Statistical Strategies for Establishing Etiologic Heterogeneity of Tumors

描述(申请人提供)：这项建议的基本前提是，基于解剖部位的癌症类型可能包含病因上不同的亚型。事实上，近年来出现了大量关于这一点的证据。该提案的目标是开发一种策略，以最佳地识别这些病因学上不同的肿瘤亚型，并开发实现这一目标所需的统计技术。除了澄清癌症病因学之外，这种方法还提供了一种更强大的策略来检测新的风险因素，通过将研究重点放在具有不同病因学的子类型上来发现这些新的风险因素。我们的研究计划是由一个关于双原发恶性肿瘤发生的关键新结果所推动的。我们表明，独立发生的癌症对的肿瘤亚型之间的优势比与潜在人群直接相关。 子类型的风险异质性。因此，来自双原发灶研究的数据可以用来从病因学的角度确定最佳的肿瘤亚型。在这项建议中，我们以这一结果为基础来开发多变量聚类技术，以优化结果簇的病因学异质性(目标1)。我们将开发类似的技术，在已知风险因素的基础上创建亚型，最大限度地提高病因异质性的程度，用于无法获得或无法获得多原发癌数据的情况下(目标2)。我们将从统计能力的角度确定将分型作为发现新风险因素的一种战略的影响(目标3)。最后，我们将开发免费提供的软件，使其他调查人员能够轻松访问 我们开发的方法(目标4)。这项研究最终将导致一个调查病因学异质性的概念框架，以及一套用于进行DAT分析的统计工具。

DESCRIPTION (provided by applicant): The fundamental premise of this proposal is that cancer types based on anatomic site may contain sub-types that are etiologically distinct. Indeed a lot of evidence for this has emerged in recent years. The goal of the proposal is to develop a strategy for optimally identifying such etiologically distinct tumor sub-types, and to develop the statistical techniques needed to accomplish this. In addition to clarifying cancer etiology, such an approach offers the promise of a more powerful strategy for detecting new risk factors, by focusing studies to discover these new risk factors on the sub-types that possess distinct etiology. Our research plan is motivated by a crucial new result regarding the occurrence of double primary malignancies. We show that the odds ratio linking tumor sub-types of pairs of independently occurring cancers is directly related to the underlying population risk heterogeneity of the sub-types. Consequently data from studies of double primaries can be used to determine optimal tumor sub-classification from an etiologic perspective. In this proposal we build upon this result to develop multivariate clustering techniques that optimize the etiologic heterogeneity of the resulting clusters (Aim 1). We will develop analogous techniques for creating sub-types that maximize the degree of etiologic heterogeneity on the basis of known risk factors for use in settings where data on multiple primary cancers are unavailable or unobtainable (Aim 2). We will determine the implications of the use of sub-typing as a strategy for detecting new risk factors from the perspective of statistical power (Aim 3). Finally, we will develop freely-available software to allow other investigators easy access to the methods that we develop (Aim 4). The research will lead ultimately to a conceptual framework for investigating etiologic heterogeneity, and a suite of statistical tools for conducting the dat analyses.