Methodology and design for biostatistics

我的研究议程涉及最优和实验设计以及生存分析领域。在优化设计中，人们决定在哪里或何时进行测量，以便从收集的数据中做出可能的“最佳”推断。为了说明的目的，考虑一项临床试验，其中50人接受降压治疗。我们能否开发设计，使我们能够“最佳”估计：(I)受治疗影响的人数比例，(Ii)治疗效果的时间，和/或(Iii)治疗前后的平均血压。实验设计的一个有趣应用是在基因组学中。在过去的20年里，已经开发了许多新技术来研究我们的DNA以及我们DNA中的变异，这些变异可能表明我们容易患上癌症等疾病。这些实验往往非常昂贵，因此它们的设计是确保尽可能收集到最高质量数据的关键问题。 生存分析领域关注的是对在某些方面不完整的“生命周期”的观察。例如，当跟踪受试者从他们的疾病开始时，一些可能会失去随访。所有这类受试者所知道的是，他们活了下来，直到他们从研究中消失。我提出的生存分析研究关注的是观察到的“寿命”往往比预期的更长，这是因为受试者被选入研究组的方式。简单地说，观察到的是较长的幸存者，而不是较短的幸存者，分析中必须进行调整，以适应这种选择偏差。这些调整有助于改进我们的推断，因此，在考虑疾病的自然病史时，它们具有重要的影响。

My research agenda concerns the areas of optimal and experimental design and survival analysis. In optimal design one determines where or when measurements should be taken in order to make the "best" inference possible from the data that will be collected. For the purpose of illustration, consider a clinical trial where 50 people are administered a blood pressure lowering treatment. Can we develop designs that will allow us to "best" estimate: (i) the proportion of people affected by the treatment, (ii) the timing of the treatment effect and/or, (iii) the before- and after-change average blood pressures. An interesting application of experimental design is in genomics. In the last 20 years many new technologies have been developed for studying our DNA and variations in our DNA that might suggest predisposition to diseases such as cancer. These experiments are often very costly hence their design is a key issue in ensuring that the best quality data possible can be collected. The area of survival analysis is concerned with observations of "lifetimes" that are incomplete in some way. For example, when following subjects from the onset of their disease some may be lost to follow-up. All that is known for such subjects is that they survived up until they were lost from the study. My proposed research in survival analysis is concerned with observed "lifetimes" that tend to be longer than expected because of the way the subjects were selected into the study group. Simply, longer survivors are observed, shorter ones are not, and adjustments in the analysis must be made to accommodate this selection bias. These adjustments help improve our inference and, hence, they have important implications when considering the natural history of a disease.