Statistical methodology in the health sciences

我的研究的主要目标是开发和评估可能适用于健康科学研究的统计方法。这项研究的大部分涉及的方法，可适用于当标准的统计技术是不合适的，由于违反基本假设的发展。这方面的一个例子是我过去和继续研究的分析方法的集群随机试验分配完整的社会单位，而不是个人的治疗。 由于一个集群中的个体不能被视为统计上独立的，标准的分析方法不适用，并可能导致误导性的结论。 我在这一领域的大部分研究都集中在开发考虑集群内依赖性的方法上，这样一来，所得到的推论就具有有效和高效的统计特性。 我的教科书“聚类随机试验的设计和分析”，与尼尔·克拉尔合著（阿诺德出版公司，2000）提出了一个全面的处理这一主题。中文翻译版的这一文本已于2006年出版。 第二个例子涉及的发展推理方法的观察者之间的协议，在小到中等规模的样本。 虽然kappa统计量在心理学和健康科学中被广泛用于测量观察者间的一致性，但用于显著性检验和置信区间构建的传统方法需要的样本量经常超出许多研究者可用的范围。 我在这里的总体目标是开发一个统一的框架，用于推断观察者间的一致性，以适应多个评分者和多个结果类别。几乎所有的问题，我解决源于我的经验，作为一个生物统计学家与研究人员在健康科学合作。

The principal objective of my research is to develop and evaluate statistical methodology that may be usefully applied to health sciences research. Most of this research involves the development of methods that may be applied when standard statistical techniques are inappropriate, due to the violation of underlying assumptions. An example of this is my past and continuing research on methods of analysis for cluster randomization trials which allocate intact social units rather than individuals to treatment. Since individuals within a cluster cannot be regarded as statistically independent, standard methods of analysis do not apply, and could result in misleading conclusions. Much of my research in this area has been focused on the development of methods that take into account within-cluster dependencies in such a way that the resulting inferences have valid and efficient statistical properties. My textbook "Design and Analysis of Cluster Randomization Trials", co-authored with Neil Klar (Arnold Publishing Co., 2000) presents a comprehensive treatment of this subject.The Chinese translated edition of this text was published in 2006. A second example involves the development of inference methods for interobserver agreement in samples of small to moderate size. Although the kappa statistics is widely used in both psychology and the health sciences to measure interobserver agreement, traditional methods for significance-testing and confidence interval construction require sample sizes that are frequently beyond the range available to many investigators. My overall objective here has been to develop a unified framework for inferences concerning interobserver agreement that accommodates both multiple raters and multiple outcome categories. Virtually all the problems I address emanate from my experience as a biostatistician collaborating with researchers in the Health Sciences.