DOI
https://doi.org/10.25772/JN2K-RK26
Defense Date
2022
Document Type
Dissertation
Degree Name
Doctor of Philosophy
Department
Biostatistics
First Advisor
Robert A. Perera
Second Advisor
Roy T. Sabo
Third Advisor
Nitai D. Mukhopadhyay
Fourth Advisor
Jessica G. LaRose
Fifth Advisor
Sunny Jung Kim
Abstract
Response-Adaptive Randomization (RAR) uses outcome data from participants earlier in the trial to skew the allocation of incoming participants towards the better performing treatment arm. However, RAR does not work as intended if the outcomes are distal, as too many (if not all) participants are accrued and randomized before primary outcomes are observed. Thus, we propose different algorithms that use earlier repeated measures of the outcome to predict participants' final repeated measurements (primary outcome) until they’re replaced by the observed primary outcome. These predictions (via predictive modeling) guide RAR by allowing calculation of the target allocation ratio sooner than RAR guided by the primary outcome only. We develop a weighted target allocation ratio that down-weights predicted primary outcomes relative to observed primary outcomes to account for their uncertainty. To evaluate the proposed design, various growth models are compared under a variety of conditions, including different standardized effect sizes, accrual rates, and outcome change patterns. Finally, we also compare our proposed method to a published design in which posterior probabilities from a Bayesian linear mixed model guide adaptive allocation, with a power-loss mitigator embedded (based on Bayesian predictive power) to prevent substantial power loss due to unequal treatment group sizes. Our proposed methods allow the current allocation ratio to more closely reach the optimal target allocation ratio by adaptively allocating sooner than RAR via primary outcome while retaining similar levels of power and Type I error rate and can be suitable alternatives to the already developed Bayesian method.
Rights
© The Author
Is Part Of
VCU University Archives
Is Part Of
VCU Theses and Dissertations
Date of Submission
8-10-2022