Auditory Profile-based Hearing-aid Fitting: A Proof-of-concept study

Objective: The clinical characterization of hearing deficits for hearing-aid fitting purposes is typically based on the pure-tone audiogram only. In a previous study, a group of hearing-impaired listeners were tested using a comprehensive test battery designed to tap into different aspects of hearing. A data-driven analysis of the data yielded four clinically relevant patient subpopulations or 'auditory profiles'. In the current study, profile-based hearing-aid settings were proposed and evaluated to explore their potential for providing more targeted hearing-aid treatment. Design: Four candidate hearing-aid settings were implemented and evaluated by a subset of the participants tested previously. The evaluation consisted of multi-comparison preference ratings carried out in realistic sound scenarios. Results: Listeners belonging to the different auditory profiles showed different patterns of preference for the tested hearing-aid settings that were largely consistent with the expectations. Conclusion: The results of this proof-of-concept study support further investigations into a stratified, profile-based hearing-aid fitting with wearable hearing aids.

Hearing loss (HL) is typically treated with hearing aids (HA). The primary purpose of HAs is 18 to provide gain to the input signal to compensate for reduced audibility. In addition, modern 19 HA incorporate advanced signal processing algorithms for noise suppression (Chung 2004). 20 As a consequence, numerous parameters need to be adjusted as part of the hearing-aid fitting 21 process. 22 In current clinical practice, the assessment of the hearing deficits of a patient relies mainly on 23 pure-tone audiometry. Based on a fitting rule that typically only uses the audiogram of the 24 patient as input, the HA amplification is then adjusted. For example, the "National Acoustic 25 Laboratories -Nonlinear 2" fitting rule (Keidser et al. 2011) is commonly used for this. The 26 NAL-NL2 rule relies on a combination of empirical knowledge and modelling aimed at 27 maximizing the effective speech audibility. Even though this can provide a reasonable overall 28 solution, there are also patients whose hearing difficulties are not captured by the audiogram 29 and who may therefore benefit from other fitting strategies ( HA features, which are not yet incorporated into existing fitting rules. For example, noise 32 reduction and directional processing are currently activated based on lifestyle considerations 33 rather than audiological factors. Although advanced HA features can improve the signal-to-34 noise ratio (SNR), preference for these settings can vary substantially across listeners, possibly 35 because of unwanted speech distortions that these algorithms typically also introduce (Neher 36 et al. 2016). Therefore, it is possible that the individualized adjustment of speech enhancement 37 algorithms could improve HA outcome, for example for patients with poor speech intelligibility 38 in challenging environments. 39 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2020.04.14. to provide non-linear gain according to an audibility-based prescription formula. In HAS-III 63 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not peer-reviewed)
The copyright holder for this preprint . https://doi.org/10.1101/2020.04.14.20036459 doi: medRxiv preprint applied. Furthermore, in HAS-II and HAS-III advanced HA features were activated to provide 65 about 2.5 dB of SNR improvement under noisy conditions (see Method, Error! Reference 66 source not found., Error! Reference source not found. for more details). is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not peer-reviewed)
The copyright holder for this preprint . https://doi.org/10.1101/2020.04.14. In summary, Profile-A and -B listeners preferred audibility-based gain prescriptions, whereas 114 Profile-C and -D listeners preferred loudness-based gain prescriptions. Besides, SNR 115 improvement was beneficial for Profile-C listeners (with a high degree of SI-related deficits). 116 Overall, these initial findings provide a useful basis for further investigations into profile-based 117 HA fitting strategies that will include field studies with wearable devices and speech 118 intelligibility asssessments. 119 120 Seven listeners participated in the current study (N=2 in each subgroup except for Profile-B, 121 N=1). All of them had previously completed a comprehensive auditory test battery (Sanchez-122 Lopez et al. 2020b), based on which they had been classified as belonging to one of the four 123 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Committee for the Capital Region of Denmark H-16036391. 125
For the study, a hearing-aid simulator (HASIM) was used that consisted of three stages: A 126 beamforming stage, a noise reduction stage and an amplitude compression stage (see Table S  The multi-comparison of the HAS was realized using the SenselabOnline software (SenseLab 139 dept. 2017). On a given trial, six stimuli were presented to the listener: An anchor resembling 140 a 'broken' hearing aid, a clinically representative HAS, and the four candidate HAS (I, II, III  141 and IV). The multi-comparisons were performed sequentially across several trials. In each case, 142 a 20-sec audio file corresponding to a given sound scenario processed with the HASIM was 143 played back (Figure S 2). The listeners then used a slider ranging from 0 to 100 to rate the 144 sound of each HAS. The question posed to the listeners was "Which hearing aid would you 145 choose?". When giving their ratings, they were instructed to focus on overall preference rather 146 than on specific attributes such as noise annoyance or speech clarity. 147 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.