Machine Learning Versus Simple Clinical Models for Cochlear Implant Outcome Prediction.

Area of Science:

• Otolaryngology
• Biomedical Engineering
• Data Science

Background:

• Cochlear implantation is a primary treatment for severe to profound hearing loss.
• Predicting individual cochlear implant (CI) outcomes remains a significant challenge despite standardized procedures.
• Existing predictive models for CI outcomes often lack accuracy and generalizability.

Purpose of the Study:

• To evaluate the predictive performance of simple and complex statistical and machine learning models for cochlear implant outcomes.
• To compare these models against a Null model baseline using pre-implantation variables.
• To determine if model complexity influences the accuracy of predicting CI success.

Main Methods:

• Retrospective analysis of 236 postlingual sensorineural hearing loss patients with residual hearing.
• Utilized Generalized Linear Models (GLM), Elastic Net, XGBoost, Random Forest, and ensemble methods.
• Data split into training (70%), validation (15%), and testing (15%) cohorts.

Main Results:

• All evaluated models showed comparable predictive performance, with minor differences in root mean squared errors and mean absolute errors.
• Model complexity did not significantly enhance predictive accuracy over simpler statistical approaches.
• Pre-implantation clinical variables demonstrated limited predictive validity for CI outcomes, despite all models outperforming the Null model.

Conclusions:

• Simple statistical models are as effective as complex machine learning models for predicting cochlear implant outcomes with current pre-implantation data.
• Pre-implantation clinical factors have limited power in predicting the success of cochlear implantation.
• Further research is necessary to identify more robust predictors for cochlear implant outcomes.