Statistical Methods for Cancer Biomarkers

描述（由申请人提供）：生物标志物在医学研究中发挥着重要作用，并且在包括癌症在内的各种疾病的检测、诊断、分期、治疗和管理中提倡使用它们。使用新的基因组和蛋白质组学技术，新的生物标志物不断被发现。生物标志物被一些人视为推进癌症转化研究的最关键组成部分。它们是FDA阐述的关键路径的核心，用于加强药物开发。有几个领域的统计考虑将提高生物标志物的效用。首先，许多研究人员认为，单一的生物标志物不足以预测，需要一组生物标志物。这导致了如何以"最佳"方式联合收割机组合生物标志物的问题。第二，许多候选生物标志物可用作随机试验的替代终点。最后，对于许多生物标志物，单调性的假设是先验已知的，例如，生物标志物水平的增加与疾病风险的增加相关。具有利用这些知识来改善生物标志物的效用的可用统计程序将是有价值的。在这项资助中，我们将为癌症研究中的生物标志物数据开发几种新的统计建模程序。具体目标1：开发生物标志物的半参数和非参数多元保序回归建模程序。(a)二元保序回归模型的两阶段估计程序及伴随的非参数和半参数轮廓似然比推断程序。(b)删失数据有序分类协变量的保序估计方法。具体目标2：在单项试验和多项试验框架中开发用于分析替代终点的统计方法。(a)替代终点分析的反事实建模方法。(b)替代终点分析的收缩估计方法。(c)替代终点分析的半竞争风险方法。具体目标3：开发混合模型平均方法和伴随的基于投影的框架，用于组合生物标志物以优化预测准确性。公共卫生相关性：癌症生物标志物的统计方法。生物标志物是可以从患者身上获得的测量值，例如从简单的血液测试中获得。生物标志物不能测量患者的感受或症状，但它们对于早期发现疾病或监测治疗效果或评估新治疗是否有效可能非常有用。这项资助的目的是开发分析生物标志物数据集的有效方法，从数据中获得最多的信息。

DESCRIPTION (provided by applicant): Biomarkers play an important role in medical research, and their use is being advocated in the detection, diagnosis, staging, treatment and management of a variety of diseases, including cancer. Using novel genomic and proteomic technologies, new biomarkers are constantly being discovered. Biomarkers are viewed by some as the most critical component in moving forward translational research in cancer. They are central to the critical path, articulated by the FDA, for enhancing drug development. There are several areas where statistical consideration would enhance the utility of biomarkers. First, many researchers have suggested that a single biomarker will be insufficient for prediction and that a panel of biomarkers will be necessary. This leads to the question of how to combine the biomarkers in an "optimal" manner. Second, many of the candidate biomarkers might be used as surrogate endpoints in randomized trials. Finally, for many biomarkers assumptions of monotonicity are a-priori known, e.g. increasing levels of biomarker are associated with increased disease risk. It would be valuable to have available statistical procedures that make use of this knowledge to improve the utility of the biomarkers. In this grant, we will be developing several new statistical modeling procedures for biomarker data in cancer studies. They are summarized in the following aims: Specific Aim 1: Development of semiparametric and nonparametric multivariate isotonic regression modelling procedures for biomarkers. (a) Two-stage estimation procedures for bivariate isotonic regression models and attendant nonparametric and semiparametric profile likelihood ratio inference procedures. (b) Isotonic estimation methods with ordered categorical covariates for censored data. Specific Aim 2: Development of statistical methods for the analysis of surrogate endpoints in a single-trial and multiple-trial framework. (a) Counterfactual-based modelling approach to the analysis of surrogate endpoints. (b) Shrinkage estimation approaches to the analysis of surrogate endpoints. (c) Semi-competing risks methodology for the analysis of surrogate endpoints. Specific Aim 3: Development of hybrid model averaging methods and attendant projection-based framework for combining biomarkers to optimize predictive accuracy. PUBLIC HEALTH RELEVANCE: Statistical Methods for Cancer Biomarkers. Biomarkers are measurements that can be taken from patient, for example from a simple blood test. Biomarkers do not measure how the patients feel or symptoms they have, never-the-less they may be very useful to give early detection of a disease, or to monitor how the treatment is working, or to assess whether a new treatment if effective. The aim of this grant is to develop valid methods of analyzing datasets of biomarkers, that get the most information out of the data.