Power analysis tools for biomarker discovery with heterogeneous diseases

描述（由申请人提供）：癌症生物标志物研究在学术界、工业界和政府蓬勃发展，受到技术进步、研究积累和数据整合的推动。尽管有大量的生物标志物发现研究，但令人惊讶的是，很少有研究检查了发现研究的样本量要求，并且今天存在非常少的公开可用的工具，允许研究人员进行功效分析以确定生物标志物发现研究的样本量要求。功效分析工具的缺乏表明，大多数生物标志物研究都是在没有适当功效分析的情况下进行的，这增加了样本量不足的风险。此外，还没有解释疾病异质性的工具，这已经显示出显著增加样本量要求并显著改变用于生物标志物选择的不同分析策略的相对功效。例如，对于100例病例和100例对照，普通的t检验检测到99%的同质性疾病的生物标志物，但对于异质性疾病仅为18%。因此，如果疾病是异质性的，使用t检验和同质性疾病把握度的生物标志物研究将只有预期把握度的五分之一。不太常用的方法，如部分AUC，对异质性疾病表现良好，但对同质性疾病表现不佳。其影响是重大的：为早期检测异质性疾病而鉴定生物标志物的研究需要不同的统计选择方法和比疾病是同质的情况更大的样本量。因此，之前的生物标志物发现研究可能未能识别出生物标志物，因为它们对于异质性疾病的能力不足，或者因为它们使用的选择方法不适合异质性疾病。 该项目将开发公开可用的工具，以帮助统计学家和非统计学家规划生物标志物研究。该项目 目标1：开发用于规划生物标志物发现研究的功效分析工具。这些工具将使研究人员能够确定样本量要求，并检查发现研究的多种分析方法，同时适当考虑疾病异质性。目标2：扩展工具以支持构建生物标志物特征的功效分析。

DESCRIPTION (provided by applicant): Cancer biomarker research is flourishing across academia, industry and government, buoyed by technological advancements, accumulation of research and data integration. Despite the large number biomarker discovery studies there have been surprisingly few studies that have examined sample size requirements for discovery studies and very minimal publicly available tools exist today that allow researchers to conduct power analyses to determine sample size requirements for biomarker discovery studies. The lack of power analysis tools suggests that most biomarker studies are undertaken without proper power analyses, increasing the risk of inadequate sample sizes. Further, there are no tools that account for disease heterogeneity, which has been shown to dramatically increase sample size requirements and significantly alter the relative power of different analytic strategie for biomarker selection. For example, with 100 cases and 100 controls, the ordinary t-test detects 99% of biomarkers for a homogeneous disease, but only 18% for a heterogeneous disease. Thus, a biomarker study using the t-test and powered for a homogeneous disease would have only one-fifth the anticipated power if the disease is heterogeneous. Less commonly-used methods, such as the partial AUC, perform well for heterogeneous diseases but poorly for homogeneous diseases. The implications are significant: studies to identify biomarkers for the early detection of heterogeneous diseases require different statistical selection methods and larger sample sizes than if the disease were homogeneous. Thus, previous biomarker discovery studies may have failed to identify biomarkers because they were underpowered for a heterogeneous disease or because they used selection methods that were inappropriate for heterogeneous diseases. This project will develop publicly available tools to help both statisticians and non-statisticians in the planning of biomarker studies. The project has the following specific aims: Aim 1: Develop power analysis tools for planning biomarker discovery studies. These tools will enable researchers to determine sample size requirements and examine multiple analytic methods for discovery research while properly accounting for disease heterogeneity. Aim 2: Extend the tools to support power analyses for construction of biomarker signatures.