Cheng Yu, Fine Jason P, Kosorok Michael R
Department of Statistics, University of Pittsburgh, Pittsburgh, Pennsylvania 15260, USA.
Biometrics. 2009 Jun;65(2):385-93. doi: 10.1111/j.1541-0420.2008.01072.x. Epub 2008 May 11.
The work is motivated by the Cache County Study of Aging, a population-based study in Utah, in which sibship associations in dementia onset are of interest. Complications arise because only a fraction of the population ever develops dementia, with the majority dying without dementia. The application of standard dependence analyses for independently right-censored data may not be appropriate with such multivariate competing risks data, where death may violate the independent censoring assumption. Nonparametric estimators of the bivariate cumulative hazard function and the bivariate cumulative incidence function are adapted from the simple nonexchangeable bivariate setup to exchangeable clustered data, as needed with the large sibships in the Cache County Study. Time-dependent association measures are evaluated using these estimators. Large sample inferences are studied rigorously using empirical process techniques. The practical utility of the methodology is demonstrated with realistic samples both via simulations and via an application to the Cache County Study, where dementia onset clustering among siblings varies strongly by age.
这项工作的灵感来自于犹他州基于人群的卡什县老龄化研究，该研究关注痴呆症发病中的同胞关系关联。由于只有一小部分人口会患上痴呆症，大多数人在没有患痴呆症的情况下死亡，因此出现了一些复杂情况。对于这种多变量竞争风险数据，标准的独立右删失数据依赖分析可能并不适用，因为死亡可能会违反独立删失假设。双变量累积风险函数和双变量累积发病率函数的非参数估计量从简单的不可交换双变量设置调整为可交换聚类数据，这是卡什县研究中大型同胞关系所需要的。使用这些估计量来评估随时间变化的关联度量。通过经验过程技术对大样本推断进行了严格研究。通过模拟以及应用于卡什县研究，用实际样本证明了该方法的实际效用，在卡什县研究中，兄弟姐妹之间痴呆症发病的聚类情况随年龄变化很大。
