Loading...

Likelihood Method for Randomized Time-to-Event Studies with All-or-None Compliance

Causal Inference in Survival analysis

by Irene Hudson (Author) Zhaojing Gong (Author) Patrick Graham (Author)

Research Paper (postgraduate) 2009 154 Pages

Statistics

Summary

Estimating causal effects in clinical trials often suffers from treatment non-compliance and missing outcomes. In time-to-event studies, it is more complicated because of censoring, the mechanism of which may be non-ignorable. While new estimators have recently been proposed to account for non-compliance and missing outcomes, few papers have specifically considered time-to-event outcomes, where even the intention-to-treat (ITT) estimator is potentially biased for estimating causal effects of assigned treatment.

In this paper we develop a likelihood based method for randomized clinical trials (RCTs) with non-compliance for time-to-event data and adapt the EM algorithm according to derive the maximum likelihood estimators (MLEs). In addition, we give formulations of the average causal effect (ACE) and compliers average causal effect (CACE) to suit survival analysis.

To illustrate the likelihood-based method (EM algorithm), a breast cancer trial data was re-analysed using a model, which assumes that the failure times and censored times both follow Weibull and Lognormal distributions, respectively (termed the NIGN-WW model and NIGN-LL model).

Details

Pages
154
Year
2009
ISBN (Book)
9783668438637
File size
1013 KB
Language
English
Catalog Number
v357300
Institution / College
University of Canterbury – Department of Mathematics and Statistics
Grade
A
Tags
Causal inference Noncompliance Survival analysis Maximum likelihood EM algorithm Weibull distribution HIP breast cancer trial CACE ITT analysis

Authors

Share

Previous

Title: Likelihood Method for Randomized Time-to-Event Studies with All-or-None Compliance