Marginal quantile regression methods for censored multiple event times
In many clinical trials or epidemiologic studies it is not uncommon to observe multiple events or failures for the same study subject. Examples include the sequence of tumor recurrences, asthmatic attacks, epileptic seizures or infection episodes in an individual. Multiple event times provide additional valuable information, but, at the same time, may introduce more complicated issues into analysis. A major difficulty involved is to specify a proper dependence structure between duration times from the same individual.
This thesis develops statistical models and methods for the analysis of censored recurrent event times, where the censoring variables are usually always observable. Through marginal approach, which bypass the difficulty of specifying correct joint distribution of error, a censored quantile regression model that parallels that of Powell's (1984, 1986) censored quantile regression model in econometrics study is proposed for multiple event times with the accelerated failure time model in survival analysis. A modified convex loss function with small positive threshold is used for estimator estimation and covariance matrix estimation. The large sample properties and numerical study are provided for this method.