Identifying existing, FDA-approved drugs with clinically protective effects against coronavirus disease 2019 using a big data approach

项目总结/摘要 2019冠状病毒病（COVID-19）是一种国家和全球公共卫生紧急情况。因为 致病病毒是新的，目前的治疗选择非常有限，有效的疫苗 可能是1-2年后。因此，迫切需要针对该疾病的有效治疗剂。 虽然正在开发新药，但这一进程缓慢，而且需要大量资源。简而言之- 从中期来看，一个上级策略是重新利用现有的药物来治疗这种疾病。超过 已经被美国食品和药物管理局（FDA）批准的100种药物已经在体外，在计算机上， 或理论上对SARS-CoV-2，引起COVID-19的病毒，或炎症过度的 引起的免疫反应目前尚不清楚的是，其中有多少具有显著的保护作用 在实际的病人身上，因为只有一小部分这些药物在临床试验中。这些代理人中的大多数是 慢性药物，因此有数百万美国人已经在使用它们。第一个目标 这项研究的目的是评估这些药物对严重的 COVID-19并发症，同时调整已知的风险因素和混杂因素。第二个目标是 寻找药物或药物组合与特定人口统计学之间的其他相互作用， 和/或可能是保护性的或有害的临床亚组。改变医疗保健数据库， 的COVID-19研究数据库，包含最新的健康保险索赔数据约一- 三分之一的美国人。使用该数据库，本研究将评估这些药物对风险的影响 COVID-19阳性患者的四个重要结果：需要住院，使用 机械通气休克死亡结果将根据风险因素进行风险调整， 用于预测COVID-19的不良结果。本研究将进一步挖掘数据， 药物或药物组合与不同患者亚群之间的三阶相互作用 使用一种新的机器学习方法，称为可行解算法（FSA）。FSA使 研究人员发现回归模型中的高阶统计相互作用，这导致了 识别子组和复杂性，这在传统回归中并不总是显而易见的 模型如果结果显示候选药物具有高度保护作用，这些药物可以优先用于 前瞻性临床研究。显示有害作用的药物可以考虑停药， 感染或高危患者。

Project Summary/Abstract Coronavirus Disease 2019 (COVID-19) is a national and global public health emergency. Because the causative virus is novel, the present options for treatment are extremely limited, and an effective vaccine could be 1-2 years away. Thus, there is an urgent need for efficacious therapeutics against the disease. While development of new drugs is under way, that process is slow and resource-intensive. In the short- to-medium term, a superior strategy is to repurpose already existing drugs to treat the disease. Over 100 drugs already approved by the Food and Drug Administration (FDA) have shown in vitro, in silico, or theoretical effect against SARS-CoV-2, the virus that causes COVID-19, or the hyperinflammatory immune response it provokes. What is unclear is how many of these have a significant, protective effect on actual patients, as only a tiny fraction of these drugs is in clinical trials. Most of these agents are chronic medications, and thus there are millions of Americans who are already using them. The first aim of this study is to assess the degree of protection any of these drugs confers against the serious complications of COVID-19 while adjusting for known risk factors and confounders. The second aim is to search for additional interactions between drugs or combinations of drugs and specific demographic and/or clinical subgroups that could be protective or harmful. The Change Healthcare Database, a part of the COVID-19 Research Database, contains up-to-date health insurance claims data for about one- third of all Americans. Using this database, this study will evaluate the impact of these drugs on the risk of four important outcomes in patients who are COVID-19-positive: need for hospitalization, use of mechanical ventilation, shock, and death. Results will be risk-adjusted for the risk factors already well established to predict poor outcomes in COVID-19. This study will further mine the data for second- and third-order interactions between drugs or combinations of drugs and different subpopulations of patients using a novel machine learning method called the Feasible Solution Algorithm (FSA). The FSA enables the researcher to uncover higher-order statistical interactions in regression models, which leads to the identification of subgroups and complexities that are not always apparent with traditional regression models. If the results show candidate drugs with highly protective effects, these can be prioritized for prospective clinical studies. Drugs that show harmful effects can be considered for discontinuation in infected or high-risk patients.