Estimation of progression-free survival in the randomized discontinuation trial design.

Abstract

TO THE EDITOR: Nosov et al recently published an article reporting on a study with a randomized discontinuation trial (RDT) design. Inthisdesign,allpatients initially receivedtivozanib for16weeks.After16 weeks, patients who responded (response was considered as 25% tumor shrinkage) continued on tivozanib, those with progressive disease were removed from the study, and those with stable disease were randomly assigned to either continue tivozanib or to be given a placebo. In Figure 4 of the article, a Kaplan-Meier curve is presented as an estimate of the progression-free survival (PFS) rate of all patients who were treated with tivozanib, measured from the time of initial dosing. However, this estimate may be biased, given that the method described was to simply censor patients who were assigned to the placebo at the time of random assignment. The reason for the bias is that the censoring is selectively performed only in the stable disease subgroup. If these patients have subsequent survival rates that are lower than those with initial tumor shrinkage (a likely possibility), the estimate will be positively biased. The solution is to combine the PFS estimate for the first 16 weeks (based on all patients) with a weighted average of the PFS rates beyond 16 weeks intherespondingandstabledisease/continually treatedsubgroups, asdescribedinthearticlebyRatainetal (seealsothearticlebyKarrisonet al). In the study by Nosov et al, 78 patients had an initial response and 118 had stable disease. The extent of the bias would depend on the magnitude of the difference in subsequent PFS in these two subgroups.