Statistical Methods

饮食是许多癌症的危险因素。然而，饮食危险因素的适当潜伏期尚不清楚。NHS数据库的一个优势是可以在30年内获得重复的饮食信息。该应用程序的一个目标(目标1)是使用指数平滑加权函数获得在30年期间对重复测量的饮食进行加权的最佳方法。这一方法也将适用于非饮食暴露，如吸烟和长期使用激素。 癌症流行病学的另一项创新是提供了多个肿瘤标志物，可以根据肿瘤类型改进特定癌症的流行病学，并帮助确认关联的因果关系。然而，随着肿瘤标志物的数量变大，肿瘤亚群的数量也变得越来越大。在目标2中，我们提出了生存分析方法来估计危险因素与肿瘤类型之间的双向交互作用以及危险因素与肿瘤类型组合之间的三向交互作用。 随着在病例对照数据集中测量的新风险因素的出现，癌症风险预测的改进越来越多。该应用程序的目标3将病例/对照数据集的新风险因素的信息与预期数据集中的标准风险因素相结合，以改进风险预测。对于乳腺癌来说，绝经年龄是一个重要的预测因素。然而，只有自然绝经和双侧卵巢切除术的女性才知道这一点。在目标4中，我们试图估计其他外科绝经妇女的自然绝经年龄，以减少偏差并提高乳腺癌风险预测的精确度。在目标5中，我们将扩展在上一轮赠款中开发的结肠癌风险预测模型，为近端癌、远端癌和直肠癌开发单独的模型。我们还考虑了使用区间删失生存方法分析复发性结直肠腺瘤预后数据的新方法。 该项目将与项目1-3密切互动，以提高我们对女性乳腺癌、结直肠癌和卵巢癌病因的了解。它还与其他项目共享由核心A(队列跟踪和数据库维护)、核心B(确认癌症和死因)、核心C(生物检验管理)和核心D(领导和数据分析)提供的强有力的行政和科学基础设施。

Diet is a risk factor for many cancers. However, the appropriate latent period for dietary risk factors is unknown. A strength of the NHS database is the availability of repeat dietary information over 30+ years. One goal of this application (aim 1) is to obtain optimal methods of weighting repeated measures of diet over a 30 year period using an exponential smoothing weighting function. This approach will also be applied to non-dietary exposures such as cigarette smoking and hormone use over a long period of time. Another innovation in cancer epidemiology is the availability of multiple tumor markers which can refine the epidemiology of specific cancers according to tumor type, and help confirm the causality of an association. However, as the number of tumor markers gets large, the number of subsets of tumors also gets large. In aim 2, we propose survival analysis methods to estimate the 2-way interaction between risk factors and tumor types as well as 3-way interactions between risk factors and combinations of tumor types. Improvements in cancer risk prediction are increasingly occurring with novel risk factors measured in case-control datasets. Aim 3 of this application combines information on novel risk factors in case/control datasets with standard risk factors in prospective datasets to improve risk prediction. For breast cancer, an Important predictor is age at menopause. However, this is only known for women with natural menopause and bilateral oophorectomy. In aim 4, we seek to estimate age at natural menopause among other surgical menopause women to reduce bias and enhance precision of breast cancer risk prediction. In aim 5, we will extend the colon cancer risk prediction model developed in the previous cycle of this grant, by developing separate models for proximal cancer, distal cancer and rectal cancer. We also consider novel methods of analysis of recurrent colorectal adenoma outcome data using interval censored survival methods. This Project will interact closely with Projects 1-3 to improve our understanding of the etiology of breast, colorectal and ovarian cancers in women. It also shares with the other Projects a strong administrative and scientific infrastructure provided by Cores A (cohort follow-up and data base maintenance), B (confirmation of cancer and cause of death), C (management of the biospecimens) and D (leadership and data analysis).