All ETDs from UAB

Advisory Committee Chair

Leslie A McClure

Advisory Committee Members

Virginia J Howard

Charles R Katholi

Nita A Limdi

David T Redden

Document Type

Dissertation

Date of Award

2015

Degree Name by School

Doctor of Philosophy (PhD) School of Public Health

Abstract

All clinical trials must deal with protocol deviations that occur during the course of the study. One of the most important deviations is non-compliance to treatment assignment. Intention to treat (ITT) is the most commonly employed method to deal with non-compliance in a clinical trial; however, it provides biased estimates of the effect of receiving the treatment. Other methods such as per protocol (PP) and as treated (AT) provide alternatives to ITT. PP and AT, assume an all-or-nothing compliance situation. However, the possibility of being partially compliant to a treatment is common. We investigate possible approaches to incorporating partial compliance data into design and analysis of a clinical trial. We examine the practice of dichotomizing partial compliance in order to use PP, AT, and the instrumental variables (IV) methods. We show that, under assumptions favorable to the use of PP, AT, and IV, dichotomizing the partial compliance data provides biased estimates, reduces power, and in some cases in-flates type I error rates. We also investigate the use of these methods within a factorial design trial, in which participants may experience increased non-compliance due to being randomized to multiple treatments simultaneously. We investigate three methods that use partial compliance data in a linear regres-sion model as a covariate. We show that under certain assumptions, these methods will provide unbiased estimates and improve the power of a test of the treatment effect with-out inflating type I error. These methods may have reduced power or inflated type I error rates when the assumptions are not met. We developed a novel way to use compliance information in an on-going clinical trial to increase study power by utilizing sample size re-estimation (SSR) and internal pilot (IP) methods, using an estimate of average compliance in the study population. An IP is used to correct the negative effects of misspecifying the average compliance at the initial sample size estimation. We showed that this method can help a study maintain the desired level of power in the study. If compliance in the population is low, the necessary sample size may become quite large.

Included in

Public Health Commons

Share

COinS