Related Experiment Video
Updated: Mar 20, 2026
08:06
Robotic Cochlear Implantation for Direct Cochlear Access
Published on: June 16, 2022
4.2K
Advancing clinical utility of artificial intelligence: lessons from developing a model to predict cochlear implant
Valentina Carducci1, Horacio Sanchez-Trigo1,2,3, Yesmeen Elgabori1
1Department of Otolaryngology-Head and Neck Surgery, Mayo Clinic, Rochester, MN 55905, United States.
JAMIA Open
|March 19, 2026
Summary
A new machine-learning model predicts cochlear implant (CI) eligibility using routine hearing tests, outperforming current methods. This AI tool offers a practical approach to improve hearing loss treatment decisions.
Area of Science:
- Artificial Intelligence in Medicine
- Audiology and Hearing Science
- Machine Learning for Healthcare
Background:
- A significant gap exists between developing artificial intelligence (AI) models and their practical implementation in clinical settings.
- Routine hearing tests are crucial for assessing hearing loss but require efficient interpretation for complex decisions like cochlear implant (CI) candidacy.
Purpose of the Study:
- To develop a machine-learning model for predicting cochlear implant (CI) eligibility using data from standard audiological evaluations.
- To illustrate key strategies for translating AI models into clinical utility: reformulation, metric selection, and handling data variability.
Main Methods:
- Extracted data from adult patients undergoing audiometric and CI candidacy testing at Mayo Clinic (2011-2023).
- Developed regression models to predict AzBio and Consonant-Nucleus-Consonant (CNC) scores from audiogram features, then reformulated into binary classification tasks using clinical thresholds.
- Assessed input data variability by analyzing test-retest differences in isophonemes scores.
Main Results:
- Binary classification models achieved 90.4% sensitivity and 80.2% positive predictive value (PPV) at specific thresholds, outperforming existing referral heuristics.
- Application to over 50,000 historical audiograms identified patient subgroups with high PPV and varying sensitivity.
- Variability analysis revealed that noise in key predictors limits attainable precision.
Conclusions:
- AI models can be effectively translated into clinically actionable tools through reformulation, appropriate metric selection, and addressing data variability.
- The study presents a scalable framework for implementing AI in healthcare to enhance decision-making across diverse clinical scenarios.
- These AI-driven insights can improve the efficiency and accuracy of identifying candidates for cochlear implant surgery.
Keywords:
artificial intelligence in health careaudiometric assessmentclinical decision-makingcochlear implant eligibilitymachine learning
