Research of Statistical Data Analysis Process Based on Harmonaization of Experimental and Olservational Studies

该项目的目的是研究和开发一些既可用于实验研究又可用于观测研究的方法。特别是，在实验研究中，我们从观察性研究的角度对每个实验中的数据进行了后验分析。此外，在观察研究中，我们提供了一些因果推理的方法和途径，其中包括树状结构的方法，即CART、Survival CART和MARS。在项目研究中，我们得到了以下结果和一些富有成效的发现：提供了实验研究质量的评估过程和实验的再调查周期，特别是实验研究的诊断；作为实验数据的后验分析，我们提供了从样本到个体的推理方法，例如基于树形结构方法构建个体档案，从医疗(药物)到疾病(药物)到治愈(药物)到患者；在生存分析中，我们建立了考虑长生存期的调整比例风险模型，并将该模型应用于癌症数据。针对目前脑出血在临床试验中的应用，我们提出了一些在并行试验和国外临床试验之间架起桥梁的模型和方法，并通过一些实际数据说明了这些方法的有效性，特别是对剂量-反应关系的推断进行了广泛而系统的研究，从累积分布函数估计的角度提出了一些模型和/或方法，并通过对文献数据进行了大量的模拟和试验，验证了模型和方法的有效性。

The purposes of the project are to research and develop some methodologies which can be used in both experimental and observational studies. Particularly, in experimental studies we have focused on a posteriori analysis of data in each experiment from the view of observational studies. Further, in observational studies, we provided some methods and approaches of cause-effect inference, which include tree structured methods, i.e., CART, Survival CART and MARS. In the project research, we could obtain the following results and some productive findings :We have provided the process of evaluating quality of experimental studies and re-investigated cycle of experiments, especially diagnosis of experimental studies.As a posteriori analysis of the experimental data, we have provided inference methods from "samples to individuals", for instance, from "medical treatments (drugs) to disease" to "cure (drugs) to patient" based by constructing individual profiles based on tree structured methods.In survival analysis, we have developed adjusted proportional hazard model which took account of long survivors for ordinary model, and applied the model to cancer data. Consequently, the model had exhaustive and data-adaptive properties, so it was shown to have useful applications in medical and/or clinical studies.For recent issues of ICH in clinical trials, we have proposed some models and approaches to bridging between concurrent trial and foreign clinical trials, and shown the usefulness of usefulness of the approaches by applying to some practical data.Particularly, inference of dose-response relationships have been widely and systematically investigated, some models and/or approaches have been proposed from the view of estimation of cumulative distribution function, and the utility of the models and/or approaches have been shown by conducting many simulations and trials to literature data.