Nonparametric Treatment Comparisons for Interval-censored Failure Time Data

Download or read book Nonparametric Treatment Comparisons for Interval-censored Failure Time Data written by Qiang Zhao and published by . This book was released on 2004 with total page 156 pages. Available in PDF, EPUB and Kindle. Book excerpt: Interval-censored failure time data commonly occur in follow-up studies. One of the main goals of such studies is to compare the distributions of the failure time among different treatment groups. By interval-censored data, we mean that the failure time of interest is not completely observed for some individuals. Instead, we only know that it belongs to a certain interval. This dissertation consists of three parts discussing the problem of comparing survival functions using nonparametric approaches based on two types of interval-censored data. First, Chapter 2 discusses the treatment comparison based on current status data, a special case of interval-censored data in which the failure time for each individual is either left- or right-censored. This part of research was motivated by comparing incidence rates of occult tumors, which is one of the main objectives of tumorigenicity experiments. For the problem, a common practice is to assume that tumors are either lethal or nonlethal (Hoel and Walburg, 1972; Sun, 1999), to fit incidence rate data to certain parametric or semiparametric models (Dewanji et al., 1993), or to treat tumors with intermediate, but known lethality (Lagakos and Louis, 1988). We propose a simple nonparametric test procedure for the incidence rate comparison that allows tumors to have intermediate and unknown lethality. This method also allows the distributions of death times to depend on treatments or doses. The proposed method is applied to a dataset arising from a tumorigenicity experiment. Chapters 3 and 4 consider the comparison of survival functions for case 2 interval-censored failure time data, which often occur in clinical trials and epidemiological studies. In Chapter 3, we generalize the most commonly used log-rank test (Mantel, 1966) for right-censored data to interval-censored data that may contain mixed types of observations: exactly observed, right-censored, left-censored, and interval-censored. Numerical studies are conducted to evaluate and compare the proposed test with existing methods, and the results indicate that the proposed test works well. We apply the method to a dataset arising from an AIDS cohort study, which motivated the study. A main drawback of most existing nonparametric test procedures for interval-censored data, including the one proposed in Chapter 3, is that they are ad-hoc and no asymptotic properties are established for the test statistics although they may be intuitive or simple. For those procedures, the derivation and calculation of the variance estimation are usually heuristical and complicated. In Chapter 4, we present a class of nonparametric tests for interval-censored data without exact observations and establish their asymptotics. These tests are generalizations of the log-rank test given in Peto and Peto (1972). Estimation of the variances of the test statistics is straightforward. The simulation results show that the proposed procedures perform satisfactorily. For illustration, we then apply the methods to a dataset from a cancer study.