Eligibility criteria design for Alzheimer's trials with real-world data and explainable AI

利用真实数据和可解释的人工智能设计阿尔茨海默病试验的资格标准

• 批准号: 10608470
• 负责人: Jiang Bian
• 金额: $ 82.02万
• 依托单位: WEILL MEDICAL COLL OF CORNELL UNIV
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-02-01 至 2027-11-30
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10608470
• 关键词: Adoption; Alabama; Algorithms; Alzheimer' s Disease; Alzheimer' s disease patient; Artificial Intelligence; Characteristics; Clinical; Clinical Research; Clinical Trials; Collection; Data; Data Science; Detection; Documentation; Effectiveness; Eligibility Determination; Ensure; Event; Florida; General Population; Individual; Libraries; Machine Learning; Methodology; Methods; Modeling; Natural Language Processing; Online Systems; Outcome; Participant; Patient-Focused Outcomes; Patients; Pharmaceutical Preparations; Population; Process; Research; Research Personnel; Safety; Serious Adverse Event; Site; Target Populations; Treatment Effectiveness; Treatment Efficacy; Work; cohort; computable phenotypes; data resource; design; electronic health record system; high dimensionality; improved; interest; novel; patient oriented; patient safety; privacy preservation; prototype; study population; success; tool; trait; trial design; trial enrollment; user centered design

ABSTRACT Clinical trials are often conducted under idealized and rigorously controlled conditions to ensure internal validity (maximizing potential treatment efficacy) while balancing patient safety (e.g., serious adverse events [SAEs]); but these conditions—often reflected in trials’ eligibility criteria—paradoxically, limits (1) the ability to identify the “right” study populations of the trials, and (2) the trials’ generalizability to the target population in real-world settings. Low generalizability has long been a concern, including for Alzheimer's disease (AD) trials. AD trial participants are systematically younger than AD patients in the general population, where eligibility criteria design issues are arguably the biggest yet modifiable barriers. The FDA has launched numerous initiatives to improve trial design and enrollment practices, such as using enrichment strategies (e.g., “use patient characteristic to select a study population in which detection of a drug effect [or safety event] is more likely than it would be in an unselected population”), so that the trial participants can better reflect the real-world target population and the trials are more likely to succeed. However, there are significant gaps between the need to improve AD trial eligibility criteria design and ways available to fulfill the need in practice. On the other hand, rapid adoption of electronic health record (EHR) systems has made large collections of real-world data (RWD) that reflect the characteristics and outcomes of the patients being treated in real-world settings, available for research. The increasing availability of RWD combined with the advancements in artificial intelligence (AI), especially machine learning (ML) offer untapped opportunities to generate real-world evidence (RWE) to support eligibility criteria design for AD trials, due to a number of key methodological gaps: (1) the lack of validated computable phenotyping (CP) and natural language processing (NLP) algorithms and tools that can accurately define the populations (e.g., AD patients) of interest and extract key relevant patient characteristics and outcomes of interest (e.g., trial endpoints such as MoCA and safety profile such as SAEs) from RWD, (2) the lack of ways to identify the desired study populations (and corresponding eligibility criteria), considering the impact of criteria to potential treatment effectiveness, patient safety, and study generalizability, and (3) the need of an easy-to-use toolbox to support trialists’ eligibility criteria design process. We propose (1) novel causal- principled, explainable AI (XAI) approaches to generate RWE to facilitate AD trial eligibility criteria design, and (2) to create the web-based ALZHEIMER'S DISEASE ELIGIBILITY EXPLAINER (ADEP) tool. We will leverage two large RWD resources, the OneFlorida+ (~19 million patients from Florida, Georgia, and Alabama) and INSIGHT (~12 million New Yorkers) clinical research networks (CRNs) contributing to the national Patient-Centered Clinical Research Network (PCORnet). The success of this project will establish (1) a novel, generalizable, and XAI-based trial enrichment framework with large collections of distributed RWD, and (2) a prototype toolbox that can provide RWE to eligibility criteria design, balancing effectiveness and patient safety in the target population.

抽象的 临床试验通常是在理想化和严格控制的条件下进行的，以确保内部有效性 （最大化潜在的治疗效率）在平衡患者安全的同时（例如，严重的不良事件[SAE]）； 但是，这些条件（通常在试验的资格标准中反映）偏见，限制（1）识别能力 “右”研究人群的试验人群，以及（2）试验对现实世界中目标人群的概括性 设置。长期以来，人们一直担心较低的普遍性，包括阿尔茨海默氏病（AD）试验。广告试验 参与者在一般人群中比AD患者系统年轻，那里的资格标准设计 问题可以说是最大但可修改的障碍。 FDA发起了许多倡议以改进 试验设计和注册实践，例如使用富集策略（例如，“使用患者特征 选择一个研究人群，其中检测药物作用[或安全事件]比在 未选择的人群”），以便审判参与者可以更好地反映现实世界的目标人群和 试验更有可能成功。但是，改善广告试验的需求之间存在很大的差距 资格标准设计和可满足实践需求的方法。另一方面，快速采用 电子健康记录（EHR）系统制作了大量的现实数据（RWD），以反映 在现实世界中接受治疗的患者的特征和结果，可用于研究。这 RWD的可用性增加与人工智能（AI）的进步相结合，尤其是 机器学习（ML）提供了未开发的机会，以生成真实的证据（RWE）来支持 由于许多关键的方法论差距：（1）缺乏经过验证的资格标准设计。 可计算的表型（CP）和自然语言处理（NLP）算法和工具可以 准确定义感兴趣的人群（例如AD患者），并提取关键相关患者特征 RWD的感兴趣的结果（例如，MOCA和安全性等试验终点），（2） 缺乏确定所需的研究人群（以及相应合格标准）的方法，考虑到 标准对潜在治疗有效性，患者安全性和研究概括性的影响，以及（3）需求 一个易于使用的工具箱，以支持试验者的资格标准设计过程。我们提出了（1）新的因果关系 - 校长，可解释的AI（XAI）方法来生成RWE，以促进AD试用资格标准设计，并且 （2）创建基于网络的阿尔茨海默氏病资格解释器（ADEP）工具。我们将利用两个 大型RWD资源，Oneflorida+（来自佛罗里达州，佐治亚州和阿拉巴马州的1900万患者）和Insight （约1200万纽约人）临床研究网络（CRN），以国家为中心的国家 临床研究网络（PCORNET）。该项目的成功将建立（1）小说，可推广的，并且 基于XAI的试验丰富框架，具有大量的分布式RWD，以及（2）一个原型工具箱 可以为目标人群中的资格标准设计，平衡效率和患者安全，为RWE提供RWE。