (e.g. audiogram, tuning fork) [14, 20]. Even among general otolaryngologists, only 3% of 13,662 patients with bilateral moderate-to-profound sensorineural hearing loss have been referred for a CI evaluation. In addition, the proportion of patients who have met hybrid/EAS criteria during initial CI evaluations at Vanderbilt University Medical Center has remained similar from 2013–2015 (25%) relative to 2015–2020 (19%) [17, 21]. Such comparisons serve as markers for referring providers being aware of expanded criteria.
Patients also lack awareness around their hearing loss, CI-related benefits, and candidacy criteria. For instance, in the 2014 National Health Interview survey, a third of 40 million individuals who noted “a little trouble hearing” to being “deaf” had never seen a provider for their hearing impairment. Moreover, 28% of respondents acknowledged that they never had a formal hearing evaluation [22]. The majority of patients cannot identify the degree of their hearing loss so self-reported hearing status cannot be used as a reliable indicator of when to refer patients [23]. In a national survey of over 15,000 adults, 31% of respondents with hearing difficulty reported that they had “never heard” of a CI [24].
Other barriers to care include complex care pathways, demographic disparities like race, lack of social support, diminished cognitive function, insurance, and increased distance from a CI center. Innovative programs like the same-day CI evaluation and surgery, remote programming, and targeted outreach for disadvantaged groups (e.g. African Americans) are tackling some of these barriers [25, 26].
Current Screening Guidelines
In efforts to streamline the screening of potential candidates, several tools have been created that assist referring providers in their clinical decision-making. Zwolan et al. developed the 60/60 guideline, which recommends that patients with a pure-tone average (PTA) of at least 60 dB HL and word-recognition score (WRS) of 60% or below in the better hearing ear should be referred for a CI [7•]. Extracted from data for 529 patients at the University of Michigan, this guideline has 96% sensitivity and 66% specificity for traditional CI candidates. The 60/60 guideline is the most commonly used for screening patients and is recommended by entities like the American Cochlear Implant Alliance.
Other screening guidelines have varying thresholds for similar audiometric measures such as PTA of at least 70 dB HL and dissatisfaction with hearing aids and PTA of at least 75 dB HL and/or WRS up to 40% [11, 12]. So et al. found that cut-offs of 65 dB HL for PTA and 70% for WRS yielded the highest sensitivity, specificity, and accuracy in identifying CI candidates at their center [9]. Another screening procedure that has been reported indicates that CI referral can be considered if the difference between the PTA and maximum WRS exceeds 8, having a sensitivity of 87% and specificity of 91% for 185 subjects [10]. Going a step further, a logistic regression model using data from 252 patients has yielded a confusion matrix with sensitivities and specificities at various thresholds for unaided speech discrimination scores and PTA [8]. This methodology allows centers to choose their own cut-offs for identifying traditional CI patients.
When applying an external dataset, these screening guidelines are only modestly successful at identifying potential candidates [27]. The 60/60 guideline has been found to have the best balance of sensitivity and specificity among the major proposed screening criteria. However, these screening tools, including the 60/60 guideline, have been primarily designed for traditional hearing loss patients and exclude expanded criteria. They continue to pose complex guidelines and have rigid cut-offs which cannot easily evolve with the ever-changing landscape of adult CI indications.
Role of Machine Learning
Machine learning methods facilitate a non-linear approach for integrating large-scale heterogenous patient datasets. This aspect can be particularly useful when building a national CI referral guideline as testing can vary among practices. For instance, patients who receive an audiogram at Vanderbilt University Medical Center standardly have unaided air conduction thresholds at 125, 250, 500, 1000, 2000, 3000, 4000, 6000, and 8000 Hz. Other centers, where resources and time can be more limited, may only attain a subset of these thresholds. Machine learning thus offers an avenue for integrating varying patient-related metrics in order to build an optimal CI referral guideline that can be applied in multiple settings and limit disruption of a practice’s existing workflow.
In addition, machine learning allows for the building of an algorithm that can be iteratively improved especially as CI indications continue to evolve. The algorithm can be tailored on the backend by a team of experts, and referring providers and patients will not have to know or recall the latest criteria. An automated referral algorithm, where users simply have to input variables for an individual and a customized recommendation for whether to proceed with CI referral, thus mitigates critical barriers such as the lack of clarity surrounding candidacy criteria.
Application of Machine Learning to Cochlear Implant Candidacy Referrals
An interdisciplinary team at Vanderbilt University Medical Center developed a novel machine learning-based referral guideline for adult CI candidates in 2022 [13•]. A random forest classification model was trained on 587 patients who underwent CI evaluation at Vanderbilt between 2015 and 2020 and who also had complete preoperative audiometric (i.e. WRS, unaided air conduction thresholds) and demographic (i.e. age, insurance) data, the input variables for the model. Notably, the cohort used to build the machine learning-based referral algorithm included patients meeting traditional and expanded criteria. In comparison to the commonly used 60/60 guideline, the random forest model was found to have higher sensitivity, specificity, and accuracy in identifying CI candidates (Table 1). Bootstrap cross-validation confirmed that this approach was potentially generalizable with consistent results.
In the random forest model, preoperative WRS, preoperative audiometric thresholds at 3000, 2000, and 125 Hz, and age at CI evaluation were found to have the largest impact on candidacy. Current screening guidelines, including the 60/60, primarily use audiometric data. As more CI referrals will be made for individuals with moderate hearing loss, preserved low-frequency hearing, SSD, and AHL, insurance may play a larger factor in candidacy and should be considered in a referral guideline moving forward.
What Does the Future Hold?
Artificial intelligence (AI), which refers to technology that allows computers to perform tasks which typically require human intelligence, and machine learning, an application of AI, are terms that are now commonly used in the healthcare space. Within the field of otolaryngology, many clinicians have embraced technologies like voice recognition software for transcribing notes and AI-based triaging and scheduling tools. In a 2023 survey study, 82% of otolaryngologists reported using AI to assist with decision-making, and 74% were comfortable using AI-proposed treatment recommendations [28]. Publications on the intersection of AI and otolaryngology have also exponentially risen in the last several years [29•].
Specifically in otology and neurotology, several applications of AI and machine learning have been described. Volume-based measurements of vestibular schwannomas and predictions of tumor recurrence using AI have been shown to be superior compared to traditional statistical modeling [30, 31]. Moreover, AI can assist with diagnosis of middle ear pathology. For instance, when compared to certain groups of providers, a deep learning algorithm had higher accuracy in diagnosing pediatric tympanic membrane effusions (96% vs. 65%) [32]. AI has other applications to improve diagnosis and treatment of vestibular disorders [33].
While AI and machine learning have the potential to revolutionize many aspects of otolaryngology, caution must be applied. The derived algorithms are only as good as the quality and integrity of the data that are being used. For example, if a CI referral guideline is primarily built using data from traditional candidates, then a SSD or AHL patient may not be appropriately referred as the algorithm has not seen such examples. Furthermore, underlying biases can exist from physicians who have trained the AI algorithms. In one study, over 129,000 images of cutaneous lesions were utilized to train an AI-based model that differentiated malignancy from benign lesions, with the model’s performance being comparable to that of expert dermatologists [34]. On further review, the authors found that the algorithm identified malignancy at a higher rate if the image had a ruler. Underlying bias in the training data, where a malignant lesion was more likely to have a ruler, ultimately influenced the model’s output. Importantly, while AI and machine learning can have tremendous applications in medicine, they are not a replacement for clinicians. With respect to CI candidacy, future research needs to externally validate a machine learning-based referral algorithm using a diverse patient population from multiple centers and evaluate its uptake in routine clinical settings. This algorithm will be used as a screening tool, and potential CI candidates will still undergo appropriate audiologic and surgical evaluations prior to receiving their implant.
Conclusion
Large gaps in CI care delivery currently exist, with less than 2% of eligible adults receiving implants under traditional and expanded candidacy criteria. Lack of understanding of CI-related benefits and indications, both among providers and patients, significantly contributes to the underutilization of CIs. A recently developed novel machine learning-based CI referral algorithm has been developed that shows high sensitivity, specificity, and accuracy in predicting candidacy among adult patients. The use of machine learning helps automate the referral process and can mitigate the lack of clarity surrounding candidacy, especially as CI indications continue to evolve. Further studies are needed to externally validate machine learning-based algorithms and assess their outcomes and impact.
Data Availability
No datasets were generated or analysed during the current study.
References
Papers of particular interest, published recently, have been highlighted as: • Of importance
Cunningham LL, Tucci DL. Hearing loss in adults. N Engl J Med. 2017;377(25):2465–73. https://doi.org/10.1056/NEJMra1616601.
Livingston G, Huntley J, Sommerlad A, Ames D, Ballard C, Banerjee S, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. Lancet. 2020;396(10248):413–46. https://doi.org/10.1016/S0140-6736(20)30367-6.
Wilson BS, Dorman MF. Cochlear implants: a remarkable past and a brilliant future. Hear Res. 2008;242(1–2):3–21. https://doi.org/10.1016/j.heares.2008.06.005.
Tang L, Thompson CB, Clark JH, Ceh KM, Yeagle JD, Francis HW. Rehabilitation and psychosocial determinants of cochlear implant outcomes in older adults. Ear Hear. 2017;38(6):663–71. https://doi.org/10.1097/AUD.0000000000000445.
Gaylor JM, Raman G, Chung M, Lee J, Rao M, Lau J, Poe DS. Cochlear implantation in adults: a systematic review and meta-analysis. JAMA Otolaryngol Head Neck Surg. 2013;139(3):265–72. https://doi.org/10.1001/jamaoto.2013.1744.
Mosnier I, Bebear JP, Marx M, Fraysse B, Truy E, Lina-Granade G, et al. Improvement of cognitive function after cochlear implantation in elderly patients. JAMA Otolaryngol Head Neck Surg. 2015;141(5):442–50. https://doi.org/10.1001/jamaoto.2015.129.
• Zwolan TA, Schvartz-Leyzac KC, Pleasant T. Development of a 60/60 guideline for referring adults for a traditional cochlear implant candidacy evaluation. Otol Neurotol. 2020;41(7):895–900. https://doi.org/10.1097/MAO.0000000000002664. A retrospective study that led to the development of the 60/60 guideline.
Ngombu SJ, Ray C, Vasil K, Moberly AC, Varadarajan VV. Development of a novel screening tool for predicting cochlear implant candidacy. Laryngoscope Investig Otolaryngol. 2021;6(6):1406–13. https://doi.org/10.1002/lio2.673.
So RJ, Padova D, Bowditch S, Agrawal Y. Candidacy for cochlear implantation: validating a novel cochlear implant candidacy calculator against gold-standard, in-clinic audiometric assessments. Laryngoscope Investig Otolaryngol. 2022;7(3):835–9. https://doi.org/10.1002/lio2.804.
Hoppe U, Hast A, Hocke T. Audiometry-based screening procedure for cochlear implant candidacy. Otol Neurotol. 2015;36(6):1001–5. https://doi.org/10.1097/MAO.0000000000000730.
Gubbels SP, Gartrell BC, Ploch JL, Hanson KD. Can routine office-based audiometry predict cochlear implant evaluation results? Laryngoscope. 2017;127(1):216–22. https://doi.org/10.1002/lary.26066.
Lupo JE, Biever A, Kelsall DC. Comprehensive hearing aid assessment in adults with bilateral severe-profound sensorineural hearing loss who present for cochlear implant evaluation. Am J Otolaryngol. 2020;41(2):102300. https://doi.org/10.1016/j.amjoto.2019.102300.
• Patro A, Perkins EL, Ortega CA, Lindquist NR, Dawant BM, Gifford R, et al. Machine learning approach for screening cochlear implant candidates: comparing with the 60/60 guideline. Otol Neurotol. 2023;44(7):e486–91. https://doi.org/10.1097/MAO.0000000000003927. A large retrospective study that reported the first application of machine learning for screening cochlear implant candidates.
Sorkin DL. Cochlear implantation in the world’s largest medical device market: utilization and awareness of cochlear implants in the United States. Cochlear Implants Int. 2013;14(Suppl 1):S4-12. https://doi.org/10.1179/1467010013Z.00000000076.
Sorkin DL, Buchman CA. Cochlear implant access in six developed countries. Otol Neurotol. 2016;37(2):e161-4. https://doi.org/10.1097/MAO.0000000000000946.
Nassiri AM, Sorkin DL, Carlson ML. Current estimates of cochlear implant utilization in the United States. Otol Neurotol. 2022;43(5):e558–62. https://doi.org/10.1097/MAO.0000000000003513.
Patro A, Lindquist NR, Tawfik KO, O’Malley MR, Bennett ML, Haynes DS, et al. A five-year update on the profile of adults undergoing cochlear implant evaluation and surgery-are we ng better? Otol Neurotol. 2022;43(9):e992–9. https://doi.org/10.1097/MAO.0000000000003670.
Barnes JH, Yin LX, Marinelli JP, Carlson ML. Audiometric profile of cochlear implant recipients demonstrates need for revising insurance coverage. Laryngoscope. 2021;131(6):E2007-E. https://doi.org/10.1002/lary.29334.
Henkin Y, Shapira Y, Yaar Soffer Y. Current demographic and auditory profiles of adult cochlear implant candidates and factors affecting uptake. Int J Audiol. 2021. https://doi.org/10.1080/14992027.2021.1941327.
Force USPST, Krist AH, Davidson KW, Mangione CM, Cabana M, Caughey AB, et al. Screening for hearing loss in older adults: US preventive services task force recommendation statement. JAMA. 2021;325(12):1196–201. https://doi.org/10.1001/jama.2021.2566.
Holder JT, Reynolds SM, Sunderhaus LW, Gifford RH. Current profile of adults presenting for preoperative cochlear implant evaluation. Trends Hear. 2018;22:2331216518755288. https://doi.org/10.1177/2331216518755288.
Mahboubi H, Lin HW, Bhattacharyya N. Prevalence, characteristics, and treatment patterns of hearing difficulty in the United States. JAMA Otolaryngol Head Neck Surg. 2018;144(1):65–70. https://doi.org/10.1001/jamaoto.2017.2223.
Angara P, Tsang DC, Hoffer ME, Snapp HA. Self-perceived hearing status creates an unrealized barrier to hearing healthcare utilization. Laryngoscope. 2021;131(1):E289–95. https://doi.org/10.1002/lary.28604.
Marinelli JP, Sydlowski SA, Carlson ML. Cochlear implant awareness in the United States: a national survey of 15,138 adults. Semin Hear. 2022;43(4):317–23. https://doi.org/10.1055/s-0042-1758376.
Patro A, Haynes DS, Perkins EL. Same-day patient consultation and cochlear implantation: patient experiences and barriers to implementation. Otol Neurotol. 2022;43(8):e820–3. https://doi.org/10.1097/MAO.0000000000003627.
Sims S, Houston L, Schweinzger I, Samy RN. Closing the gap in cochlear implant access for African-Americans: a story of outreach and collaboration by our cochlear implant program. Curr Opin Otolaryngol Head Neck Surg. 2017;25(5):365–9. https://doi.org/10.1097/MOO.0000000000000399.
Lee DS, Herzog JA, Walia A, Firszt JB, Zhan KY, Durakovic N, et al. External validation of cochlear implant screening tools demonstrates modest generalizability. Otol Neurotol. 2022;43(9):e1000–7. https://doi.org/10.1097/MAO.0000000000003678.
Asokan A, Massey CJ, Tietbohl C, Kroenke K, Morris M, Ramakrishnan VR. Physician views of artificial intelligence in otolaryngology and rhinology: a mixed methods study. Laryngoscope Investig Otolaryngol. 2023;8(6):1468–75. https://doi.org/10.1002/lio2.1177.
• Bur AM, Shew M, New J. Artificial intelligence for the otolaryngologist: a state of the art review. Otolaryngol Head Neck Surg. 2019;160(4):603–11. https://doi.org/10.1177/0194599819827507. A literature review of the use of artificial intelligence in otolaryngology.
Cass ND, Lindquist NR, Zhu Q, Li H, Oguz I, Tawfik KO. Machine learning for automated calculation of vestibular schwannoma volumes. Otol Neurotol. 2022;43(10):1252–6. https://doi.org/10.1097/MAO.0000000000003687.
Abouzari M, Goshtasbi K, Sarna B, Khosravi P, Reutershan T, Mostaghni N, et al. Prediction of vestibular schwannoma recurrence using artificial neural network. Laryngoscope Investig Otolaryngol. 2020;5(2):278–85. https://doi.org/10.1002/lio2.362.
Crowson MG, Bates DW, Suresh K, Cohen MS, Hartnick CJ. “Human vs Machine” validation of a deep learning algorithm for pediatric middle ear infection diagnosis. Otolaryngol Head Neck Surg. 2023;169(1):41–6. https://doi.org/10.1177/01945998221119156.
Yeh SC, Huang MC, Wang PC, Fang TY, Su MC, Tsai PY, Rizzo A. Machine learning-based assessment tool for imbalance and vestibular dysfunction with virtual reality rehabilitation system. Comput Methods Programs Biomed. 2014;116(3):311–8. https://doi.org/10.1016/j.cmpb.2014.04.014.
Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM, Thrun S. Corrigendum: dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017;546(7660):686. https://doi.org/10.1038/nature22985.
Author information
Authors and Affiliations
Contributions
AP wrote the main manuscript text, and MHF and DSH edited and reviewed the manuscript.
Corresponding author
Ethics declarations
Conflict of Interest
The authors declare no competing interests.
Human and Animal Rights and Informed Consent
All reported studies/experiments with human or animal subjects performed by the authors have been previously published and complied with all applicable ethical standards (including the Helsinki declaration and its amendments, institutional/national research committee standards, and international/national/institutional guidelines).
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Patro, A., Freeman, M.H. & Haynes, D.S. Machine Learning to Predict Adult Cochlear Implant Candidacy. Curr Otorhinolaryngol Rep (2024). https://doi.org/10.1007/s40136-024-00511-7
Accepted:
Published:
DOI: https://doi.org/10.1007/s40136-024-00511-7