Using Machine Learning to Identify Ophthalmology Subspecialty Care and Advance Workforce Research with the IRIS®...

Area of Science:

• Ophthalmology
• Machine Learning
• Health Informatics

Background:

• The American Academy of Ophthalmology's IRIS® Registry (Intelligent Research in Sight) contains deidentified patient data.
• Classifying ophthalmologists into subspecialties is crucial for workforce research and policy.
• Current methods for subspecialty identification may not be efficient or scalable.

Purpose of the Study:

• To develop and evaluate machine learning models for identifying ophthalmology subspecialists.
• To classify ophthalmologists into general and subspecialty categories using practice data.
• To assess the performance of different machine learning approaches for subspecialty classification.

Main Methods:

• Utilized deidentified data from 9032 ophthalmologists in the IRIS Registry (2013-2023).
• Employed random forest models for binary and multispecialty classification based on diagnosis, procedure, and prescription codes.
• Assessed model performance using area under the receiver operating characteristic curve (AUROC), F1 scores, and Matthews correlation coefficient.

Main Results:

• Machine learning models achieved high accuracy in identifying ophthalmology subspecialists.
• Retina (AUROC: 0.981) and oculofacial (AUROC: 0.975) subspecialties showed the highest classification accuracy.
• The procedure-only random forest model demonstrated superior performance (AUROC: 0.903) compared to diagnosis-only or prescription-only models.

Conclusions:

• Machine learning models utilizing IRIS Registry data can provide near real-time assessment of ophthalmic subspecialty care.
• Identifying subspecialty physicians through practice patterns offers insights into eye care delivery trends.
• These findings have implications for ophthalmic workforce research and policy development.