Machine Learning to Characterize Speech Recognition and Quality of Life Outcomes in Adult Cochlear Implant Users

Adult cochlear implant (CI) users display enormous variability in both speech recognition (SR) and quality of life (QOL) outcomes. Traditional demographic and audiologic factors together explain <40% of outcome variability. Moreover, SR and QOL are only weakly correlated. This study used a supervised machine-learning approach to develop patient profiles to explain both SR and QOL outcomes in CI users.

Prospective cohort study.

Single-institution, tertiary care center.

Forty-nine experienced adult CI users.

Supervised machine learning to identify patient profiles discriminating outcome groups.

SR was measured using the Perceptually Robust English Sentence Test Open-Set (PRESTO). QOL was measured using the Nijmegen Cochlear Implant Questionnaire (NCIQ). Participants were median-split based on PRESTO and NCIQ scores, resulting in 4 outcome groups: High-SR/High-QOL, Low-SR/Low-QOL, High-SR/Low-QOL, and Low-SR/High-QOL. Demographic, auditory, and cognitive-linguistic assessments were collected to define profiles for the 4 outcome groups.

The 4 outcome groups demonstrated different factor profiles. Notably, the factors that most strongly distinguished the groups included duration of hearing loss, age at onset of hearing loss, speed of phonological and lexical processing, working memory, and nonverbal reasoning.

A battery of assessments can help characterize the profiles of CI users who experience favorable SR and QOL outcomes versus those who do not. This type of profiling analysis provides a potential tool for identifying individuals who are at high risk for experiencing poor SR and/or QOL outcomes, and those who may benefit from more specific counseling or rehabilitation.