STATISTICAL METHODOLOGY FOR CANCER CLINICAL TRIALS

描述（申请人摘要）：本提案的长期目标是 确定I期临床试验的有效和伦理上有吸引力的设计 试验，确定适当的估计程序的最大耐受 剂量（MTD），并通过提供最先进的设计来影响实践 供研究人员使用的软件。 背景：I期临床试验通常规模很小，不受控制， 对人类受试者的连续研究，旨在确定最大耐受量 一种实验性药物的剂量。也许是因为I期临床试验是 通常是非随机的，不涉及大样本， 假设驱动的统计考虑往往被忽视。然而，在这方面， 不能准确找到正确MTD的试验可能导致不充分的 剂量水平（从有效性的角度来看）被传递到进一步的 测试或在高毒性剂量水平被传递到后期阶段的试验。 “传统”设计已经流行了一段时间， 三人一组进行治疗，剂量递增或递减， 他们的反应。这样的方法简单地将MTD识别为 数据，因此估计不相关。其他人则采取了更为正式的 方法，将MTD作为剂量-反应曲线的未知参数。 然后问题变成分位数估计。这是我们的方法， 接受这个提议。参数贝叶斯方法（例如，持续 重新评估方法;剂量递增和过量控制）和非参数 方法（例如，随机游走规则）已经被提出作为允许 感兴趣的分位数的有效估计。 具体目标I推导出估计分位数的贝叶斯最优设计 在指定的剂量水平不超过规定的 分位数我们将这个问题扩展到贝叶斯序贯设计，在 同样的限制，在具体目标二中。具体目标III扩展了具体目标I 和11通过处理非二进制有序反应从世卫组织毒性 规模我们建议使用比例优势模型来推导约束 贝叶斯最优设计及其序贯模拟。具体目标四： 使用随机游走规则为试验开发非参数设计， 顺序毒性量表我们将探讨适当的估算程序， 在具体目标VI中，我们打算对以下内容进行正式比较： I期临床试验的现有方法与开发的方法 在具体目标I-V中，我们打算使用最先进的计算方法， 设施，以找到感兴趣的道德参数的确切分布。 最后，我们开发了界面友好的前端软件，方便进行 第一阶段的临床试验。

DESCRIPTION (Applicant's abstract): The long-term goal of this proposal is to determine efficient and ethically attractive designs for phase I clinical trials, determine appropriate estimation procedures for the maximum tolerated dose (MTD), and to impact practice by providing state-of-the art design software for use by investigators. Background: Phase I clinical trials are typically very small, uncontrolled, sequential studies of human subjects designed to determine the maximum tolerate dose of an experimental drug. Perhaps because phase I clinical trials are generally non- randomized, do not involve large samples, and are not hypothesis-driven, statistical considerations have often been ignored. However, trials that do not accurately find the correct MTD may result in inadequate dose levels (from the standpoint oi effectiveness) being passed on to further testing or in highly toxic dose levels being passed on to later phase trials. "Conventional" designs have been popular for some time, where patients are treated in groups of three, and doses are escalated or de-escalated depending on their responses. Such methods simply identify an MTD as a function of the data, and hence estimation is not relevant. Others have taken a more formal approach, by treating the MTD as an unknown parameter of a dose-response curve. The problem then becomes one of quantile estimation. This is the approach we take in this proposal. Parametric Bayesian methods (e.g., continual reassessment method; escalation with overdose control) and nonparametric methods (e.g., random walk rules) have been proposed as designs that allow efficient estimation of a quantile of interest. Specific Aim I derives the Bayesian optimal design for estimation of a quantile under a constraint that the assigned dose levels do not exceed a specified quantile. We extend this problem into a Bayesian sequential design, under the same constraint, in specific Aim II. Specific Aim III extends Specific Aims I and 11 by dealing with non-binary ordinal responses from the WHO toxicity scale. We propose to use a proportional odds model to derive the constrained Bayesian optimal design and its sequential analog. Specific Aim IV proposes to use a random walk rule to develop a nonparametric design for a trial with ordinal toxicity scale. We will explore appropriate estimation procedures in Specific Aim V. In Specific Aim VI, we intend to do a formal comparison of existing methodology for phase I clinical trials with the methodology developed in Specific Aims I-V. We intend to use state-of-the-art computational facilities to find exact distributions of ethical parameters of interest. Finally, we develop user-friendly front-end software to facilitate the conduct of phase I clinical trials in Specific Aim VII.