Statistical Methods and Issues for Implementing Adaptive Phase I Trials

描述(由申请人提供)：本提案中的六个目标是由需要重点研究的适应性第一阶段试验设计中的重要和相关问题推动的。这项提案中的研究人员带来了统计方法、应用临床试验经验和计算机编程技能的独特强大组合，以影响肿瘤学临床试验的未来。拟议研究的成功完成将大大增强现有的第一阶段方法，并为新的适应性第一阶段试验方法提供新的见解，这对方法学和应用统计学家都将是重要的。最重要的是，我们的发现将与NIH的任务相关，NIH的任务是做出改善人们健康和拯救生命的重要发现。目的1涉及持续再评估方法(CRM)模型构建中的基本问题，这些问题在已发表的文献中被忽视，但直接影响试验的成功。目标2、3和4共享一个潜在的主题，即通过将更多的患者信息纳入剂量发现过程来提高I期试验的效率。这些信息既与患者的既往病史有关，也与试验期间他们的癌症治疗情况有关，这些数据通常是出于一般临床目的而收集的，可能会提供有关制剂毒性概况的额外信息，但在I期研究中往往被忽略。Aim 2提出了四种将患者异质性纳入适应性设计的建模方法，Aim 3研究了将非剂量限制毒性纳入MTD估计的两种方法，而Aim 4研究了将非单调疗效模式纳入其中的方法。目的5提出了在I期试验完成登记后推断每个剂量的DLT率的方法，这些信息在试验完成并找到推荐的MTD后很少考虑，但提供了围绕所选MTD及其邻近剂量的不确定性信息。缺乏可自由使用和可修改的软件仍然是将适应性I期试验设计应用于常规临床实践的主要障碍。因此，本提案包含第六个最终目标，横跨提案的所有四年，仅侧重于本提案中所述的所有方法以及尚未纳入单一软件包的现有方法的SA和R方案编制。通过这一最终目标，我们将提供一套软件包，满足从事第一阶段设计方法学工作的研究人员的一般需求和管理实际第一阶段试验的特定日常需求，最终目标是使适应性第一阶段试验设计在未来十年发表的肿瘤学试验中司空见惯。 与公共卫生相关：第一阶段试验是发现新的治疗癌症的药物的关键的第一步，无论是单独使用还是与现有药物联合使用。成功的药物发现取决于卓越的临床试验设计，以及实施这些设计的免费软件。这项建议将审查适应性第一阶段试验设计所需的改进，并提供用户友好的软件，促进将这些改进实施到实际的临床研究中。这些发现将与NIH的任务相关，并为那些设计适应性I阶段试验或管理实际适应性I阶段试验的人提供新的见解。

DESCRIPTION (provided by applicant): The six aims in this proposal are motivated by important and relevant issues in the design of adaptive Phase I trials where focused research is needed. The investigators in this proposal bring a uniquely strong combination of statistical methodology, applied clinical trial experience, and computer programming skills to impact the future of oncology clinical trials. Successful completion of the proposed research will substantially augment existing Phase I methodology and provide new insight into novel adaptive Phase I trials approaches that will be important to both methodologic and applied statisticians. Most important, our findings will be relevant to the NIH mission of making important discoveries that improve peoples health and save lives. Aim 1 concerns fundamental issues in model construction for the Continual Reassessment Method (CRM) that are ignored in published literature but have a direct impact on the success of the trial. Aims 2, 3, and 4 share an underlying theme of improving the efficiency of Phase I trials by incorporating additional patient information into the dose-finding process. This information relates to both the prior history of the patient as well as the treatment of their cancer during the trial, data that are routinely collected for general clinical purposes that could impart additional information about the toxicity profile of an agent, but that are often ignored in Phase I studies. Aim 2 proposes four modeling approaches to incorporate patient heterogeneity in adaptive designs, Aim 3 investigates two approaches for incorporating non-dose-limiting toxicities into the estimation of the MTD, while Aim 4 examines approaches to incorporate non-monotonic efficacy patterns. Aim 5 proposes methods for inference about the DLT rate for each dose after a Phase I trial has completed enrollment, information that is rarely considered once a trial is completed and the recommended MTD is found, but provides information for the uncertainty surrounding the selected MTD and its neighboring doses. The lack of freely available and modifiable software remains the major barrier to the implementation of adaptive Phase I trial designs into routine clinical practice. Therefore, this proposal contains a final, sixth aim, spanning all four years of the proposal, focused solely on the programming, in both SAS and R, of all methods described in this proposal, as well as existing methods that have yet to be housed in a single software package. Through this final aim, we will provide a suite of software packages that meet the general needs of researchers working on Phase I design methodology and the specific day-to-day needs of those who administer actual Phase I trials, with the eventual goal of making adaptive Phase I trial designs commonplace in oncology trials published in the coming decade. PUBLIC HEALTH RELEVANCE: Phase I trials are a crucial first step in the discovery of new agents, either alone or in combination with existing agents, for the treatment of cancer. Successful drug discovery hinges on superior clinical trial designs, as well as freely available software to implement those designs. This proposal will examine needed improvements to adaptive Phase I trial designs and provide user-friendly software that facilitates the implementation of those improvements into actual clinical research. The findings will be relevant to the NIH mission and provide new insight to those who either design adaptive Phase I trials or administer actual adaptive Phase I trials.