New Statistical Models for Recurrent Event Data

描述（由申请人提供）：在生物医学和公共卫生机构的许多纵向研究中，复发性事件很普遍。处理重复事件时的一个主要目标是将事件之间发生的时间与相关的伴随变量联系起来。这些知识为制定公共卫生政策、倡导适当的社会行为、确定适当的医疗行动或规定行动方案提供指导。除了伴随变量对事件间发生时间的影响外，还需要考虑其他三个方面。首先，当事件发生时，通常采取干预措施以延长事件的再次发生。其次，累积事件发生对主体的弱化效应需要得到解释。第三，由于一个被试可能有不止一个事件，相互发生的时间可能受到未观察到的因素的影响，这些因素导致这些时间之间的关联。Pena和Hollander（2002）提出了一类一般的随机模型，它同时包含了这些方面，并包含了许多现有的专业强度模型。对于这类新的和一般的模型，本项目的目的是：(a)发展和检查适当的统计方法来估计有关的模型参数；(b)研究经常性事件模型中数据应计方面的影响；(c)开发验证模型、执行拟合优度检验和诊断模型的方法；(d)将由此产生的新统计方法应用于实际的生物医学和公共卫生数据集。该项目的结果将为生物医学和公共卫生从业人员提供广泛的随机模型和新的统计方法，适用于分析公共卫生和生物医学研究中出现的复发事件数据，例如肿瘤发生、癌症研究、心理健康等。

DESCRIPTION (provided by applicant): Recurrent events are prevalent in many longitudinal studies in biomedical and public health settings. A major goal when dealing with recurrent events is to relate the event inter-occurrence times to relevant concomitant variables. Such knowledge provides guidance in formulating public health policies, advocating proper social behavior, determining appropriate medical actions, or dictating courses of action. Aside from the effects of concomitant variables on the event inter-occurrence times, three other aspects need to be considered. First, when the event occurs, an intervention is usually instituted to prolong the reoccurrence of the event. Second, the weakening effect of accumulating event occurrences on the subject needs to be accounted for. Third, since there may be more than one event for a subject, the inter-occurrence times may be affected by unobserved factors, which induce association among these times. A general class of stochastic models which simultaneously incorporate these aspects, and which subsumes many existing specialized intensity models, has been proposed in Pena and Hollander (2002). For this new and general class of models, this project aims (a) to develop and examine appropriate statistical methods for estimating relevant model parameters; (b) to examine the impact of the data accrual aspects of the recurrent event model; (c) to develop methods for validating the model, performing goodness-of-fit tests, and diagnosing the model; and (d) to apply the resulting new statistical methods to real biomedical and public health data sets. The results of this project will provide biomedical and public health practitioners a wide class of stochastic models and new statistical methods appropriate for analyzing recurrent event data arising in public health and biomedical studies such as for instance tumor occurrence, cancer studies, mental health, and many others.