Statistical Methods in Epidemiology

今年的工作主要涉及两个项目。首先，对于双变量失效时间数据的回归方法，我们建立了统计模型，并通过模拟研究对其性能进行了评估。此外，所提出的方法已被应用于WHI激素治疗试验。其次，我们提出了一种改进的家族史评分来预测乳腺癌的风险。这个项目是由姐妹研究项目推动的，所有参与者都至少有一个姐妹患有乳腺癌。在将家族史视为是/否变量的传统方法下，无法区分10个姐妹中有1个患有乳腺癌的女性与3个姐妹中有3个患有乳腺癌的女性的风险。建议的家族史评分考虑了患病姐妹的数量、姐妹总数和姐妹的年龄。在我们的模拟和数据分析中，提出的家族史评分对乳腺癌风险具有强大的预测性。

The work in this year mainly involve two projects. First, for regression methods for bivariate failure time data, we have developed a statistical model and evaluated its performance through simulation studies. Also, the proposed methods have been applied to the WHI hormone therapy trials. Second, we proposed a improved family history score for risk prediction of breast cancer. This project was motivated by the Sister Study, where all participants have at least one sister with breast cancer. With traditional approach of considering family history as a yes/no variable, there is no ability to distinguish the risks of a women with 1 sister out of 10 with breast cancer versus a women with 3 sister out of 3 with breast cancer. The proposed family history score takes number of diseased sister, total number of sisters and ages of sisters into account. In our simulation and data analysis, the proposed family history score showed powerful predictiveness for breast cancer risk.