Nonparametric and Semiparametric Analysis of Failure-time Data with Time-varying Group Status: Applications to HIV Genomic Data

• Activity Number: 426
• Type: Contributed
• Date/Time: Thursday, August 12, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Thomas LaFramboise*+
• Companies: Harvard University
• Keywords: time-dependent covariates ; HIV genomics ; survival analysis ; U-statistics ; interval censoring

Abstract

Studies in which we wish to analyze the effect of group status on failure time may be complicated by the fact that the group status may vary over time. Moreover, failure for each subject may sometimes be observed only at certain time points--a potentially different set of time points for each patient. For example, in an HIV study a subject's virus(es) may be genotyped at varying time points during his or her treatment. If we define failure to be the occurrence of a detectable mutation at a specific site on the HIV genome, then the situation described above applies--we can only specify an interval during which the mutation occurred. We describe non- and semi-parametric approaches to the analysis of this type of data. Rather than consider one failure time for each subject, we record time to event from each patient observation, thereby accommodating the changing value of the covariate. Constructing one-sample U-statistics in this setting, we automatically account for the obvious correlations between the multiple failure times recorded for each individual. We apply our method to HIV viral genomic data from a collection of three recent clinical trials.