Personalized Predictions for Glaucoma Progression Using Artificial Intelligence for Electronic Health Records

使用电子健康记录人工智能对青光眼进展进行个性化预测

• 批准号: 10400077
• 负责人: Sophia Ying Wang
• 金额: $ 22.92万
• 依托单位: STANFORD UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-06-01 至 2026-02-28
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10400077
• 关键词: Adherence; Advisory Committees; Affect; Algorithms; Area; Artificial Intelligence; Blindness; Calibration; Characteristics; Clinic; Clinical; Clinical Data; Code; Communities; Complex; Computers; Data; Data Sources; Databases; Development; Disease; Documentation; Electronic Health Record; Ensure; Evaluation; Future; Glaucoma; Goals; Health; Image; Individual; Informatics; Intervention; Measurement; Medicine; Mentorship; Methods; Mission; Modeling; Names; Natural Language Processing; Operative Surgical Procedures; Ophthalmic examination and evaluation; Ophthalmology; Outcome; Patient Care; Patients; Pattern; Performance; Persons; Pharmaceutical Preparations; Phenotype; Physicians; Physiologic Intraocular Pressure; Process; Public Health; Readability; Recording of previous events; Registries; Research; Research Personnel; Scientist; Structure; Techniques; Technology; Testing; Text; Translating; Treatment outcome; Variant; Vertebral column; Veterans Health Administration; Vision; Work; artificial intelligence algorithm; base; biomedical informatics; career; clinical data warehouse; clinically relevant; cohort; data sharing; demographics; evidence base; high risk; improved; individualized medicine; innovation; insight; large scale data; machine learning model; machine learning prediction; medication compliance; optimal treatments; personalized medicine; personalized predictions; precision medicine; prediction algorithm; predictive modeling; prevent; prospective; structured data; success; support tools; tool

Project Summary/Abstract: Glaucoma is the leading cause of irreversible blindness, affecting over 60 million people worldwide. Glaucoma patients vary widely in their presentation, with some retaining long-term disease stability, and others progressing quickly to vision loss. If glaucoma patients at highest risk of progression could be identified early, clinicians could better personalize their treatment approaches. Many clinical factors that affect glaucoma progression, such as intraocular pressure, treatment history, and medication adherence, are documented within the free-text notes of the electronic health records (EHR) and are not in large-scale administrative claims databases. Recent advances in artificial intelligence (AI) and natural language processing (NLP) have enabled the integration of the rich and complex EHR data into highly accurate predictive algorithms for health outcomes in medicine and surgery. We hypothesize that we can extend these AI and NLP techniques to build predictive algorithms for glaucoma progression that outperform traditional models reliant on only administrative features. The goal of this project is to build and evaluate predictive algorithms for glaucoma progression using large-scale EHR data, while developing Dr Wang's expertise in AI and NLP, advancing her career as an independent clinician scientist. Aim 1 focuses on using the structured clinical data within the EHR, which are numeric or coded and readily machine-readable, to build baseline machine learning models predicting glaucoma progression requiring surgery. Aim 2 focuses on using and augmenting clinical named entity recognition tools to integrate information from EHR free text into AI models predicting glaucoma progression to surgery. Aim 3 focuses on understanding, explaining, and evaluating the performance of AI algorithms in a real-world prospective setting, by evaluating their performance on key subpopulations, their reliance on key features, and investigating potential areas of bias in a new cohort of glaucoma patients. This proposal is innovative in developing AI-based predictive algorithms for glaucoma progression using numeric and textual clinical data uniquely available in the EHR. The tools and methods Dr Wang will build and evaluate will substantially impact the ophthalmology field by enabling evidence-based tailoring of treatment approaches to patients' unique clinical characteristics, a step towards precision medicine. Furthermore, the careful evaluation of AI predictive algorithms on a new cohort of patients will provide insights into their performance on key subpopulations and reliance on key features, which is critical to advancing our understanding of possible limitations of deploying AI in the clinical workflow. Dr. Wang's career and research will advance under the primary mentorship of Dr. Tina Hernandez-Boussard, a national leader in informatics and expert in using NLP on EHR to improve patient care. Her outstanding Advisory Committee, including clinician-investigators Drs. Pershing, Stein, Chang, and Goldberg, will ensure Dr. Wang's success in becoming an independent clinician-investigator integrating ophthalmology and informatics.

项目摘要/摘要：青光眼是导致不可逆失明的主要原因，影响超过 6000 万人 世界各地的人们。青光眼患者的表现差异很大，有些患者仍患有长期疾病 稳定，而其他人则迅速进展为视力丧失。如果进展风险最高的青光眼患者能够 及早发现，临床医生可以更好地个性化他们的治疗方法。许多临床因素 影响青光眼进展的因素包括眼压、治疗史和药物依从性 记录在电子健康记录 (EHR) 的自由文本注释中，并且规模不大 行政索赔数据库。人工智能 (AI) 和自然语言的最新进展 处理（NLP）使丰富而复杂的 EHR 数据集成为高度准确的 医学和外科健康结果的预测算法。我们假设我们可以扩展这些 AI 和 NLP 技术可构建优于传统青光眼进展的预测算法 仅依赖于管理功能的模型。该项目的目标是构建和评估预测 使用大规模 EHR 数据进行青光眼进展的算法，同时开发王博士的 人工智能和自然语言处理方面的专业知识，推动了她作为独立临床科学家的职业生涯。目标 1 侧重于 使用 EHR 中的结构化临床数据（这些数据是数字或编码且易于机器读取） 建立基线机器学习模型来预测需要手术的青光眼进展。目标 2 重点关注 使用和增强临床命名实体识别工具将 EHR 自由文本中的信息集成到 AI 中 预测青光眼进展至手术的模型。目标 3 侧重于理解、解释和 通过评估人工智能算法在现实世界的前瞻性环境中的性能 关于关键亚群、他们对关键特征的依赖，并调查新群体中潜在的偏见领域 青光眼患者。该提案在开发基于人工智能的预测算法方面具有创新性 使用 EHR 中唯一可用的数字和文本临床数据来确定青光眼进展。工具 王博士将建立和评估的方法将对眼科领域产生重大影响 根据患者独特的临床特征制定基于证据的治疗方法，迈出了一步 精准医疗。此外，对新一批患者的人工智能预测算法进行了仔细评估 将深入了解他们在关键亚群上的表现以及对关键特征的依赖，这一点至关重要 加深我们对在临床工作流程中部署人工智能可能存在的局限性的理解。王医生的 职业和研究将在蒂娜·埃尔南德斯·布萨尔博士（Tina Hernandez-Boussard）的主要指导下取得进展，她是一位国家 信息学领域的领导者和在 EHR 上使用 NLP 来改善患者护理的专家。她出色的咨询 委员会，包括临床医生研究员 Drs。潘兴、斯坦因、张和戈德堡将确保王博士的 成功成为一名整合眼科和信息学的独立临床医生-研究者。