The two-sample problem in the accelerated failure time model.

A test is proposed for comparing two groups for difference in survivor distributions. This test allows for general censoring and for concomitant information in the form of categorical covariates. It assumes that the underlying model is accelerated failure time. The scale-change coefficient is estimated by finding a suitable transformation of the data to make the covariate-specific hazards equal. Martingale limit results are used to derive the asymptotic distribution of the estimate so that confidence interval type tests may be used. A relative risk estimate is also proposed for the two-sample case where the underlying model is accelerated failure time. The estimate, a ratio of integrated weighted hazards, and its asymptotic properties are derived using the method of influence functions. Simulation methods are used to assess the estimate for the log-logistic and Weibull survivor functions with uniform censoring. The estimate is shown to behave well for small values of the scale change parameter.PhDBiological SciencesBiostatisticsUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/128171/2/8812966.pd