Loading...
Bayesian Adaptive Designs for Phase III Trials with Time-to-event Endpoint in the Case of Non-proportional Hazards
Thewarapperuma, Nadeesha
Thewarapperuma, Nadeesha
Citations
Altmetric:
Abstract
Results from the phase II clinical trial can be used to guide the design of a phase III clinical trial. When prior results, such as those from the phase II trial, indicate that the proportional hazards assumption between the treatment and control arm does not hold, it is not appropriate to design a phase III trial using traditional approaches since it can lead to incorrect sample sizes. An insufficient sample size may result in a study with inconclusive results, and too large a sample size will waste valuable resources. Sample size calculations and clinical trial designs are well-established for data with continuous or dichotomous endpoints, or for time-to-event data that meet the traditional assumptions, such as the assumption of proportional hazards or the assumption that survival times follow an exponential distribution. There is growing interest in research for clinical trial designs under a non-proportional hazard setting, and in this dissertation, we consider two designs. Treatment benefits under both designs are measured as an improvement in longevity in survival times for patients randomized to the treatment arm compared to survival times for patients randomized to the control arm.The first design is under the proportional time setting, and treatment benefit is measured as an increase in longevity, usually by a factor greater than one. For example, with a factor of 1.5, a patient randomized to the control arm may have a survival time of 3 months, while a similar patient randomized to the treatment arm may have a survival time of 4.5 months. In the second paper, we relax some of the assumptions proposed under the first design. That is, that the treatment benefit is instantaneous following trial commencement, and the magnitude of the treatment benefit is preserved over the course of the trial. In this design, we consider two scenarios of interest, and which can be observed in a real-life setting. The first is a treatment benefit with a delayed effect where it may take some time to observe the results of the novel treatment, and the second is a treatment with diminishing benefits, where the benefit is immediate but tapers off over the course of the trial. Finally, we introduce an R package to implement the clinical trial design proposed under the first paper.
Description
Date
2022-08-31
Journal Title
Journal ISSN
Volume Title
Publisher
University of Kansas
Collections
Archive Status
This item contains archived web content.
Files
Research Projects
Organizational Units
Journal Issue
Keywords
Biostatistics, Statistics, Health sciences, adaptive designs, Bayesian trials, non-proportional hazards, sample size, time-to-event