Quantifying the Utility of Adaptive Designs

Abstract

Adaptive designs (AD) are a broad class of trial designs that allow pre-planned modifications to be made to a trial as patient data is accrued, without undermining its validity or integrity. AD’s may lead to improved efficiency, patient-benefit and power of the trial. However these advantages may be attenuated by a delay in observing the primary outcome variable. In the presence of such delay, we have to choose whether to (a) pause recruitment until requisite data is accrued for the interim analysis, leading to a longer trial completion period; or (b) continue to recruit patients, which may result in a large number of participants who do not effectively benefit from the interim analysis. In this case, little work has investigated the size of outcome delay that results in the realised efficiency gains of AD’s being negligible compared to classical fixed-sample alternatives. This thesis therefore covers different kinds of AD’s and the impact on them of outcome delay. The thesis first explores Simon’s two-stage design for single-arm trials with Bernoulli data. A selection of recently conducted phase II oncology trials is used to assess the impact of delay in practice, while delay optimal designs are also proposed. This work is then extended to group-sequential designs with Normally distributed outcome data. It is observed that for two-arm group-sequential designs, even small levels of outcome delay can have a significant impact on the trial’s efficiency. To obtain maximum efficiency gain from introducing interim analyses into a simple RCT, it is argued the delay length should not be more than 25% of the total recruitment length. The next part of the thesis shifts to focusing on sample size re-estimation(SSR), a design where the variable to optimize is not the expected sample size. Accordingly, we propose an alternative metric to evaluate the efficiency of a SSR design and assessed its efficiency through extensive simulation. The findings indicate that delay has very little impact on SSR trials. However, it is observed that if the sample size has been over-estimated at the beginning of the trial, outcome delay can quickly reduce the trial efficiency to a large extent. Finally, in light of the thesis findings, we discuss the implications of using the ratio of the total recruitment length to the outcome delay as a measure of the utility of different adaptive designs