Abstract
Group sequential designs are used to potentially shorten randomized clinical trials and thereby reduce subject burden, improve safety, and save time and resources. Clinical trials comparing treatments for systemic lupus erythematosus (SLE) might adopt such designs if the ordinal outcome scales for SLE, such as the Systemic Lupus Activity Measure and Systemic Lupus Erythematosus Disease Activity Index, were more like continuous outcome scales with interval properties. After describing the basic features of sequential trials and highlighting some major issues in their design, we propose approaches that mitigate these issues. In particular, high-speed computing has accelerated advances in sequential design, making available a variety of designs that can be implemented with minimal technical support. The challenge now is to understand the concepts behind such flexible designs and then to apply them to improve studies of SLE.
Highlights
Terminating a clinical trial as soon as a robust result becomes evident is an ethical and practical imperative and minimizes exposure of volunteer participants to potentially ineffective or toxic treatment
Group sequential clinical trial designs are a means to this end
A decrease in systolic blood pressure from 240 to 200 mm Hg has a different meaning and clinical significance than a decrease from 140 to 100 mm Hg. Both the absolute and the relative changes are needed to interpret a 40 mm Hg blood pressure drop. This paper addresses this problem in the context of a sequential randomized clinical trial
Summary
Terminating a clinical trial as soon as a robust result becomes evident is an ethical and practical imperative and minimizes exposure of volunteer participants to potentially ineffective or toxic treatment. Because the O’BrienFleming rule accounts for this correlation, the actual overall type I error is 5 %, even though the sum of the alpha values is 6.6 % We pay for this with a small increase in total sample size; if a one-stop design needs 1000 subjects, this sequential design needs 1024 subjects, a 2.4 % increase. The O’Brien-Fleming test increases sample size, but provides three chances to stop early, but not for futility—that is, stopping early because the treatment difference is so small that gathering more data as planned has little or no chance to reject the null hypothesis [5]. Bayesian designs High-speed computing allows us to explore many sets of cutoff Z-values to either reject the null hypothesis or declare futility. European Consensus Lupus Activity Measure aListing of items for each scale, and minimum and maximum scores obtained from Lam and Petri [1]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
