Functional association models for multivariate survival processes

• 成果类型: Article
• 署名作者: Yan, J; Fine, JP
• 署名单位: University of Iowa; University of Wisconsin System; University of Wisconsin Madison; University of Wisconsin System; University of Wisconsin Madison
• 刊物名称: JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
• ISSN/ISSBN: 0162-1459
• DOI: 10.1198/016214504000001286
• 发表日期: 2005
• 页码: 184-196
• 关键词: PROPORTIONAL HAZARDS MODEL; varying-coefficient models; additive risk model; failure-time data; longitudinal data; regression-models; estimating equations; linear-models; LIFE-TABLES; distributions

摘要

We consider multivariate temporal processes that are continuously observed within overlapping time windows. The intended application is in censored multistate and multivariate survival settings, where point processes are continuously observed. These data differ from other functional data, like longitudinal data, which are discretely observed at irregular times. Existing functional approaches to survival processes use intensity models, which require smoothing and depend critically on the choice of smoothing parameters, similarly to discretely observed data. In this article we study functional mean and association regression models for the point processes, with unspecified time-varying coefficients. The continuous observation scheme is exploited; the coefficients may be estimated nonparametrically by extending generalized estimating equations to continuously observed data. The estimators automatically converge at the parametric rate without smoothing, unlike with discretely observed data. Uniform consistency and weak convergence is established with empirical process techniques. The nonparametric estimators yield new tests for covariate effects, parametric submodeling of these effects, and goodness-of-fit testing. Simulation studies and an analysis of familial aggregation of alcoholism illustrate the methodology's practical utility.