Studying exceptional treatment non-responders and genetics to predict treatment response in rheumatoid arthritis

项目总结/摘要 类风湿性关节炎（RA）患者治疗的一个主要挑战是确定最佳治疗方法。 几种有效的生物疾病缓解抗风湿药物（bDMARD）可用于RA，反映了 治疗的进展和RA的异质性;患者的亚组响应，而其他人没有。之前 研究集中在对肿瘤坏死因子抑制剂（TNFi）反应良好的患者， bDMARD，在寻找可用于临床护理的预测因子方面的成功有限。这项建议旨在 通过采取不同的方向来弥补知识上的差距。本研究的目的是关注 特殊bDMARD无应答者，定义和表征接受≥3类 bDMARD用于RA。我们将测试临床电子健康记录数据（EHR）或 基因组数据可以从TNFi应答者中鉴定出特殊的无应答者。在目标1中，我们利用来自 一个约16 K RA患者的EHR队列，以确定与异常无应答相关的临床因素 使用传统的流行病学方法。同时，我们将应用机器学习的方法， 主题建模将使我们能够评估更广泛的功能的预测性。的实例 功能包括使用自然语言从文本注释中提取的账单代码、处方和医疗概念。 语言处理在目标2中，我们将测试RA遗传风险因素是否可用于一个亚组的患者， Aim 1和其他炎性关节炎，如轴性脊柱关节病，可以预测异常的非- 对bDMARD治疗的反应。作为aim 2的一部分，我们还将纳入任何预测性临床因素 目标1中通过传统或主题建模方法确定。总体假设是， 特殊的无应答者可能比对TNFi有应答的患者更少“类RA”， 风险等位基因和典型的RA特征。这一定义提供了一种新的方法， 表型RA，重点是那些对治疗反应差的人。这项研究意义重大，因为 筛查不仅在临床上有帮助，而且还可以识别出未来新药研究的目标患者 目标的这种方法是创新的，因为它考虑了当代的数据， “机会”失败3类bDMARD，而在过去只有有限的数量可用。 这些数据将使用传统的流行病学模型和较新的方法进行检查，如主题 可以集成更广泛的数据类型的建模。最后，本提案旨在预测 届时，患者将带着基因数据前来就诊，作为其医疗记录的一部分。

PROJECT SUMMARY/ABSTRACT A major challenge in caring for patients with rheumatoid arthritis (RA) is determining the optimal therapy. Several effective biologic disease modifying anti-rheumatic drugs (bDMARDs) are available for RA, reflecting both advances in therapy, and the heterogeneity of RA; subsets of patients respond while others do not. Prior studies focused on patients with a good response to tumor necrosis factor inhibitor (TNFi), the most common bDMARD, with limited success in finding predictors that can be used in clinical care. This proposal seeks to address that gap in knowledge by taking a different direction. The objective of this study is to focus on exceptional bDMARD non-responders, defining and characterizing patients who have been on ≥3 classes of bDMARDs for RA. We will test whether data available in clinical electronic health record data (EHR) or genomic data can identify exceptional non-responders from TNFi responders. In Aim 1, we leverage data from an EHR cohort of ~16K RA patients to determine clinical factors associated with exceptional non-response using traditional epidemiologic approaches. As well, we will apply approaches using machine learning and topic modeling that will enable us to evaluate the predictiveness of a broader range of features. Examples of features include billing codes, prescriptions, and medical concepts extracted from text notes using natural language processing. In Aim 2, we will test whether RA genetic risk factors available in a subset of patients in Aim 1, and those of other inflammatory arthritides, e.g. axial spondyloarthropathy, can predict exceptional non- response to bDMARD therapy. As part of aim 2, we will also incorporate any predictive clinical factors identified in Aim 1 through the traditional or topic modeling approach. The overarching hypothesis is that the exceptional non-responders may be less “RA-like” than patients who respond to TNFi, with fewer RA genetic risk alleles and classic RA features from the narrative notes. This definition provides a new way to sub- phenotype RA, focusing on those that will have a poor response to therapy. This study is significant because a screen will be helpful not only in the clinic but can also identify patients to target for future studies of novel drug targets. This approach is innovative because it considers contemporary data where patients now have more “opportunity” to fail 3 classes of bDMARDs, where in the past there were only a limited number available. These data will be examined both using traditional epidemiologic models and newer approaches such as topic modeling that can integrate a broader range of data types. Finally, this proposal is designed to anticipate a time when patients will come for their visit with genetic data as part of their medical record.