Two-phase outcome-dependent sampling designs for sequential survival time analysis

Lin, Tzuemn-Renn (2018) Two-phase outcome-dependent sampling designs for sequential survival time analysis. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (5MB)

Abstract

In some observational studies, the covariates of interest might be expensive to measure although the outcome variable could easily be obtained. In this situation, a costefficient two-phase outcome-dependent sampling design could be employed to measure the expensive covariate for more informative subjects. In phase one, all members of a random sample from a population or a cohort are measured for the outcome variable and inexpensive covariates. In phase two, a subset of the cohort is selected based on the outcome variable, and the expensive covariate is measured only for the selected individuals. Case-cohort design is a commonly used outcome-dependent sampling design in time-to-event analyses. In generalized case-cohort design, in which the selection probability depends only on the event indicator, a random subsample of individuals who experienced the event are selected, along with a random subsample of those with censored event times. It was previously shown that when the selection probability at phase two depends on observed event time and censoring time in addition to the event indicator, the efficiency of the design might increase. Efficient design has a lower variance of the coefficient estimate of the expensive covariate in the regression model. In this study, we consider bivariate sequential time-to-event data, which consists of gap times between two events observed in sequence, as the outcome variables. The objective of this study is to investigate efficient two-phase sampling designs for a predetermined phase two sample size. We consider sampling designs depending on the event indicators and gap times. A likelihood-based method is used to estimate the associations between the expensive covariate and the two gap times. We show that when the selection probability at phase two depends on the two observed gap times and censoring times in addition to their event indicators, the efficiency of the design might improve.

Actions (login required)

View Item