使用机器学习识别眼科亚专科护理和IRIS®注册（视力智能研究）推进劳动力研究

IF 3.2 Q1 OPHTHALMOLOGY

Ophthalmology science Pub Date : 2025-06-14 DOI:10.1016/j.xops.2025.100855

Ju Hyun Jeon MS , Ju-Yeun Lee MD, PhD , Tobias Elze PhD , Joan W. Miller MD , Alice C. Lorch MD, MPH , Mei-Sing Ong PhD , Ann Chen Wu MD, MPH , David G. Hunter MD, PhD , Isdin Oke MD, MPH

{"title":"使用机器学习识别眼科亚专科护理和IRIS®注册（视力智能研究）推进劳动力研究","authors":"Ju Hyun Jeon MS , Ju-Yeun Lee MD, PhD , Tobias Elze PhD , Joan W. Miller MD , Alice C. Lorch MD, MPH , Mei-Sing Ong PhD , Ann Chen Wu MD, MPH , David G. Hunter MD, PhD , Isdin Oke MD, MPH","doi":"10.1016/j.xops.2025.100855","DOIUrl":null,"url":null,"abstract":"<div><h3>Purpose</h3><div>To develop machine-learning models to identify ophthalmology subspecialists using deidentified patient data from a large database.</div></div><div><h3>Design</h3><div>Cross-sectional.</div></div><div><h3>Participants</h3><div>All ophthalmologists participating in the American Academy of Ophthalmology's IRIS® Registry (Intelligent Research in Sight) from 2013 to 2023 were classified under one of the following general or subspecialty categories: comprehensive, cataract, cornea, glaucoma, retina, oculofacial, pediatric, or neuro-ophthalmology.</div></div><div><h3>Methods</h3><div>We collected the diagnosis, procedure, and prescription codes linked to each ophthalmologist. We performed binary subspecialty classification using random forest models with fivefold cross validation and multispecialty classification using 4 approaches (diagnosis only, procedure only, prescription only, and combined).</div></div><div><h3>Main Outcome Measures</h3><div>Model performance was assessed using area under the receiver operating characteristic curve (AUROC), F1 scores, and Matthews correlation coefficient.</div></div><div><h3>Results</h3><div>The study included 9032 ophthalmologists. Classification accuracy differed by subspecialty (AUROC, retina: 0.981; oculofacial: 0.975; pediatric: 0.972; glaucoma: 0.937; cornea: 0.932; neuro: 0.912; cataract: 0.861; and comprehensive: 0.760). The procedure-only random forest model had better performance (AUROC, 0.903) than the diagnosis-only (0.880) and prescription-only (0.835) model.</div></div><div><h3>Conclusions</h3><div>Machine learning models leveraging the IRIS Registry can provide a near real-time assessment of the landscape of ophthalmic subspecialty care. Identifying subspecialty physicians through practice patterns may provide valuable insights into the future trends of eye care delivery with implications for workforce research and policy interventions.</div></div><div><h3>Financial Disclosure(s)</h3><div>Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.</div></div>","PeriodicalId":74363,"journal":{"name":"Ophthalmology science","volume":"5 6","pages":"Article 100855"},"PeriodicalIF":3.2000,"publicationDate":"2025-06-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Using Machine Learning to Identify Ophthalmology Subspecialty Care and Advance Workforce Research with the IRIS® Registry (Intelligent Research in Sight)\",\"authors\":\"Ju Hyun Jeon MS , Ju-Yeun Lee MD, PhD , Tobias Elze PhD , Joan W. Miller MD , Alice C. Lorch MD, MPH , Mei-Sing Ong PhD , Ann Chen Wu MD, MPH , David G. Hunter MD, PhD , Isdin Oke MD, MPH\",\"doi\":\"10.1016/j.xops.2025.100855\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><h3>Purpose</h3><div>To develop machine-learning models to identify ophthalmology subspecialists using deidentified patient data from a large database.</div></div><div><h3>Design</h3><div>Cross-sectional.</div></div><div><h3>Participants</h3><div>All ophthalmologists participating in the American Academy of Ophthalmology's IRIS® Registry (Intelligent Research in Sight) from 2013 to 2023 were classified under one of the following general or subspecialty categories: comprehensive, cataract, cornea, glaucoma, retina, oculofacial, pediatric, or neuro-ophthalmology.</div></div><div><h3>Methods</h3><div>We collected the diagnosis, procedure, and prescription codes linked to each ophthalmologist. We performed binary subspecialty classification using random forest models with fivefold cross validation and multispecialty classification using 4 approaches (diagnosis only, procedure only, prescription only, and combined).</div></div><div><h3>Main Outcome Measures</h3><div>Model performance was assessed using area under the receiver operating characteristic curve (AUROC), F1 scores, and Matthews correlation coefficient.</div></div><div><h3>Results</h3><div>The study included 9032 ophthalmologists. Classification accuracy differed by subspecialty (AUROC, retina: 0.981; oculofacial: 0.975; pediatric: 0.972; glaucoma: 0.937; cornea: 0.932; neuro: 0.912; cataract: 0.861; and comprehensive: 0.760). The procedure-only random forest model had better performance (AUROC, 0.903) than the diagnosis-only (0.880) and prescription-only (0.835) model.</div></div><div><h3>Conclusions</h3><div>Machine learning models leveraging the IRIS Registry can provide a near real-time assessment of the landscape of ophthalmic subspecialty care. Identifying subspecialty physicians through practice patterns may provide valuable insights into the future trends of eye care delivery with implications for workforce research and policy interventions.</div></div><div><h3>Financial Disclosure(s)</h3><div>Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.</div></div>\",\"PeriodicalId\":74363,\"journal\":{\"name\":\"Ophthalmology science\",\"volume\":\"5 6\",\"pages\":\"Article 100855\"},\"PeriodicalIF\":3.2000,\"publicationDate\":\"2025-06-14\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Ophthalmology science\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S2666914525001538\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"OPHTHALMOLOGY\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Ophthalmology science","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2666914525001538","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"OPHTHALMOLOGY","Score":null,"Total":0}

引用次数: 0

摘要

目的开发机器学习模型，利用来自大型数据库的未识别患者数据来识别眼科专科医生。从2013年到2023年，所有参加美国眼科学会IRIS®注册（视力智能研究）的眼科医生被归类为以下一般或亚专业类别之一：综合、白内障、角膜、青光眼、视网膜、眼面、儿科或神经眼科。方法收集每位眼科医生的诊断、手术和处方代码。我们使用随机森林模型进行二元亚专科分类，该模型具有五重交叉验证，并使用4种方法（仅诊断、仅手术、仅处方和联合）进行多专科分类。主要结果测量采用受试者工作特征曲线下面积（AUROC）、F1评分和马修斯相关系数评估模型的性能。结果纳入9032名眼科医生。亚专科分类准确率存在差异(视网膜AUROC: 0.981；oculofacial: 0.975;儿童:0.972;青光眼:0.937;角膜:0.932;神经:0.912;白内障:0.861;综合：0.760)。纯程序随机森林模型（AUROC, 0.903）优于纯诊断模型（0.880）和纯处方模型（0.835）。结论利用IRIS Registry的机器学习模型可以对眼科亚专科护理的情况进行近乎实时的评估。通过实践模式识别亚专科医生可能为眼科护理服务的未来趋势提供有价值的见解，并对劳动力研究和政策干预产生影响。财务披露专有或商业披露可在本文末尾的脚注和披露中找到。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

查看原文本刊更多论文

Using Machine Learning to Identify Ophthalmology Subspecialty Care and Advance Workforce Research with the IRIS® Registry (Intelligent Research in Sight)

Purpose

To develop machine-learning models to identify ophthalmology subspecialists using deidentified patient data from a large database.

Design

Cross-sectional.

Participants

All ophthalmologists participating in the American Academy of Ophthalmology's IRIS® Registry (Intelligent Research in Sight) from 2013 to 2023 were classified under one of the following general or subspecialty categories: comprehensive, cataract, cornea, glaucoma, retina, oculofacial, pediatric, or neuro-ophthalmology.

Methods

We collected the diagnosis, procedure, and prescription codes linked to each ophthalmologist. We performed binary subspecialty classification using random forest models with fivefold cross validation and multispecialty classification using 4 approaches (diagnosis only, procedure only, prescription only, and combined).

Main Outcome Measures

Model performance was assessed using area under the receiver operating characteristic curve (AUROC), F1 scores, and Matthews correlation coefficient.

Results

The study included 9032 ophthalmologists. Classification accuracy differed by subspecialty (AUROC, retina: 0.981; oculofacial: 0.975; pediatric: 0.972; glaucoma: 0.937; cornea: 0.932; neuro: 0.912; cataract: 0.861; and comprehensive: 0.760). The procedure-only random forest model had better performance (AUROC, 0.903) than the diagnosis-only (0.880) and prescription-only (0.835) model.

Conclusions

Machine learning models leveraging the IRIS Registry can provide a near real-time assessment of the landscape of ophthalmic subspecialty care. Identifying subspecialty physicians through practice patterns may provide valuable insights into the future trends of eye care delivery with implications for workforce research and policy interventions.