Multiarm, multistage randomized controlled trials with stopping boundaries for efficacy and lack of benefit: An update to nstage

Alexandra Blenkinsop,Babak Choodari-Oskooei

doi:10.1177/1536867x19893616

Alexandra Blenkinsop, Babak Choodari-Oskooei

Open Access

https://doi.org/10.1177/1536867x19893616

Copy DOI

Abstract

Royston et al.’s (2011, Trials 12: 81) multiarm, multistage (MAMS) framework for the design of randomized clinical trials uses intermediate outcomes to drop research arms early for lack of benefit at interim stages, increasing efficiency in multiarm designs. However, additionally permitting interim evaluation of efficacy on the primary outcome measure could increase adoption of the design and result in practical benefits, such as savings in patient numbers and cost, should any efficacious arm be identified early. The nstage command, which aids the design of MAMS trial designs, has been updated to support this methodological extension. Operating characteristics can now be calculated for a design with binding or nonbinding stopping rules for lack of benefit and with efficacy stopping boundaries. An additional option searches for a design that strongly controls the familywise error rate at the desired level. We illustrate how the new features can be used to design a trial with the drop-down menu, using the original comparisons from the MAMS trial STAMPEDE as an example. The new functionality of the command will serve a broader range of trial objectives and increase efficiency of the design and thus increase uptake of the MAMS design in practice.

Highlights

Multiarm, multistage (MAMS) clinical trial designs for time-to-event outcomes result in increased efficiencies in time and resources over traditional two-arm designs (Royston et al 2011) and have been successfully implemented in trials investigating therapies in many disease areas, including oncology (Sydes et al 2009, 2012; Parmar et al 2017). Barthel, Royston, and Parmar (2009) developed the nstage command to assist those designing such a trial. nstage calculates the required sample size and operating characteristics with an intuitive menu-driven approach
To illustrate the updates and demonstrate how the new output from nstage can be interpreted, we present an example below that uses the design specification for the original comparisons in the STAMPEDE trial, which started as a six-arm four-stage MAMS design with I = D (Sydes et al 2012; Parmar et al 2008)
We have demonstrated in this article how the nstage command allows easy specification and implementation of efficacy stopping boundaries to a MAMS design and gives the investigator the appropriate information required to calculate and control the relevant operating characteristics of the design with minimal computation

Summary

Introduction

Multistage (MAMS) clinical trial designs for time-to-event outcomes result in increased efficiencies in time and resources over traditional two-arm designs (Royston et al 2011) and have been successfully implemented in trials investigating therapies in many disease areas, including oncology (Sydes et al 2009, 2012; Parmar et al 2017). Barthel, Royston, and Parmar (2009) developed the nstage command to assist those designing such a trial. nstage calculates the required sample size and operating characteristics with an intuitive menu-driven approach. Multistage (MAMS) clinical trial designs for time-to-event outcomes result in increased efficiencies in time and resources over traditional two-arm designs (Royston et al 2011) and have been successfully implemented in trials investigating therapies in many disease areas, including oncology (Sydes et al 2009, 2012; Parmar et al 2017). The command was updated in 2015 to increase functionality; the update included new features such as estimation of the familywise error rate (FWER) and improved estimation of the correlation between the test statistics of treatment effects (Bratton, Choodari-Oskooei, and Royston 2015). The nature of the design results in an increased probability of a single clinical trial protocol identifying an effective regimen, saves considerable time and resources, and requires fewer patients compared with multiple independent two-arm trials (Parmar, Carpenter, and Sydes 2014). Some examples of internationally known trials designed under the MAMS framework are STAMPEDE, which is the largest ever trial conducted in prostate cancer (Sydes et al 2012), and RAMPART.

Methods

Discussion

Conclusion