Development and Evaluation of a Computable Phenotype for Normal Tension Glaucoma

IF 4.6 Q1 OPHTHALMOLOGY

Ophthalmology science Pub Date : 2025-06-18 DOI:10.1016/j.xops.2025.100858

Fountane Chan MD , Wei-Chun Lin MD, PhD , Alan Tang , Benjamin Y. Xu MD, PhD , Sophia Y. Wang MD, MS , Michael V. Boland MD, PhD , Catherine Q. Sun MD , Sally Baxter MD, MSc , Brian Stagg MD, MS , Michelle Hribar PhD , Aiyin Chen MD

{"title":"Development and Evaluation of a Computable Phenotype for Normal Tension Glaucoma","authors":"Fountane Chan MD , Wei-Chun Lin MD, PhD , Alan Tang , Benjamin Y. Xu MD, PhD , Sophia Y. Wang MD, MS , Michael V. Boland MD, PhD , Catherine Q. Sun MD , Sally Baxter MD, MSc , Brian Stagg MD, MS , Michelle Hribar PhD , Aiyin Chen MD","doi":"10.1016/j.xops.2025.100858","DOIUrl":null,"url":null,"abstract":"<div><h3>Purpose</h3><div>To develop a computable phenotype for normal tension glaucoma (NTG) to enhance disease identification from electronic health records (EHRs).</div></div><div><h3>Design</h3><div>Retrospective cohort study.</div></div><div><h3>Subjects</h3><div>Deidentified EHR data from an academic medical center identified 1851 patients aged ≥40 years, with glaucoma and available clinical notes.</div></div><div><h3>Methods</h3><div>Of these 1851 patients, 200 were randomly selected for a chart review to receive gold standard diagnoses. Four rule-based NTG computable phenotypes were developed and tested. Phenotype 1 relied on NTG International Classification of Diseases (ICD)-9 and ICD-10 codes. Phenotype 2 incorporated structured intraocular pressure (IOP) data and medication lists. Phenotype 3 used only structured IOP data. Phenotype 4 combined structured IOP and medication data natural language processing (NLP) to extract IOP values and NTG mentions from chart notes. Internal and external validation were performed.</div></div><div><h3>Main Outcome Measures</h3><div>F1 score, sensitivities, specificities, positive predictive value (PPV), negative predictive value (NPV), and accuracy.</div></div><div><h3>Results</h3><div>Chart review identified NTG in 30% of patients, and only 7% had NTG ICD codes. Phenotype 1 had an F1 of 36.8%, sensitivity 24.1%, specificity 97%, PPV 77.8%, NPV 74.9%, and accuracy 75.1%. Compared with ICD codes, phenotypes 2 and 3 had F1 of 66.7% and 69.8%, sensitivity 77.6% and 89.7%, specificity 76.3% and 71.1%, PPV 58.4% and 57.1%, NPV 88.8% and 94.1%, and accuracy of 76.7% and 76.7%, respectively. Incorporating NLP, phenotype 4 had the best performance with an F1 of 77.4%, sensitivity 82.8%, specificity 86.7%, PPV 72.7%, NPV 92.1%, and accuracy 85.5%. Phenotypes 2 to 4 increase NTG case detection fourfold compared with phenotype 1.</div></div><div><h3>Conclusions</h3><div>Normal tension glaucoma phenotypes using NLP achieved the best overall performance, and those incorporating structured data perform better than ICD codes alone. The NTG ICD code-based phenotype is highly specific but lacks sensitivity. Insights from this study may inform the development of computable phenotypes for other disease subtypes within broader disease categories.</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 100858"},"PeriodicalIF":4.6000,"publicationDate":"2025-06-18","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/S2666914525001563","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

Abstract

Purpose

To develop a computable phenotype for normal tension glaucoma (NTG) to enhance disease identification from electronic health records (EHRs).

Design

Retrospective cohort study.

Subjects

Deidentified EHR data from an academic medical center identified 1851 patients aged ≥40 years, with glaucoma and available clinical notes.

Methods

Of these 1851 patients, 200 were randomly selected for a chart review to receive gold standard diagnoses. Four rule-based NTG computable phenotypes were developed and tested. Phenotype 1 relied on NTG International Classification of Diseases (ICD)-9 and ICD-10 codes. Phenotype 2 incorporated structured intraocular pressure (IOP) data and medication lists. Phenotype 3 used only structured IOP data. Phenotype 4 combined structured IOP and medication data natural language processing (NLP) to extract IOP values and NTG mentions from chart notes. Internal and external validation were performed.

Main Outcome Measures

F1 score, sensitivities, specificities, positive predictive value (PPV), negative predictive value (NPV), and accuracy.

Results

Chart review identified NTG in 30% of patients, and only 7% had NTG ICD codes. Phenotype 1 had an F1 of 36.8%, sensitivity 24.1%, specificity 97%, PPV 77.8%, NPV 74.9%, and accuracy 75.1%. Compared with ICD codes, phenotypes 2 and 3 had F1 of 66.7% and 69.8%, sensitivity 77.6% and 89.7%, specificity 76.3% and 71.1%, PPV 58.4% and 57.1%, NPV 88.8% and 94.1%, and accuracy of 76.7% and 76.7%, respectively. Incorporating NLP, phenotype 4 had the best performance with an F1 of 77.4%, sensitivity 82.8%, specificity 86.7%, PPV 72.7%, NPV 92.1%, and accuracy 85.5%. Phenotypes 2 to 4 increase NTG case detection fourfold compared with phenotype 1.

Conclusions

Normal tension glaucoma phenotypes using NLP achieved the best overall performance, and those incorporating structured data perform better than ICD codes alone. The NTG ICD code-based phenotype is highly specific but lacks sensitivity. Insights from this study may inform the development of computable phenotypes for other disease subtypes within broader disease categories.

Financial Disclosure(s)

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

查看原文本刊更多论文

正常张力型青光眼可计算表型的发展和评价

目的建立正常张力性青光眼（NTG）的可计算表型，以提高电子健康记录（EHRs）对疾病的识别。设计回顾性队列研究。来自学术医疗中心的确定的电子病历数据确定了1851例年龄≥40岁的青光眼患者和可用的临床记录。方法从1851例患者中随机抽取200例进行图表回顾，以获得金标准诊断。开发并测试了四种基于规则的NTG可计算表型。表型1依赖于NTG国际疾病分类(ICD)-9和ICD-10代码。表型2纳入了结构化的眼内压（IOP）数据和药物清单。表型3仅使用结构化IOP数据。表型4结合结构化IOP和药物数据自然语言处理（NLP）从图表注释中提取IOP值和NTG提及。进行了内部和外部验证。主要结局指标f1评分、敏感性、特异性、阳性预测值（PPV）、阴性预测值（NPV）和准确性。结果在30%的患者中发现了NTG，只有7%的患者有NTG ICD代码。表型1的F1为36.8%，敏感性24.1%，特异性97%，PPV 77.8%, NPV 74.9%，准确性75.1%。与ICD编码相比，表型2和3的F1分别为66.7%和69.8%，敏感性分别为77.6%和89.7%，特异性分别为76.3%和71.1%，PPV分别为58.4%和57.1%，NPV分别为88.8%和94.1%，准确性分别为76.7%和76.7%。结合NLP，表型4表现最佳，F1为77.4%，敏感性为82.8%，特异性为86.7%，PPV为72.7%，NPV为92.1%，准确性为85.5%。表型2 ~ 4与表型1相比，NTG病例检出率增加了4倍。结论NLP对正常张力型青光眼的综合诊断效果最好，结合结构化数据的青光眼优于单独使用ICD编码的青光眼。基于NTG ICD编码的表型具有高度特异性，但缺乏敏感性。这项研究的见解可能为更广泛疾病类别中其他疾病亚型的可计算表型的发展提供信息。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。

求助全文

约1分钟内获得全文求助全文

来源期刊