Ko, Jin Hui
(2010)
Statistical Issues in the Design and Analysis of Sequentially Randomized Trials.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Adaptive treatment strategies are comprehensive methods for treating chronic diseases according to patients' needs and responses. They are useful in the treatment of diseases such as cancer or AIDS, where the treatment is frequently modified to adapt to the patients' health status. In the first part of this dissertation, we consider two commonly used randomization designs in clinical trial, namely, up-front randomized trial and sequentially randomized trial used to compare treatment strategies. Up-front randomization is the classical method of randomization where patients are randomized at the beginning of the study to pre-specified strategies. On the other hand, in sequentially randomization trials patients are randomized sequentially to available treatment options over the duration of the therapy as they become eligible to receive them. We compare the efficiency of the traditional up-front randomized trials to that of sequentially randomized trials for comparing adaptive treatment strategies both analytically and numerically based on a continuous outcome. In the second part of the dissertation, we consider analyzing right-censored survival data from two-stage sequentially randomized trials. In such analysis, it is often of interest to use median residual lifetime as the summary parameter to assess the treatment effectiveness. However, estimation of the median residual lifetime from sequentially randomized trials is not as straightforward because of its sequential randomization structure. We propose methods for estimating strategy-specific median residual life function from a two-stage sequentially randomized trial. Two types of estimators are proposed by inverting the inverse-probability-weighted estimated survival function and by using inverse-probability-weighted estimating equation function. We provide methods for estimating variances of these estimators and compare them through a simulation study. Our simulation study shows that both methods produce approximately unbiased estimators in large samples. We demonstrate our methods by applying them to a sequentially randomized leukemia clinical trial data set. Diseases such as cancer, leukemia, depression, and AIDS are major causes of morbidity and mortality in the United States. Medical research in the recent times has focused on finding optimal treatment strategies to manage or eradicate these diseases to reduce individual and community burdens. Statistical methodologies proposed in this dissertation will help appropriately design and analyze trials to develop effective treatment strategies and thus will be of significant use for improving public health in the United States and around the world.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
28 September 2010 |
| Date Type: |
Completion |
| Defense Date: |
26 July 2010 |
| Approval Date: |
28 September 2010 |
| Submission Date: |
4 July 2010 |
| Access Restriction: |
5 year -- Restrict access to University of Pittsburgh for a period of 5 years. |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
School of Public Health > Biostatistics |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Adaptive Treatment Strategies; Dynamic Treatment Regime; Inverse Probability Weighting; Sequential Randomization; Median Residual Life Function; Two-Stage Randomization Design |
| Other ID: |
http://etd.library.pitt.edu/ETD/available/etd-07042010-103236/, etd-07042010-103236 |
| Date Deposited: |
10 Nov 2011 19:49 |
| Last Modified: |
15 Nov 2016 13:45 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/8262 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}