Early Cochlear Implant Outcomes Predict Long-Term Speech Recognition.

OBJECTIVES: Early identification of cochlear implant (CI) users at risk for poor audiological outcomes may allow for timely intervention to optimize long-term CI benefit. This longitudinal study aims to develop and evaluate logistic regression models that can predict patients' 12-months speech recognition performance from their early speech recognition scores and other patient factors. DESIGN: This retrospective cohort study included 625 postlingually deafened, adult CI users with bilateral hearing loss from 2 tertiary CI centers. Logistic regressions were fit to model the likelihood of a clinically significant improvement in consonant-nucleus-consonant (CNC) word scores at 12-months postimplantation. All models included sex, race, age at implantation, duration of deafness, and pre-CI CNC score. The model performance benefit of adding CNC improvement status (improved versus not improved, with improvement defined as scoring beyond the upper limit of the 95% confidence interval of the preoperative score) at 1- or 3-months postactivation to the baseline models was evaluated by comparing the area under the receiver operating characteristic curve (AUC). Data were combined across institutions and then separated into training and validation cohorts at a 3:1 ratio. Models were developed using the training cohort (n = 469) and then applied to the validation cohort (n = 156). RESULTS: Of the 625 patients included, 513 patients (82%) demonstrated improvement in CI-only CNC word recognition scores at 12-months postimplantation compared with their preimplantation scores. Early improvement was strongly associated with long-term outcomes. Patients were more likely to improve by 12 months if they improved by 1 month (odds ratio = 45.72) or 3 months (odds ratio = 22.22) postimplantation. Model performances were similar across the training and validation cohorts. In the validation cohort, the model using 3-months data had the highest predictive accuracy (AUC = 0.92), followed by the 1-month model (AUC = 0.88), and the baseline preoperative model (AUC = 0.77). Based on the regression coefficients obtained from the training cohort, equations can successfully estimate the probability of CNC improvement at 12 months postimplantation for CI patients. CONCLUSIONS: Models that included CNC score improvements at 1 month demonstrated good predictive discrimination, while those incorporating 3-months improvements showed excellent discrimination in both the training and validation cohorts. This study highlights the importance of assessing early post-CI speech recognition improvement and proposes a regression model that clinicians can use in real-time to provide early estimates of their patients' probability of 12-months CNC improvement.