Regression models for truncated survival data
研究了离散时间右截断生存数据的非参数最大似然估计，并提出了针对分类协变量的离散时间回归模型，同时将Cox模型扩展到连续时间的逆向风险函数。
Right truncated survival data measured in discrete time are considered. A simple derivation of Turnbull's (1976) non-parametric maximum likelihood estimate of the cumulative distribution function is given. Regression models in discrete time are considered for the analysis of right truncated and left censored survival data when categorical or ordered categorical covariates are present. Standard programs, with risk sets appropriately defined, that compute maximum likelihood estimates for regression coefficients from discrete data, may be used to compute maximum likelihood estimates of the proposed model parameters. Dynamic asymptotic methods of Arjas & Haara (1987) permit the establishment of asymptotic joint normality of the estimates. In the continuous case, an extension of Cox's model to retro-hazard functions is established using a time transformation. Asymptotic properties are derived following Andersen et al. (1985) and Andersen & Gill (1982).