Cochlear implantation in postlingual deaf adults is time-sensitive towards positive outcome: clinical utility of advanced machine learning techniques.

Objectives: We investigated the effects of preoperative factors on outcomes of CIs in postlingually deaf adults using a general linear model (GLM) and a nonlinear Regression Forest Regression (RFR) model. Study Design: Postoperative monosyllabic word recognition scores (WRS) served as the dependent variable to predict. Predictors included duration of deafness, duration of auditory deprivation (duration of deafness without hearing aid use), age at implantation, preoperative hearing threshold and monosyllabic WRS in quiet. Patients: Postlingually deaf adults (n = 120) who received CI, which was fully inserted, without any inner ear abnormalities or combined disabilities and with follow-up of more than 2 years. Methods: The prediction accuracy was evaluated with the mean absolute error (MAE) as well as the Pearson's correlation coefficient between the true WRS and predicted WRS. To determine the importance of predictors, we measured increase in the MAE when a given variable was omitted in the regression model relative to when it was included. We used a leave-one-out cross-validation to avoid bias related to inclusion of the test data into the training procedure. Results: The fitting of GLMs resulted in prediction performance with correlation coefficient r=0.7 and MAE of 15.6+/-9.5). On the other hand, the RFR machine learning yielded superior prediction performance to the GLM with r=0.96 and MAE of 6.1+/-4.7 (t=9.8; p<0.00001). Computation of the importance showed that the contribution of DAD to the prediction was the highest (MAE increase when omitted: 12.1), followed by duration of deafness (8.6) and AgeCI (8.3). In a subsequent analysis, a subgroup of patients with DAD ≤ 10 years showed higher postCI WRS and smaller variation than those with DAD > 10 years. Conclusion: The current study on clinical utility of machine learning on auditory outcomes of CIs in postlingually deaf adults demonstrated that an advanced nonlinear classifier yields a highly accurate prediction ability with an error of +/-6 in WRS. Our finding also suggests that CI should be implemented no later than a sensitive period (10 years) after deafness to lead to successful outcome. Finally, our machine learning technique has the potential for patient counseling and predicting benefit from CI to patients. [ABSTRACT FROM AUTHOR]