On nonlinearity prediction of Hearing Aid behavioural measures using Speech Auditory Brainstem Responses and Decision Trees

Authors: Emanuele Perugia1 , Ghada BinKhamis1,2 , Josef Schlittenlacher1 and Karolina Kluk1
1Manchester Centre for Audiology and Deafness, University of Manchester, Manchester, UK
2King Fahad Medical City, Riyadh, Saudi Arabia

Background: Speech perception outcomes of individual hearing aid (HA) fitting are usually assessed using speech-in-noise (SIN) tests (such as BKB-SIN and VCV) or self-report questionnaires (e.g SSQ-Speech). An objective (non-behavioural) outcome measure is desirable for infants or individuals with disabilities. The speech-ABR to short consonant-vowel stimuli is a possible and clinically applicable measure; however, BinKhamis et al. (2019) found that speech-ABR is not a linear predictor of performance on aided BKB-SIN and VCV or of aided self-reported benefit (SSQ-Speech). We revisited BinKhamis et al. (2019) moving beyond the linear relations through decision trees. We aimed to predict aided SIN performance with unaided speech-ABR in quiet from which Speech-ABR in noise was subtracted.

Methods: Data were obtained from 98 HA users (age 18-60 years) with sensorineural hearing loss. Speech-ABRs were recorded in 2×2 conditions: aided vs. unaided and quiet vs. background noise. Regression trees were fitted to BKB-SIN and VCV data using as variables pure tone average (PTA, from 250 Hz to 8 kHz), age, amplitudes and latencies of the main speech-ABR peaks.

Results: The Root Mean Square Errors (RMSE) for the BKB-SIN and VCV predictions were 4.34 and 4.72 dB SNR, respectively. The PTA yielded the first split in both trees: the participants with PTA < 43 dB HL performed better on both SIN tests (dark blue rectangles); while participants with PTA > 57 (for BKB) or 63 (for VCV) dB HL performed worse (dark red rectangles). Speech-ABR peak amplitudes predicted performances for those with PTAs in-between: lower SNR was associated with larger peak D and, counter-intuitively, smaller peak E amplitude differences between quiet and noise. The peaks D and E are evoked by the vowel formant transition period and are associated with the fundamental frequency of the stimulus.

Conclusions: Speech-ABR might be suitable as an objective outcome measure only in the participants within a specific and restricted range of PTA, hence it does not seem be feasible for clinical application.

Acknowledgment. The authors are supported by the NIHR Manchester Biomedical Research Centre.

References. BinKhamis, et al. (2019). Trends Hear, 23:1–20.