Correcting Overall Survival for the Impact of Crossover Via a Rank-Preserving Structural Failure Time (RPSFT) Model in the RECORD-1 Trial of Everolimus in Metastatic Renal-Cell Carcinoma

Abstract

Clinical trials in oncology often allow patients randomized to placebo to cross over to the active treatment arm after disease progression, leading to underestimation of the treatment effect on overall survival as per the intention-to-treat analysis. We illustrate the statistical aspects and practical use of the rank-preserving structural failure time (RPSFT) model with the Fleming-Harrington family of tests to estimate the crossover-corrected treatment effect, and to assess its sensitivity to various weighting schemes in the RECORD-1 trial. The results suggest that the benefit demonstrated in progression-free survival is likely to translate into a robust overall survival benefit.

A pooled analysis of overall survival in COMFORT-I and COMFORT-II, 2 randomized phase III trials of ruxolitinib for the treatment of myelofibrosis

“…Here, we used the RPSFT method to evaluate the impact of crossover on the OS analysis; RPSFT is an accepted method to correct for crossover in randomized trials and has been used in other studies. 17,18 The RPSFT analysis supports the hypothesis that the crossing over of placebo-or BAT-treated patients to ruxolitinib may have led to underestimating the actual effect of ruxolitinib on survival.…”

“…Here, we used the RPSFT method to evaluate the impact of crossover on the OS analysis; RPSFT is an accepted method to correct for crossover in randomized trials and has been used in other studies. 17,18 The RPSFT analysis supports the hypothesis that the crossing over of placebo-or BAT-treated patients to ruxolitinib may have led to underestimating the actual effect of ruxolitinib on survival.A potential confounding factor to interpretation of the survival advantage lies in the assumption that patients who were randomized to ruxolitinib may have received more supportive care, whereas treating investigators had more experience administering medications in the BAT arm and were more versed in the safety profiles of these agents. The focus on patients receiving an investigational agent may have led to comorbidities being more rapidly identified and treated, potentially affecting survival outcomes.…”

“…The crossovercorrected treatment effect was estimated using a rank-preserving structural failure time (RPSFT) model (Online Supplementary Figure S1). 17,18 OS was evaluated by the degree of spleen size reduction achieved at week 24 in a landmark analysis. 19 The reverse KaplanMeier method was used to analyze censoring patterns in each study.…”

Hard-Wired Bias

“…RECORD-1 demonstrated an improvement in progression free survival, but no change in overall survival 10 . This deficiency was attributed to crossover 10 , and the manufacturer provided modeling experiments arguing what the survival would have been were it not for crossover 11 . However, this exercise relies on assumptions, which may not be true.…”

The misguided ethics of crossover trials

“…2 ” While the trial found a statistically significant difference in time to progression, there was no difference in overall survival, which the authors felt was, “probably due to confounding by crossover.” The challenge here is that by consciously designing the trial with crossover, the investigators would have known that any uncertainty about a mortality benefit could be attributed to the confounding factor they introduced. Nevertheless, a post-hoc statistical analysis performed by the drug’s manufacturer, used a statistical model to correct for crossover, and concluded that everolimus adds 0.93 days of life for every day it is taken 3 .…”