POPULATION PHARMACOKINETICS/DYNAMICS: STATISTICAL ISSUES

癌症的临床研究正变得越来越复杂。我们对肿瘤细胞生长的分子特征的理解正在催生抗癌药物开发的新思路，这对临床研究的设计和分析提出了挑战，以评估其活性。仍然需要更好地了解常用抗癌药物以及这些新药物的药代动力学和药效学变异性的决定因素。事实上，大多数（如果不是全部的话）抗癌药物最初都是在样本量相对较小的早期临床研究中进行最严密的检查的，这就要求有一种方法能够有效而连贯地将从不同研究中收集到的信息结合起来。为了解决这些和类似的问题，本应用程序描述了用于分析和设计复杂数据研究的统计方法的持续研究和开发。第一个具体目标涉及具有混合先验的相关数据的分层建模，并将方法扩展到来自多个周期治疗和/或为多个终点收集的患者数据（例如，同时给予两种或多种药物的药代动力学）。第二个具体目标是将产生相关数据的研究信息（例如，来自同一患者的多次活检的纵向剖面或分析结果）结合起来，这些研究可能或多或少具有共同点。该应用程序描述了统计工具的开发，以便在具有此类相关结果数据的研究中进行完整的贝叶斯元分析。重要的是，统计模型还允许混合模型中的公共和研究特定组件，并以数据依赖的方式推断模型允许跨研究借鉴的程度。第三个具体目标提出了在具有混合先验的复杂生物过程建模背景下确定样本大小，采样时间和/或每个患者（或肿瘤）样本数量的最佳设计规则。该设计将允许在后续周期中对未来受试者或当前受试者的设计点（例如采样时间）进行顺序更新。提出的工作将探讨这些设计相对于其他可能的设计策略的特性。

Clinical research in cancer is becoming ever more complex. Our understanding of the molecular characteristics of tumor cell growth is spawning new ideas for anticancer drug development that present challenges for the design and analysis of clinical studies to evaluate their activity. There still exists a need for better understanding of the determinants of variability in the pharmacokinetics and pharmacodynamics of commonly used anticancer drugs, as well as of these newer agents. The fact that most, if not all, anticancer agents are initially examined most closely in early phase clinical studies having relatively small sample sizes begs for methods that will combine information gathered from separate studies efficiently and coherently. Addressing these and similar issues, this application describes continuing research and development of statistical methods for the analysis and design of studies of complex data. The first specific aim concerns hierarchical modeling of correlated data with mixture priors and expands the methods to data from patients treated over multiple cycles and/or collected for multiple end points (e.g., pharmacokinetics of two or more drugs given simultaneously). The second specific aim addresses combining information from studies that generate correlated data (e.g., longitudinal profiles or assay results from multiple biopsies from the same patient) and that may have more or less in common. The application describes the development of statistical tools to carry out full Bayesian meta-analysis across studies with such correlated outcome data. Importantly, the statistical model also allows for common and study-specific components in the mixture model and for inferring in a data-dependent way the degree to which the model allows borrowing strength across studies. The third specific aim proposes optimal design rules for determining sample size, sampling times and/or numbers of samples per patient (or tumor) in the context of modeling complex biologic processes with mixture priors. The designs will allow for sequential updating of design points (e.g., sampling times) for future subjects or current subjects in subsequent cycles. The proposed work will explore the properties of these designs relative to other possible design strategies.