Decision Rules for Multiple Endpoints in Clinical Trials

临床试验中多个终点的决策规则

• 批准号: 7142419
• 负责人: AJIT C TAMHANE
• 金额: $ 18.34万
• 依托单位: NORTHWESTERN UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2006
• 资助国家: 美国
• 起止时间: 2006-09-01 至 2008-07-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/7142419
• 关键词: clinical trials; computer program /software; computer system design /evaluation; drug screening /evaluation; experimental designs; health science research analysis /evaluation; statistics /biometry; technology /technique development

DESCRIPTION (provided by applicant): Many clinical trials compare a new treatment with a control on multiple endpoints. Protocols often specify complex decision rules that involve a combination of superiority and equivalence (non-inferiority) tests on primary, co-primary and secondary endpoints; some tests may be conditional on the outcomes of the other tests. Also, it is not uncommon to find sponsors proposing an adaptive modification of the protocol rule. Statistical properties of such rules are generally unknown and often they do not control the desired alpha level. This leads to much debate and confusion in the statistical evaluation of drug trials data. The long-term goal of the proposed research is to provide a general systematic framework for analyzing such rules and determining optimal allocation of the desired alpha level among the component tests to maximize power. Two approaches will be explored in achieving the research goals, both related to the theory of multiple comparisons. The first approach uses union-intersection and intersection-union tests, and assumes multivariate normality. The second approach uses the theory of gatekeeping procedures, and generalizes parallel and serial gatekeeping methods to what we call as multiple tree gatekeeping procedures. This approach is based on p-values of component test statistics and does not assume any particular distribution for the data. Bootstrap methodology will be used to implement the tests in many situations because of unknown correlations among the endpoints or non-normality of the data. The primary public health benefit of the proposed research is that it will facilitate correct statistical evaluation of multiple endpoints data. In particular, it will prevent approval of an inefficacious drug or a treatment which may be shown to be effective if incorrect statistical methods are used that do not properly control the type I error inflation caused by multiple tests on the endpoints.

描述（由申请人提供）：许多临床试验将新疗法与对照疗法在多个终点上进行比较。方案通常会指定复杂的决策规则，其中涉及对主要、共同主要和次要终点的优效性和等效性（非劣效性）测试的组合；一些测试可能以其他测试的结果为条件。此外，赞助商提出对协议规则进行适应性修改的情况也并不罕见。此类规则的统计属性通常是未知的，并且通常它们不能控制所需的 alpha 水平。这导致药物试验数据的统计评估中存在很多争论和混乱。所提出的研究的长期目标是提供一个通用的系统框架来分析这些规则并确定组件测试之间所需的 alpha 水平的最佳分配以最大化功率。为了实现研究目标，将探索两种方法，两者都与多重比较理论相关。第一种方法使用并交和交并检验，并假设多元正态性。第二种方法使用看门过程理论，并将并行和串行看门方法概括为我们所说的多树看门过程。该方法基于组件测试统计的 p 值，并且不假设数据的任何特定分布。由于端点之间的相关性未知或数据的非正态性，Bootstrap 方法将用于在许多情况下实施测试。拟议研究的主要公共卫生益处是它将促进对多个终点数据的正确统计评估。特别是，如果使用不正确的统计方法，不能正确控制由端点多次测试引起的 I 类错误膨胀，它将阻止批准无效的药物或可能被证明有效的治疗方法。