Date of Award

2016

Embargo Period

8-1-2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Public Health Sciences

College

College of Medicine

First Advisor

Sharon D. Yeatts

Second Advisor

Yuko Y. Palesch

Third Advisor

Bethany J. Wolf

Fourth Advisor

Leslie A. McClure

Fifth Advisor

Magdy Selin

Abstract

Covariate-adaptive randomization has been frequently used in randomized controlled trials (RCTs) because it can well balance prognostic factors between treatment groups. However when a subject is assigned a wrong covariate value or misplaced in a wrong cohort during the randomization procedure, it may not only impact the balancing of the covariate, but also influence the treatment assignment based on the assigned cohort. Furthermore, it is preferred that covariates that are adjusted during the randomization procedure should also be adjusted for in the primary analysis. It is not clear whether a corrected covariate value, if it could be ascertained, or a misclassified covariate value should be used for the analysis, since the covariate value is tied both to the randomization procedure and analytic model. In this research, the impact of such misclassification on the type I error rate, power for hypothesis testing for the treatment effect and estimation bias of the treatment effect is explored under covariate-adaptive randomization in Aim 1. In Aim 2, a latent class model, the Continuous-time Hidden Markov Model (CTHMM) is used to estimate the misclassification issue with respect to both the estimation of misclassification probabilities and model diagnosis. An AIC based approach, which is calculated from a modified full data likelihood, is developed to test the assumption of misclassification. In Aim 3, a two-stage analysis strategy is proposed, which combines the CTHMM and the Misclassification Simulation-Extrapolation method (MCSIMEX), to correct the estimation bias of the perfectly measured variable caused by covariate misclassification. We apply the proposed analysis strategy to data from the Interventional Management of Stroke III trial to demonstrate the two-stage model.

Rights

All rights reserved. Copyright is held by the author.

Share

COinS