Otology & Neurotology 36:1001Y1005 Ó 2015, Otology & Neurotology, Inc.
Audiometry-Based Screening Procedure for Cochlear Implant Candidacy *Ulrich Hoppe, *Anne Hast, and †Thomas Hocke *Department of Audiology, ENT-Clinic, University of Erlangen-Nu¨rnberg, Erlangen; and ÞCochlear Deutschland GmbH & Co KG, Karl-Wiechert-Allee, Hannover, Germany
Objective: This study defines a screening procedure for cochlear implant (CI) candidacy in hearing aid users by using simple audiometric measures. Methods: Within this retrospective study, hearing aid performance and audiometric measures in 185 subjects (318 ears) were analyzed. By means of a linear Naive Bayes classifier, the pure-tone average and the maximum monosyllabic score (PBmax) were used to predict the aided monosyllabic word score and CI candidacy. Results: The two parameters PBmax and four-frequency hearing threshold average can be used to predict speech perception
with hearing aids with reasonable accuracy for screening purposes. The classification has a sensitivity of 87% and a specificity of 91%. The classification can be represented by a simple linear formula. Conclusion: CI candidacy can be predicted based on commonly used audiometric measures. Cochlear implant candidacy may be considered if the difference between the average pure-tone threshold (in decibels) and PBmax (in percent) exceeds 8. Key Words: CandidacyVCochlear implantsVHearing aidsVIndicationsVScreeningVSpeech audiometry. Otol Neurotol 36:1001Y1005, 2015.
Cochlear implantation is a very effective established treatment of profound hearing loss. However, the number of cochlear implant recipients is much lower than prevalence data would suggest. According to estimations of the World Health Organization, the prevalence of severe (61Y80 dB) to profound (980 dB) hearing loss for adults is about 1% worldwide, with some differences across regions (1). The utilization rate of cochlear implantation in the United States for children and adults is approximately 6% of the relevant hearing-impaired population (2). In Germany, the rate of cochlear implant (CI) in hearing-impaired child and adult populations is estimated to be on a similar scale (3,4), keeping in mind that this rate does greatly depend on the clinically used indications that have evolved during the last decade. The change of indications for cochlear implantation was based on improving performance outcomes with newer-generation CI systems and/or the increased experience with the implantation of recipients with a shorter phase of hearing deprivation (5Y8) and increased levels of residual hearing (9,10) as well as new therapeutic approaches (11Y13). For
ipsilateral considerations regarding cochlear implantation in terms of speech understanding, a critical score of about 30% aided monosyllabic speech score at a conversational level is widely accepted (6,14). Because of the ongoing expansion of audiologic criteria for cochlear implantation, the evaluation process of implant candidacy has increased in complexity. This change leads to some issues regarding the referral from hearing professionals (in nonimplanting institutions) to audiologic centers with experience in cochlear implantation. The causes for the low referral rates of hearing professionals are currently in discussion (2,15Y17). According to common practice in the field, audiologic referral criteria are often simplified down to the pure-tone audiogram. This might have been motivated by rather restricted and simple CI indications in the past. However, this is inappropriate for patients with comparable good hearing threshold but unacceptable low levels of speech understanding. In particular, this patient group with severe to profound sensorineural hearing loss tends to be unsatisfied with their performance for speech understanding with their hearing aid(s). For the general ENT physician, there are two ways to proceed with those patients. First option is the referral to a hearing aid acoustician. The second option is the referral to an audiologic center with experience in both implantable hearing solutions and conventional hearing aids. The decision pathway and ultimate choice of
Address correspondence and reprint requests to Prof. Dr. Ulrich Hoppe, Audiologische Abteilung, Hals-Nasen-Ohrenklinik, Kopf- und Halschirurgie, Universita¨tsklinikum Erlangen, Waldstr. 1, D-91054 Erlangen; E-mail: [email protected] T.H. is working for a cochlear implant company.
1001
Copyright © 2015 Otology & Neurotology, Inc. Unauthorized reproduction of this article is prohibited.
1002
U. HOPPE ET AL.
referral lead directly to the aim of the proposal described in this article. Our aim was the development of a simple referral criterion for general ENT use to support the decision and choice of referral path for an individual hearing-impaired patient either to a cochlear implant center or for ongoing audiologic rehabilitation with conventional hearing aids. At this point, a time-saving and easy-to-use screening procedure is needed. This screening procedure has to be based on audiometric measures that are independent of hearing aid fitting and do not require extensive technical capabilities. According to recent findings on the relationship between hearing thresholds and speech perception abilities (18,19), the screening should incorporate both measures. We propose the assessment of two basic audiologic measures, namely, the maximum speech understanding for monosyllabics (PBmax) and the four-frequency hearing threshold average (4FPTA), as a reasonable basis for a decision on the correct referral path.
metric hearing thresholds exceeding 20 dB difference for the 4FPTA, the contralateral ear was masked via a headphone using speech-shaped noise with a level of 30 dB below the actual speech level. In a first step, the SRSaided at 65 dB was plotted in dependence of the pure-tone loss represented by the 4FPTA. By fitting a sigmoid function, a formula could be derived that describes roughly the SRSaided as a function of pure-tone loss. In a second step, the data were used to predict the possible candidacy for CI based on 4FPTA and PBmax only. By applying the 30% criterion for monosyllabic score for the aided situation, the ears were classified into two groups in the two-dimensional plane spanned by the 4FPTA and the PBmax. The first group has an SRSaided less than 30% and therefore CI candidates. The second group has an SRSaided better than 30% and therefore are provided sufficiently with conventional hearing aids. The classification was realized via linear Naive Bayes classifier. For all calculations, the MathWorks Matlab software was used.
RESULTS METHODS For this study, audiometric measurements of 185 sensorineural hearing-impaired subjects were evaluated retrospectively. Patients were referred to the audiologic center at the ENT Department Erlangen between January 2011 and February 2013 for hearing assessment and hearing aid evaluation. The data were collected as a part of clinical routine for hearing aid performance assessment. All subjects had been supplied with hearing aids and had at least 3 months’ experience using them. Approximately 50% of the subjects were supplied with hearing aids for the first time. All had completed the hearing aid selection and fitting process at least 3 months before the visit. Before aided hearing assessment, the hearing aids were technically checked. All patients attended a follow-up visit with their hearing aid professional before the visit. Only native German-speaking subjects with no known mental disabilities influencing audiometric tests were included in the analysis. Subjects with air-bone gaps of more than 5 dB were excluded from the analysis. The subject group consists of 92 males and 93 females, the age at assessment ranged from 17 to 92 years, with a mean age of 61 years. Of the 185 subjects, 133 used hearing aids bilaterally and 52 used a hearing aid unilaterally. In total, data of 318 ears were analyzed. Each ear was tested separately. The pure-tone average threshold for the frequencies 0.5, 1, 2, and 4 kHz (4FPTA) was evaluated for each ear for air conduction. Speech audiometry using Freiburg monosyllabic words was measured via headphones in the unaided condition for each ear. The speech stimuli were presented at 65 dB HL initially, and the corresponding speech recognition score (SRSunaided), correct word score in percent, was recorded. The presentation level was increased stepwise by 10 to 15 dB until a maximum score of 100% was achieved or the maximum score achievable (PBmax) below the patient’s uncomfortable loudness level was reached. The SRS with hearing aids (SRSaided) was measured in an anechoic soundproof booth, 6 6 m, with frontal sound stimulus incidence at a presentation level of 65 dB SPL via loudspeaker at 1.5 m from the patient. The measurements were performed monaurally with the contralateral ear occluded by a hearing protector (Ohropax soft). In case of interaural asym-
Figure 1 shows the aided monosyllabic SRS with a hearing aid in the free-field at a presentation level of 65 dB SPL as a function of the 4FPTA. There is a strong correlation especially in the range of moderate hearing losses. For a more convenient view of the available data in Figure 1A, a summary is presented in box plots in Figure 1B using class widths of 10 dB. In the vicinity of a 4FPTA of 60 dB HL, the largest variation was observed. At this 4FPTA, a quarter of the subjects do not exceed an SRSaided of 30%. At lower 4FPTA values, the vast majority of data are above 30% SRSaided. The SRSaided can be fitted by a sigmoid function, equation 1:
SRS aided ½% ¼ 100
eðA0 þA1 q4FPTAÞ 1 þ eðA0 þA1 q4FPTAÞ
ð1Þ
The parameters are A0=3,98 T 0,05 and A1=j0,0661 T 0,0008/dB. Figure 2 shows the maximum speech scores for Freiburg monosyllablic words (PBmax) measured in speech audiometry with circumaural headphones. The median presentation level was 100 dB SPL. The presentation level to achieve the maximum score ranged from 65 dB SPL for mild hearing losses to 130 dB SPL for profound hearing losses. A Naive Bayes classifier (20) was used for classification of two categories. The first category represents subjects (ears separately) expected to have a free-field SRSaided of less than 30% and therefore classified as potential CI candidates. The second category represents subjects with sufficient benefit from hearing aid use, which means an SRSaided equal to or above 30%. The prediction is based on the 4FPTA and the PBmax. The linear equation for the Naive Bayes classifier is given by equation 2. This can be rewritten as a criterion for the referral of a hearing aid user to a CI center,
Otology & Neurotology, Vol. 36, No. 6, 2015
Copyright © 2015 Otology & Neurotology, Inc. Unauthorized reproduction of this article is prohibited.
CI CANDIDACY AUDIOMETRY-BASED SCREENING PROCEDURE
1003
subjects, indicating a number of different influencing key factors. The particular type of cochlear damage plays an important role in the benefit from aided hearing (18). The presence of dead regions (19) within the cochlea, the slope of the audiogram, and the specific damage of the inner or outer hair cells will also influence speech perception. As pointed out by Halpin and Rauch (18), hearing aid benefit is limited by the cochlear damage that is reliably characterized by PBmax. Accordingly, the PBmax should be considered as the individual’s potential upper limit of hearing aid performance rather than an always achievable goal (21). With conventional hearing aid amplification, the desired PBmax to enable good speech recognition for conversational speech cannot be achieved by a considerable number of individuals. Only 40% of the hearing aid patients investigated by Hoppe et al. (21) achieved a performance within a 10% range near the desired PBmax. One of the limiting factors for successful amplification is the residual hearing dynamic range. Zwartenkot et al. (22) underlined this importance and showed that a minimum of 35 dB dynamic range is required for sufficient everyday listening benefit. In particular, for severe to profound hearing loss, the PBmax is achieved at levels quite near to the UCL. This possibly limits the theoretical potential of the information-carrying capacity (18) in everyday life, given that the PBmax is measured with headphones at high levels for only a short period in an otherwise quiet environment. But a hearing aid fitting must provide whole-day comfort even under more difficult and challenging noisy conditions. FIG. 1. Speech recognition scores for Freiburg monosyllablic words (A) measured monaurally (N = 318) with hearing aid in the free field at a presentation level of 65 dB SPL. The abscissa represents the average hearing loss (0.5, 1, 2, 4 kHz). In addition, the fitted sigmoid function is presented. The graph below (B) summarizes the data of the above figure (A) via box plots. The data are grouped with regard to the hearing loss range for the ear. The class widths of the boxes are 10 dB, covering the range from 25 dB HL to 95 dB HL. The numbers within the boxes represent the size of the corresponding group.
equation 3. If the left-hand side of equation 3 exceeds a critical value of 0.8855, the patient should be referred. % ð2Þ PB max ¼ 1:1299 q4FPTAj16:63% dB 4FPTA ½dBj14:72 90:8855 PB max ½%
ð3Þ
Based on the training data, the Naive Bayes classifier yields a sensitivity of 87% while having a specificity of 91%. The misclassification rate is 11%, the positive predictive value is 77%, and the negative predictive value is 95%. DISCUSSION Hearing Aid Amplification The benefit of amplification with conventional hearing aids for hearing-impaired subjects varies greatly between
FIG. 2. Maximum speech recognition scores (PBmax) for Freiburg monosyllablic words measured monaurally with circumaural headphones. The abscissa represents the average hearing loss (0.5, 1, 2, 4 kHz). All subjects (ears separately measured) who achieved a score less than 30% in the free-field condition with hearing aids at 65 dB SPL (data in Fig. 1) are represented by downward-facing triangles. All subjects (ears) achieving a score of 30% or better are represented by circles. The thick black line shows the result of linear Bayes classification, equation 2. Based on this classification, the area below the lines represents the range of being a possible CI candidate; the above area represents the range of a hearing aid candidate. For the study population, the linear Bayes classification achieves a sensitivity of 87%, together with a specificity of 91%. The thin black line represents a simplified classifier according to equation 4. Otology & Neurotology, Vol. 36, No. 6, 2015
Copyright © 2015 Otology & Neurotology, Inc. Unauthorized reproduction of this article is prohibited.
1004
U. HOPPE ET AL.
An additional factor that may influence the results as already discussed (21) is the inherent adverse bias in the investigated group because it seems reasonable to assume that the majority of patients visiting the hearing center do so because they are having trouble with their hearing aids. Other rather technical factors include the limited amplification caused by feedback, wearing comfort of ear molds, and others. Transient suppression, feedback management, and compression may provide greater comfort and ease of use but could also compromise speech perception. As a practical consequence of all limiting factors, the SRSaided at 65 dB SPL is on average 20% points below the PBmax (21). The wide variation of hearing aid benefit relative to the average hearing thresholds alone is thus a result of the different contribution of thee various influencing factors on an individual basis. Use of Monosyllabic Speech Scores and Pure-Tone Thresholds for Screening Halpin and Rauch (18) show that even a simple monosyllabic word recognition test allows for clinical conclusions regarding the success of amplification. The results of classification, shown in Figure 2, are evidently related to the classification target, the minimum SRS, which should be achievable by a hearing aid user for speech stimuli presented in the free field at 65 dB SPL. We used a criterion of 30% for Freiburg monosyllables. This target is often referred to as an audiologic indication criteria or borderline candidacy (6,14). In Germany, although it is not explicitly published by either the national health administration or the cochlear implant guideline committee, a critical score of 30% has been applied in common clinical practice in the last decade for referral for further assessment by a CI center. The applied Bayes classification for the 30% target is based on simple audiometric measures. The general ENT physicians have a reasonable referral criterion for the hearing aid patient to proceed with hearing aid therapy (referral to hearing aid acoustician) or to consult a hearing center with a comprehensive range of possible alternative therapies like cochlear implantation. Ears or Individuals? Latest findings such as cochlear implantation in patients with asymmetric hearing loss and single-sided deafness (11Y13) suggest the separate evaluation of the subjects’ ears. For this reason, the data for the 185 subjects were examined and analyzed as 318 individual ears. This may limit the interpretation of the terms sensitivity and specificity used in this study because a screening procedure should refer to subjects not to ears. Nevertheless, the monaural approach reflects current practice in CI candidacy in Germany and some other countries providing CI to subjects with asymmetric hearing loss and single-sided deafness. Candidacy Criterion The good results of the Naive Bayes classification, sensitivity 87% and specificity of 91%, lead to quite acceptable positive and negative predictive values for our patient group reported as 77% and 95%, respectively. To a small but
certain degree, the influence of the preselection of subjects remains unknown. However, our patient group covers a broad range of degrees of hearing loss with a reasonable distribution in the region of interest. For practical purposes, the referral criteria from equation 3 might be simplified. The following formula: PB max ½%G 4FPTA ½dBj8
ð4Þ
is much easier to handle, with negligible influence on classification. The corresponding linear separation is denoted as the thin line shown in Figure 2. Obviously, one data point is affected only. The use of this inequation is explained by the two examples. For a subject with a PBmax = 60%, a referral is recommended when the 4FPTA is above 68 dB HL; for a subject with a PBmax = 50%, a referral is already recommended when the 4FPTA is 58 dB HL or larger. The two parameters PBmax and 4FPTA can be used to predict speech perception with hearing aids, sufficiently. Large values for PBmax do not necessarily provide sufficient hearing aid outcome when the 4FPTA is above a certain level. On the other hand, a very low 4FPTA is also not sufficient for good hearing aid outcome if the PBmax is low. CONCLUSION An easy-to-use criterion for the referral of unsatisfied hearing aid users consulting a general ENT can be applied. If the expression PBmax[%]G4FPTA [dB]j8 is valid, the patient should be referred to a CI center for further diagnostics. This decision can be based on very simple audiometric data, namely, the averaged hearing threshold at four frequencies, 4FPTA, and the measurement of the maximum SRS via audiometer under headphones, PBmax, without any hearing aid testing. The high positive predictive value of this classification enables CI centers to handle this workload. The decision process for referral proposed in this study must not be confused with CI indication, a complex process that cannot be reduced down to a few parameters. REFERENCES 1. Mathers C, Smith A, Concha M. Global burden of hearing loss in the year 2000. WHO health data and statistics. Available at: http://www.who.int/healthinfo/statistics/bod_hearingloss.pdf. Accessed February 25, 2014. 2. Sorkin DL. Cochlear implantation in the world’s largest medical device market: utilization and awareness of cochlear implants in the United States. Cochl Impl Int 2013;14:4Y12. 3. Sohn W, Jo¨rgenshaus W. Schwerho¨rigkeit in Deutschland, Repra¨sentative Ho¨rscreening-Untersuchung bei 2000 Probanden in 11 Allgemeinpraxen. Z All Med 2001;77:S143. 4. Teschner M, Polite C, Lenarz T, Lustig L. Cochlear implantation in different health-care systems: disparities between Germany and the United States. Otol Neurotol 2013;34:66Y74. 5. Kru¨ger B, Gert J, Rost U, et al. Performance groups in adults cochlear implant users: speech perception results from 1984 until today. Otol Neurotol 2008;29:509Y12.
Otology & Neurotology, Vol. 36, No. 6, 2015
Copyright © 2015 Otology & Neurotology, Inc. Unauthorized reproduction of this article is prohibited.
CI CANDIDACY AUDIOMETRY-BASED SCREENING PROCEDURE 6. Gifford RH, Dorman MF, Shallop JK, et al. Evidence for the expansion of adult cochlear implant candidacy. Ear Hear 2010;31:186Y94. 7. Blamey P, Artieres F, Bazkent D, et al. Factors affecting auditory performance of postlinguistically deaf adults using cochlear implants: an update with 2251 patients. Audiol Neurotol 2013;18:36Y47. 8. Holden LK, Finley CC, Firszt JB, et al. Factors affecting open-set word recognition in adults with cochlear implants. Ear Hear 2013; 34:342Y60. 9. Dettman SJ, D’Costa WA, Dowell RC, et al. Cochlear implants for children with significant residual hearing. Arch Otolaryngol Head Neck Surg 2004;130:612Y8. 10. Lenarz T, Sto¨ver T, Bu¨chner A, et al. Hearing conservation surgery using the hybrid-1 electrode. Results from the first clinical trial at the medical university of Hanover. Audiol Neurotol 2009; 14:22Y31. 11. Vermeire K, Nobbe A, Schleich P, et al. Neural tonotopy in cochlear implants: an evaluation in unilateral cochlear implant patients with unilateral deafness and tinnitus. Hear Res 2008;245:98Y106. 12. Arndt S, Aschendorff A, Laszig R, et al. Comparison of pseudobinaural hearing to real binaural hearing rehabilitation after cochlear implantation in patients with unilateral deafness and tinnitus. Otol Neurotol 2011;32:39Y47. 13. Hassepass F, Aschendorff A, Wesarg T, et al. Unilateral deafness in children: audiologic and subjective assessment of hearing
14.
15. 16.
17. 18.
19. 20. 21. 22.
1005
ability after cochlear implantation. Otol Neurotol 2013;34: 53Y60. De Raeve L, Wouters J. Accessibliity to cochlear implants in Belgium: state of the art on selectio, reimbursement, habilitation, and outcomes in children and adults. Cochl Impl Int 2013;14:18Y25. Chundu S, Flynn SL. Audiogram and cochlear implant candidacyV UK perspective. Cochl Impl Int 2014;15:241Y4. Chundu S, Buhagiar R. Audiologists’ knowledge of cochlear implant and their related referrals to the cochlear implant centre: pilot study findings from UK. Cochl Impl Int 2013;14:9Y13. Kielhorn H, Scho¨nermark MP. Strategische Rahmenbedingungen fu¨r die Versorgung mit Cochleaimplantaten. HNO 2014;62:214Y8. Halpin C, Rauch SD. Clinical implications of a damaged cochlea: pure-tone thresholds vs information carrying capacity. Otolaryngol Head Neck Surg 2008;140:473Y6. Moore BC. Dead regions in the cochlea: conceptual foundations, diagnosis, and clinical applications. Ear Hear 2004;25:98Y116. Niemann H. Pattern Analysis and Understanding. 2nd ed. Berlin, Germany: Springer, 2013. Hoppe U, Hast A, Hocke T. Sprachverstehen mit Ho¨rgeraten in Abha¨ngigkeit vom Tongeho¨r. HNO 2014;62:443Y8. Zwartenkot JW, Snik AF, Mylanus EA, et al. Amplification options for patients with mixed hearing loss. Otol Neurotol 2014;35:221Y6.
Otology & Neurotology, Vol. 36, No. 6, 2015
Copyright © 2015 Otology & Neurotology, Inc. Unauthorized reproduction of this article is prohibited.
Audiometry-Based Screening Procedure for Cochlear Implant Candidacy *Ulrich Hoppe, *Anne Hast, and †Thomas Hocke *Department of Audiology, ENT-Clinic, University of Erlangen-Nu¨rnberg, Erlangen; and ÞCochlear Deutschland GmbH & Co KG, Karl-Wiechert-Allee, Hannover, Germany
Objective: This study defines a screening procedure for cochlear implant (CI) candidacy in hearing aid users by using simple audiometric measures. Methods: Within this retrospective study, hearing aid performance and audiometric measures in 185 subjects (318 ears) were analyzed. By means of a linear Naive Bayes classifier, the pure-tone average and the maximum monosyllabic score (PBmax) were used to predict the aided monosyllabic word score and CI candidacy. Results: The two parameters PBmax and four-frequency hearing threshold average can be used to predict speech perception
with hearing aids with reasonable accuracy for screening purposes. The classification has a sensitivity of 87% and a specificity of 91%. The classification can be represented by a simple linear formula. Conclusion: CI candidacy can be predicted based on commonly used audiometric measures. Cochlear implant candidacy may be considered if the difference between the average pure-tone threshold (in decibels) and PBmax (in percent) exceeds 8. Key Words: CandidacyVCochlear implantsVHearing aidsVIndicationsVScreeningVSpeech audiometry. Otol Neurotol 36:1001Y1005, 2015.
Cochlear implantation is a very effective established treatment of profound hearing loss. However, the number of cochlear implant recipients is much lower than prevalence data would suggest. According to estimations of the World Health Organization, the prevalence of severe (61Y80 dB) to profound (980 dB) hearing loss for adults is about 1% worldwide, with some differences across regions (1). The utilization rate of cochlear implantation in the United States for children and adults is approximately 6% of the relevant hearing-impaired population (2). In Germany, the rate of cochlear implant (CI) in hearing-impaired child and adult populations is estimated to be on a similar scale (3,4), keeping in mind that this rate does greatly depend on the clinically used indications that have evolved during the last decade. The change of indications for cochlear implantation was based on improving performance outcomes with newer-generation CI systems and/or the increased experience with the implantation of recipients with a shorter phase of hearing deprivation (5Y8) and increased levels of residual hearing (9,10) as well as new therapeutic approaches (11Y13). For
ipsilateral considerations regarding cochlear implantation in terms of speech understanding, a critical score of about 30% aided monosyllabic speech score at a conversational level is widely accepted (6,14). Because of the ongoing expansion of audiologic criteria for cochlear implantation, the evaluation process of implant candidacy has increased in complexity. This change leads to some issues regarding the referral from hearing professionals (in nonimplanting institutions) to audiologic centers with experience in cochlear implantation. The causes for the low referral rates of hearing professionals are currently in discussion (2,15Y17). According to common practice in the field, audiologic referral criteria are often simplified down to the pure-tone audiogram. This might have been motivated by rather restricted and simple CI indications in the past. However, this is inappropriate for patients with comparable good hearing threshold but unacceptable low levels of speech understanding. In particular, this patient group with severe to profound sensorineural hearing loss tends to be unsatisfied with their performance for speech understanding with their hearing aid(s). For the general ENT physician, there are two ways to proceed with those patients. First option is the referral to a hearing aid acoustician. The second option is the referral to an audiologic center with experience in both implantable hearing solutions and conventional hearing aids. The decision pathway and ultimate choice of
Address correspondence and reprint requests to Prof. Dr. Ulrich Hoppe, Audiologische Abteilung, Hals-Nasen-Ohrenklinik, Kopf- und Halschirurgie, Universita¨tsklinikum Erlangen, Waldstr. 1, D-91054 Erlangen; E-mail: [email protected] T.H. is working for a cochlear implant company.
1001
Copyright © 2015 Otology & Neurotology, Inc. Unauthorized reproduction of this article is prohibited.
1002
U. HOPPE ET AL.
referral lead directly to the aim of the proposal described in this article. Our aim was the development of a simple referral criterion for general ENT use to support the decision and choice of referral path for an individual hearing-impaired patient either to a cochlear implant center or for ongoing audiologic rehabilitation with conventional hearing aids. At this point, a time-saving and easy-to-use screening procedure is needed. This screening procedure has to be based on audiometric measures that are independent of hearing aid fitting and do not require extensive technical capabilities. According to recent findings on the relationship between hearing thresholds and speech perception abilities (18,19), the screening should incorporate both measures. We propose the assessment of two basic audiologic measures, namely, the maximum speech understanding for monosyllabics (PBmax) and the four-frequency hearing threshold average (4FPTA), as a reasonable basis for a decision on the correct referral path.
metric hearing thresholds exceeding 20 dB difference for the 4FPTA, the contralateral ear was masked via a headphone using speech-shaped noise with a level of 30 dB below the actual speech level. In a first step, the SRSaided at 65 dB was plotted in dependence of the pure-tone loss represented by the 4FPTA. By fitting a sigmoid function, a formula could be derived that describes roughly the SRSaided as a function of pure-tone loss. In a second step, the data were used to predict the possible candidacy for CI based on 4FPTA and PBmax only. By applying the 30% criterion for monosyllabic score for the aided situation, the ears were classified into two groups in the two-dimensional plane spanned by the 4FPTA and the PBmax. The first group has an SRSaided less than 30% and therefore CI candidates. The second group has an SRSaided better than 30% and therefore are provided sufficiently with conventional hearing aids. The classification was realized via linear Naive Bayes classifier. For all calculations, the MathWorks Matlab software was used.
RESULTS METHODS For this study, audiometric measurements of 185 sensorineural hearing-impaired subjects were evaluated retrospectively. Patients were referred to the audiologic center at the ENT Department Erlangen between January 2011 and February 2013 for hearing assessment and hearing aid evaluation. The data were collected as a part of clinical routine for hearing aid performance assessment. All subjects had been supplied with hearing aids and had at least 3 months’ experience using them. Approximately 50% of the subjects were supplied with hearing aids for the first time. All had completed the hearing aid selection and fitting process at least 3 months before the visit. Before aided hearing assessment, the hearing aids were technically checked. All patients attended a follow-up visit with their hearing aid professional before the visit. Only native German-speaking subjects with no known mental disabilities influencing audiometric tests were included in the analysis. Subjects with air-bone gaps of more than 5 dB were excluded from the analysis. The subject group consists of 92 males and 93 females, the age at assessment ranged from 17 to 92 years, with a mean age of 61 years. Of the 185 subjects, 133 used hearing aids bilaterally and 52 used a hearing aid unilaterally. In total, data of 318 ears were analyzed. Each ear was tested separately. The pure-tone average threshold for the frequencies 0.5, 1, 2, and 4 kHz (4FPTA) was evaluated for each ear for air conduction. Speech audiometry using Freiburg monosyllabic words was measured via headphones in the unaided condition for each ear. The speech stimuli were presented at 65 dB HL initially, and the corresponding speech recognition score (SRSunaided), correct word score in percent, was recorded. The presentation level was increased stepwise by 10 to 15 dB until a maximum score of 100% was achieved or the maximum score achievable (PBmax) below the patient’s uncomfortable loudness level was reached. The SRS with hearing aids (SRSaided) was measured in an anechoic soundproof booth, 6 6 m, with frontal sound stimulus incidence at a presentation level of 65 dB SPL via loudspeaker at 1.5 m from the patient. The measurements were performed monaurally with the contralateral ear occluded by a hearing protector (Ohropax soft). In case of interaural asym-
Figure 1 shows the aided monosyllabic SRS with a hearing aid in the free-field at a presentation level of 65 dB SPL as a function of the 4FPTA. There is a strong correlation especially in the range of moderate hearing losses. For a more convenient view of the available data in Figure 1A, a summary is presented in box plots in Figure 1B using class widths of 10 dB. In the vicinity of a 4FPTA of 60 dB HL, the largest variation was observed. At this 4FPTA, a quarter of the subjects do not exceed an SRSaided of 30%. At lower 4FPTA values, the vast majority of data are above 30% SRSaided. The SRSaided can be fitted by a sigmoid function, equation 1:
SRS aided ½% ¼ 100
eðA0 þA1 q4FPTAÞ 1 þ eðA0 þA1 q4FPTAÞ
ð1Þ
The parameters are A0=3,98 T 0,05 and A1=j0,0661 T 0,0008/dB. Figure 2 shows the maximum speech scores for Freiburg monosyllablic words (PBmax) measured in speech audiometry with circumaural headphones. The median presentation level was 100 dB SPL. The presentation level to achieve the maximum score ranged from 65 dB SPL for mild hearing losses to 130 dB SPL for profound hearing losses. A Naive Bayes classifier (20) was used for classification of two categories. The first category represents subjects (ears separately) expected to have a free-field SRSaided of less than 30% and therefore classified as potential CI candidates. The second category represents subjects with sufficient benefit from hearing aid use, which means an SRSaided equal to or above 30%. The prediction is based on the 4FPTA and the PBmax. The linear equation for the Naive Bayes classifier is given by equation 2. This can be rewritten as a criterion for the referral of a hearing aid user to a CI center,
Otology & Neurotology, Vol. 36, No. 6, 2015
Copyright © 2015 Otology & Neurotology, Inc. Unauthorized reproduction of this article is prohibited.
CI CANDIDACY AUDIOMETRY-BASED SCREENING PROCEDURE
1003
subjects, indicating a number of different influencing key factors. The particular type of cochlear damage plays an important role in the benefit from aided hearing (18). The presence of dead regions (19) within the cochlea, the slope of the audiogram, and the specific damage of the inner or outer hair cells will also influence speech perception. As pointed out by Halpin and Rauch (18), hearing aid benefit is limited by the cochlear damage that is reliably characterized by PBmax. Accordingly, the PBmax should be considered as the individual’s potential upper limit of hearing aid performance rather than an always achievable goal (21). With conventional hearing aid amplification, the desired PBmax to enable good speech recognition for conversational speech cannot be achieved by a considerable number of individuals. Only 40% of the hearing aid patients investigated by Hoppe et al. (21) achieved a performance within a 10% range near the desired PBmax. One of the limiting factors for successful amplification is the residual hearing dynamic range. Zwartenkot et al. (22) underlined this importance and showed that a minimum of 35 dB dynamic range is required for sufficient everyday listening benefit. In particular, for severe to profound hearing loss, the PBmax is achieved at levels quite near to the UCL. This possibly limits the theoretical potential of the information-carrying capacity (18) in everyday life, given that the PBmax is measured with headphones at high levels for only a short period in an otherwise quiet environment. But a hearing aid fitting must provide whole-day comfort even under more difficult and challenging noisy conditions. FIG. 1. Speech recognition scores for Freiburg monosyllablic words (A) measured monaurally (N = 318) with hearing aid in the free field at a presentation level of 65 dB SPL. The abscissa represents the average hearing loss (0.5, 1, 2, 4 kHz). In addition, the fitted sigmoid function is presented. The graph below (B) summarizes the data of the above figure (A) via box plots. The data are grouped with regard to the hearing loss range for the ear. The class widths of the boxes are 10 dB, covering the range from 25 dB HL to 95 dB HL. The numbers within the boxes represent the size of the corresponding group.
equation 3. If the left-hand side of equation 3 exceeds a critical value of 0.8855, the patient should be referred. % ð2Þ PB max ¼ 1:1299 q4FPTAj16:63% dB 4FPTA ½dBj14:72 90:8855 PB max ½%
ð3Þ
Based on the training data, the Naive Bayes classifier yields a sensitivity of 87% while having a specificity of 91%. The misclassification rate is 11%, the positive predictive value is 77%, and the negative predictive value is 95%. DISCUSSION Hearing Aid Amplification The benefit of amplification with conventional hearing aids for hearing-impaired subjects varies greatly between
FIG. 2. Maximum speech recognition scores (PBmax) for Freiburg monosyllablic words measured monaurally with circumaural headphones. The abscissa represents the average hearing loss (0.5, 1, 2, 4 kHz). All subjects (ears separately measured) who achieved a score less than 30% in the free-field condition with hearing aids at 65 dB SPL (data in Fig. 1) are represented by downward-facing triangles. All subjects (ears) achieving a score of 30% or better are represented by circles. The thick black line shows the result of linear Bayes classification, equation 2. Based on this classification, the area below the lines represents the range of being a possible CI candidate; the above area represents the range of a hearing aid candidate. For the study population, the linear Bayes classification achieves a sensitivity of 87%, together with a specificity of 91%. The thin black line represents a simplified classifier according to equation 4. Otology & Neurotology, Vol. 36, No. 6, 2015
Copyright © 2015 Otology & Neurotology, Inc. Unauthorized reproduction of this article is prohibited.
1004
U. HOPPE ET AL.
An additional factor that may influence the results as already discussed (21) is the inherent adverse bias in the investigated group because it seems reasonable to assume that the majority of patients visiting the hearing center do so because they are having trouble with their hearing aids. Other rather technical factors include the limited amplification caused by feedback, wearing comfort of ear molds, and others. Transient suppression, feedback management, and compression may provide greater comfort and ease of use but could also compromise speech perception. As a practical consequence of all limiting factors, the SRSaided at 65 dB SPL is on average 20% points below the PBmax (21). The wide variation of hearing aid benefit relative to the average hearing thresholds alone is thus a result of the different contribution of thee various influencing factors on an individual basis. Use of Monosyllabic Speech Scores and Pure-Tone Thresholds for Screening Halpin and Rauch (18) show that even a simple monosyllabic word recognition test allows for clinical conclusions regarding the success of amplification. The results of classification, shown in Figure 2, are evidently related to the classification target, the minimum SRS, which should be achievable by a hearing aid user for speech stimuli presented in the free field at 65 dB SPL. We used a criterion of 30% for Freiburg monosyllables. This target is often referred to as an audiologic indication criteria or borderline candidacy (6,14). In Germany, although it is not explicitly published by either the national health administration or the cochlear implant guideline committee, a critical score of 30% has been applied in common clinical practice in the last decade for referral for further assessment by a CI center. The applied Bayes classification for the 30% target is based on simple audiometric measures. The general ENT physicians have a reasonable referral criterion for the hearing aid patient to proceed with hearing aid therapy (referral to hearing aid acoustician) or to consult a hearing center with a comprehensive range of possible alternative therapies like cochlear implantation. Ears or Individuals? Latest findings such as cochlear implantation in patients with asymmetric hearing loss and single-sided deafness (11Y13) suggest the separate evaluation of the subjects’ ears. For this reason, the data for the 185 subjects were examined and analyzed as 318 individual ears. This may limit the interpretation of the terms sensitivity and specificity used in this study because a screening procedure should refer to subjects not to ears. Nevertheless, the monaural approach reflects current practice in CI candidacy in Germany and some other countries providing CI to subjects with asymmetric hearing loss and single-sided deafness. Candidacy Criterion The good results of the Naive Bayes classification, sensitivity 87% and specificity of 91%, lead to quite acceptable positive and negative predictive values for our patient group reported as 77% and 95%, respectively. To a small but
certain degree, the influence of the preselection of subjects remains unknown. However, our patient group covers a broad range of degrees of hearing loss with a reasonable distribution in the region of interest. For practical purposes, the referral criteria from equation 3 might be simplified. The following formula: PB max ½%G 4FPTA ½dBj8
ð4Þ
is much easier to handle, with negligible influence on classification. The corresponding linear separation is denoted as the thin line shown in Figure 2. Obviously, one data point is affected only. The use of this inequation is explained by the two examples. For a subject with a PBmax = 60%, a referral is recommended when the 4FPTA is above 68 dB HL; for a subject with a PBmax = 50%, a referral is already recommended when the 4FPTA is 58 dB HL or larger. The two parameters PBmax and 4FPTA can be used to predict speech perception with hearing aids, sufficiently. Large values for PBmax do not necessarily provide sufficient hearing aid outcome when the 4FPTA is above a certain level. On the other hand, a very low 4FPTA is also not sufficient for good hearing aid outcome if the PBmax is low. CONCLUSION An easy-to-use criterion for the referral of unsatisfied hearing aid users consulting a general ENT can be applied. If the expression PBmax[%]G4FPTA [dB]j8 is valid, the patient should be referred to a CI center for further diagnostics. This decision can be based on very simple audiometric data, namely, the averaged hearing threshold at four frequencies, 4FPTA, and the measurement of the maximum SRS via audiometer under headphones, PBmax, without any hearing aid testing. The high positive predictive value of this classification enables CI centers to handle this workload. The decision process for referral proposed in this study must not be confused with CI indication, a complex process that cannot be reduced down to a few parameters. REFERENCES 1. Mathers C, Smith A, Concha M. Global burden of hearing loss in the year 2000. WHO health data and statistics. Available at: http://www.who.int/healthinfo/statistics/bod_hearingloss.pdf. Accessed February 25, 2014. 2. Sorkin DL. Cochlear implantation in the world’s largest medical device market: utilization and awareness of cochlear implants in the United States. Cochl Impl Int 2013;14:4Y12. 3. Sohn W, Jo¨rgenshaus W. Schwerho¨rigkeit in Deutschland, Repra¨sentative Ho¨rscreening-Untersuchung bei 2000 Probanden in 11 Allgemeinpraxen. Z All Med 2001;77:S143. 4. Teschner M, Polite C, Lenarz T, Lustig L. Cochlear implantation in different health-care systems: disparities between Germany and the United States. Otol Neurotol 2013;34:66Y74. 5. Kru¨ger B, Gert J, Rost U, et al. Performance groups in adults cochlear implant users: speech perception results from 1984 until today. Otol Neurotol 2008;29:509Y12.
Otology & Neurotology, Vol. 36, No. 6, 2015
Copyright © 2015 Otology & Neurotology, Inc. Unauthorized reproduction of this article is prohibited.
CI CANDIDACY AUDIOMETRY-BASED SCREENING PROCEDURE 6. Gifford RH, Dorman MF, Shallop JK, et al. Evidence for the expansion of adult cochlear implant candidacy. Ear Hear 2010;31:186Y94. 7. Blamey P, Artieres F, Bazkent D, et al. Factors affecting auditory performance of postlinguistically deaf adults using cochlear implants: an update with 2251 patients. Audiol Neurotol 2013;18:36Y47. 8. Holden LK, Finley CC, Firszt JB, et al. Factors affecting open-set word recognition in adults with cochlear implants. Ear Hear 2013; 34:342Y60. 9. Dettman SJ, D’Costa WA, Dowell RC, et al. Cochlear implants for children with significant residual hearing. Arch Otolaryngol Head Neck Surg 2004;130:612Y8. 10. Lenarz T, Sto¨ver T, Bu¨chner A, et al. Hearing conservation surgery using the hybrid-1 electrode. Results from the first clinical trial at the medical university of Hanover. Audiol Neurotol 2009; 14:22Y31. 11. Vermeire K, Nobbe A, Schleich P, et al. Neural tonotopy in cochlear implants: an evaluation in unilateral cochlear implant patients with unilateral deafness and tinnitus. Hear Res 2008;245:98Y106. 12. Arndt S, Aschendorff A, Laszig R, et al. Comparison of pseudobinaural hearing to real binaural hearing rehabilitation after cochlear implantation in patients with unilateral deafness and tinnitus. Otol Neurotol 2011;32:39Y47. 13. Hassepass F, Aschendorff A, Wesarg T, et al. Unilateral deafness in children: audiologic and subjective assessment of hearing
14.
15. 16.
17. 18.
19. 20. 21. 22.
1005
ability after cochlear implantation. Otol Neurotol 2013;34: 53Y60. De Raeve L, Wouters J. Accessibliity to cochlear implants in Belgium: state of the art on selectio, reimbursement, habilitation, and outcomes in children and adults. Cochl Impl Int 2013;14:18Y25. Chundu S, Flynn SL. Audiogram and cochlear implant candidacyV UK perspective. Cochl Impl Int 2014;15:241Y4. Chundu S, Buhagiar R. Audiologists’ knowledge of cochlear implant and their related referrals to the cochlear implant centre: pilot study findings from UK. Cochl Impl Int 2013;14:9Y13. Kielhorn H, Scho¨nermark MP. Strategische Rahmenbedingungen fu¨r die Versorgung mit Cochleaimplantaten. HNO 2014;62:214Y8. Halpin C, Rauch SD. Clinical implications of a damaged cochlea: pure-tone thresholds vs information carrying capacity. Otolaryngol Head Neck Surg 2008;140:473Y6. Moore BC. Dead regions in the cochlea: conceptual foundations, diagnosis, and clinical applications. Ear Hear 2004;25:98Y116. Niemann H. Pattern Analysis and Understanding. 2nd ed. Berlin, Germany: Springer, 2013. Hoppe U, Hast A, Hocke T. Sprachverstehen mit Ho¨rgeraten in Abha¨ngigkeit vom Tongeho¨r. HNO 2014;62:443Y8. Zwartenkot JW, Snik AF, Mylanus EA, et al. Amplification options for patients with mixed hearing loss. Otol Neurotol 2014;35:221Y6.
Otology & Neurotology, Vol. 36, No. 6, 2015
Copyright © 2015 Otology & Neurotology, Inc. Unauthorized reproduction of this article is prohibited.
