Predicting Individual Hearing Aid Preference from Listening Experiences in Daily Life

Authors: Jeppe Christensen1, Juan Camilo Gil-Carvajal2, Johanne Rumley2, Melanie Lough3, Helen Whiston3, Gabrielle Saunders3

1Eriksholm Research Centre
2Oticon A/S
3Manchester Centre for Audiology and Deafness, School of Health Sciences, University of Manchester

Background: The relationship between daily life listening experiences and preference for a particular hearing aid (HA) is not well understood. More specifically, we do not know if the preference for a particular model of HA is mediated by satisfaction in specific listening situations or by overall satisfaction across many listening situations. In this study we investigate this by examining associations between HA preference, Ecological Momentary Assessments (EMAs) of real-world listening experiences, and HA datalogging (contextual information).

Methods: Experienced HA users (N = 40) with mild-to-moderate symmetrical sensorineural hearing loss provided reports of listening activity and satisfaction ratings via EMAs over 2 weeks with each of two different HA models, primarily differing in their noise reduction technology. At the end of the last wear period, participants selected their preferred HA to keep. We then used Random Forest classification models with 5-fold cross-validation balanced across participants to predict individual choices using EMAs and contextual data as features.

Results: When context was considered, a prognostic classification of single EMAs into end-of-trial preference achieved a 93.8% accuracy across participants. The contextual data particularly improved accuracy for participants with less consistent satisfaction ratings. The importance and partial dependency of features showed that EMAs completed in sound environments with predominantly low signal-to-noise ratio and listening activities related to television, people talking, non-specific listening, and music listening had the most significant impact on preference.

Conclusions: EMAs have high prognostic classification accuracy, making them ideal for individualized hearing outcome investigations and preference predictions. The combination of EMAs with contextual information about the sound environment results in predictions that are more invariant to inconsistent satisfaction ratings.