项目名称: 个体化医学中生物标记物预测能力的估计和推断
项目编号: No.11301424
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 数理科学和化学
项目作者: 马昀蓓
作者单位: 西南财经大学
项目金额: 22万元
中文摘要: 预测生物标记物能够预测病人对某一特定治疗方案的治疗效果。为了能够利用这些信息来最大程度地使病人获益,我们需要建立一套系统的统计模型,用于评估生物标记物的预测效果。目前使用的方法,大部分首先需要对生物标记物简单分层,估计每一层中各治疗组的生存曲线,通过比较生存曲线,判断生物标志物的预测效果。这类方法信息利用不充分,同时具有很大的模型偏差,特别是当生物标记物为连续度量时。在本课题中,我们将基于前期工作提出的BATE曲线(Zhou & Ma 2012),建立一个新的统计框架来弥补上述方法中的缺陷。我们的方法不仅能够直观地展现出连续生物标记物的治疗效果曲线,而且可以识别敏感人群、并为特定患者选择最优治疗方案。此外,对于超高维的潜在预测生物标记物,我们通过超高维变量选择方法对其进行筛选,同时考虑其相互之间的相关性,评价它们的综合预测能力。
中文关键词: 生物标记物;特征筛选;变量选择;变系数模型;C-统计量
英文摘要: A predictive marker is a biomarker that predicts the differential efficacy (benefit) of a particular therapy based on the value of a biomarker (e.g., only patients expressing the biomarker will respond to the specific treatment or will respond to a greater degree than those without the biomarker). To apply these exciting results to maximize patient benefit, a systematic statistical methodology is required to assess the clinical utility of promising biomarkers for predicting patients' responses to particular treatments. Most of the current statistical methods for assessing the clinical utility of a predictive biomarker are based on a comparison of estimated survival curves between a treatment and control group, stratified by the biomarker values. Such an approach may loss central information, and have serious model bias caused by misspecification in modeling. In this proposal, based on our previous work (Zhou & Ma 2012), we introduce a new concept, the BATE curve, to represent the predictive ability of a biomarker in selecting patients who respond better to one particular treatment over another treatment. On the other hand, for millions of potential predictive biomarkers, we develop a ultra-high variable seclection method, to identify real predictive biomarkers and evaluate their predictive abilities.
英文关键词: Biomarker;feature screening;variable selection;varying-coefficient model;C-statistics