Jiang Hongyu, Chappell Rick, Fine Jason P
Department of Biostatistics and Center for Biostatistics in AIDS Research, Harvard University, Boston, MA 02115, USA.
Control Clin Trials. 2003 Apr;24(2):135-46. doi: 10.1016/s0197-2456(02)00307-0.
We consider estimating the distribution of nonterminal event time that may be censored by a terminal event but not vice versa. This problem may arise in two scenarios: (1) clinical trials involving both terminal and nonterminal events, or (2) when the primary outcome of the trial is nonterminal but there exists informative dropout. Inference is complicated by the dependent censoring from mortality or informative dropout. In this paper, we show that the joint distribution of the events can be formulated such that only the association is parameterized, with the marginal distributions unspecified. A closed form estimator for the association parameter is obtained from a novel adaptation of Oakes' concordance estimating equation. Tests for independence and goodness-of-fit of the model are also developed. A computationally simple, consistent and asymptotically normal estimator for the marginal distribution of the nonterminal event is given. Simulations demonstrate that the methods work well with practical sample sizes. The proposals are illustrated with data from an AIDS clinical trial.
我们考虑估计可能被终末事件截尾但反之不然的非终末事件时间的分布。这个问题可能出现在两种情形中：（1）涉及终末和非终末事件的临床试验，或（2）当试验的主要结局是非终末的但存在信息性失访时。来自死亡或信息性失访的相依截尾使推断变得复杂。在本文中，我们表明事件的联合分布可以被构建，使得仅对关联进行参数化，而边际分布未明确指定。通过对奥克斯一致性估计方程的一种新颖改编获得了关联参数的闭式估计量。还开发了模型的独立性检验和拟合优度检验。给出了一个计算简单、一致且渐近正态的非终末事件边际分布的估计量。模拟表明这些方法在实际样本量下效果良好。用一项艾滋病临床试验的数据对这些提议进行了说明。
