The differential role of auditory and non-auditory measures for predicting speech-in-noise intelligibility: A comparison between hearing-impaired listeners with and without hearing aid supply

There exists a large variability among hearing aid users (HAUs) in rehabilitation success as reflected in measures of speech-in-noise (SIN) intelligibility, even when accounting for age and hearing loss (HL). Research has shown that standard audiological measures alone cannot comprehensively explain this variance, suggesting factors beyond these measures to contribute to the differences. Particularly the role of cognitive abilities is currently being investigated. Widely neglected so far has been the subclinical population (SCP), i.e. individuals with an age-related HL who are not yet provided with hearing aids (HA). To address these issues, the current study aims at identifying relevant factors beyond audiological ones which contribute to an improved prediction of SIN performance, and furthermore at differentiating between hearing-impaired listeners with and without HA supply at comparable levels of HL.

A multivariate analysis was conducted on data from n=333 hearing-impaired subjects from the database of the Hörzentrum Oldenburg comprising auditory measures, cognitive measures and a questionnaire with subjective ratings. These measures were entered as explanatory variables in a stepwise linear regression model for predicting speech-in-noise reception thresholds (SRTs) on group level for HA users (HAUs) (n=216) vs. non-users (SCP) (n=117). For group comparison, HAUs were measured unaided and both groups were categorized into slight (26-40 dB HL) and moderate (41-60 dB HL) degree of HL.

Preliminary results indicate that relevant predictor variables differ substantially as a function of group and level of HL. At moderate levels of HL, the audiogram was the best predictor for SRTs in both groups, accounting for ~60% variance. At slight levels, the best model in the SCP group explained ~30% variance and comprised audiological measures alone. In contrast, in the HAU group, verbal intelligence and subjective ratings accounted for two thirds of the total variance (~60%) with the audiogram explaining another third.

These results suggest that not only the degree of HL plays a role in models of performance prediction but, moreover, the provisioning with a HA. At moderate levels, audiological measures are the main predictors for both groups. For hearing-impaired subjects with a slight HL who are not provisioned with a HA, audiological measures were the best predictors as well, although leaving much unexplained variance. In contrast, for HAUs at slight levels, verbal intelligence was a better predictor than hearing ability for unaided speech understanding. This suggests that including non-audiological measures can improve patient’s performance prediction depending on HA provisioning and degree of HL.