Statistical Methods in the Design and Analysis of Clinical Trials and Assessment of Prediction with Patient-Specific Information

在接下来的五年里，我将开发新的方法来估计、测试和统计推断基于患者特定数据建立的预测模型的准确性。预测误差将针对具有各种审查类型的生存结果数据、适应治愈和未治愈患者群体的治愈率模型以及高维数据进行研究。这项研究对于健康和医学研究人员来说很重要，他们正在预测患者的预后、疾病类别或对治疗的反应。它也是统计学方法论研究的一个重要领域，需要计算方法和理论发展。我将培训一名博士生如何应对这些统计挑战，以加强这一领域的研究，为统计科学的实际应用和进步服务。我还将开发新的临床试验设计和分析方法，用于治疗效果随患者生物标志物测量而变化的情况。一种方法是假设对定义在不同生物标记物阈值上的患者亚群产生额外的治疗效果。另一种方法是非参数的，允许治疗效果随着生物标记物的变化而改变，而不需要预先指定的功能形式。门限模型方法以其清晰的实用解释，对临床研究人员非常有用。我将指导两名理科硕士。为临床研究人员建立样本量设计指南。我将培训一名博士生关于各种门槛模型的理论和方法论研究。第二种方法更一般，它以灵活的形式提供了治疗效果的直接视图。我将培训另一名博士生，教他们如何为各种类型的数据开发非参数模型。计划在这一总方向上的工作旨在通过生物标记物研究改善临床试验，并为加拿大的统计研究增加一个新的前沿。它解决了靶向治疗和个性化药物的临床试验研究所产生的紧迫的统计问题。

In the next five years, I will develop new methods in estimation, testing and making statistical inference for the accuracy of prediction models built upon patient specific data. Prediction error will be studied for survival outcome data with various types of censoring, for cure-rate models which accommodate cured and uncured patient populations and for high dimensional data. The research is important for health and medical researchers who are making predictions for patients' prognostics, disease categories or responses to therapy. It is also an important area of methodological research in statistics in need of computational methods and theoretical development. I will train a Ph.D. student in dealing with these statistical challenges to build up the research in this area for both practical use and advancement of statistical science. I will also develop new methods for clinical trial design and analysis when the treatment effect varies with patients' biomarker measurements. One approach is to assume additional treatment effects on patient subsets defined on different biomarker thresholds. The other approach is nonparametric and allows the treatment effect to change with the biomarker without a pre-specified functional form. The threshold model approach is very useful for clinical researchers for its clear practical interpretation. I will supervise two M.Sc. students in building sample size designing guidelines for clinical researchers. I will train one Ph.D. student on theoretical and methodological study on various threshold models. The second approach is more general, and it provides a direct view of the treatment effect in a flexible form. I will train another Ph.D. student on developing the nonparametric models for various types of data. The planned work in this general direction aims to improve clinical trials through biomarker study, and adds a new frontier to the statistical research in Canada. It addresses the urgent statistical issues arising from clinical trial study on targeted treatments and personalized medicine.