Use Bayesian methods to facilitate the data integration for complex clinical trials

项目摘要/摘要 这项研究计划的主要目标是开发通用和有效的fi贝叶斯统计方法来 利用复杂的临床试验数据加强药物发现。生物医学科学的快速发展是普遍的-- ING日益庞大和不同种类的与健康相关的数据，包括毒性和EFfiCacy终点，长期 生存时间和替代生物标记物Profile。尽管数据本质上是异质的，但它们服务于 相同的中心药物发现问题和多种类型的结果可以从相同的个体收集- UAL。因此，一个成功的信息整合这些产生于不同时期的大数据 复杂的临床试验可以提高假设检验的能力，加快药物发现过程，以及 加强试验的个人道德，以及其他有益的fit。然而，麻省理工学院还需要做出很大的努力--fi IGate不同平台生成的数据的差距；否则，累积的不一致和 偏差可能会扭曲复杂临床试验的统计推断。我们将解决这一重要而具有挑战性的问题 研究课题通过开发一系列新颖的贝叶斯统计方法。特别是，我们将(1)制定一项 使用患者衍生器官(PDO)和配对临床结果的联合建模方法选择 并验证了个性化医疗(2)构建了贝叶斯子群-种fi剂量优化模型。 跨多维异质数据的风险效益fit证据的大小以及(3)开发贝叶斯校准 一种集成不同距离主协议试验控制信息的网络Meta分析方法 主宰阶段。此外，我们还将开发界面友好的Web应用程序，以促进广泛应用 建议的方法在临床实践中的应用。这项提案的所有目标都是由以下实际问题驱动的 复杂的临床试验。建议的研究是一般性的，并包括各种临床试验设置， 包括肿瘤学和疫苗试验、I、II和III期试验、标准和主方案试验、长期试验 和短期结果，以及替代标记物。初步结果表明，本文提出的方法能够有效地提高图像质量。 与其他方法相比，大大减少了数据的偏差，产生了高效可靠的fi性能 现有的方法。

Project Summary/Abstract The primary goal of this research proposal is to develop general and efficient Bayesian statistical methods to enhance drug discovery using complex clinical trial data. Rapid development in biomedical sciences is generat- ing increasingly large and heterogeneous health-related data, including toxicity and efficacy endpoints, long-term survival time, and surrogate biomarker profile. Although the data are heterogeneous by nature, they serve the same central drug discovery question and multiple types of outcomes may be collected from the same individ- ual. Therefore, a successful information integration of these “big data” generated during different periods of complex clinical trials can improve the power of the hypothesis testing, speed the drug discovery process, and enhance the individual ethics of the trials, among other benefits. However, significant efforts are needed to mit- igate the gaps of the data generated from different platforms; otherwise, the accumulated inconsistencies and biases may distort the statistical inference for complex clinical trials. We will tackle this important and challenging research topic by developing a series of novel Bayesian statistical methods. In particular, we will (1) develop a jointly modeling approach using the patient-derived organoids (PDO) and the paired clinical outcome to select and verify personalized medicine (2) construct a Bayesian subgroup-specific dose optimization model to synthe- size risk-benefit evidence across multi-dimensional heterogeneous data and (3) develop a Bayesian calibrated network meta-analysis method to integrate the control information of master protocol trials during different ran- domization stages. In addition, we will develop user-friendly web apps to facilitate the widespread application of the proposed methods in clinical practice. All the aims in this proposal are driven by practical issues from complex clinical trials. The proposed research are general and encompasses a variety of clinical trial settings, including oncology and vaccine trials, phase I, II, and III trials, standard and master protocol trials, long-term and short-term outcomes, and surrogate marker. The preliminary results show that the proposed methods can substantially reduce the bias of the data and yield highly efficient and reliable performances, compared with other existing methods.