Final overall survival (OS) analysis with modeling of crossover impact in the phase III GRID trial of regorafenib vs placebo in advanced gastrointestinal stromal tumors (GIST).

Abstract

156 Background: GRID showed that regorafenib (REG) improves progression-free survival (PFS; primary endpoint) vs placebo (PBO) in patients (pts) with advanced GIST after failure of at least imatinib and sunitinib (HR 0.27; 1-sided p < 0.0001). An interim OS analysis at the time of the primary PFS analysis showed a positive trend (HR 0.77; p = 0.199) despite 85% of PBO pts crossing over to REG. An exploratory analysis modeling the impact of crossover on OS suggested a benefit for REG. We present exploratory analyses of OS comparing crossover correction results at different times, including at the planned final OS analysis. Methods: Data cut-off dates for OS analyses were 26 Jan 2012 (final PFS analysis), 31 Jan 2014, and 8 Jun 2015 (final OS analysis). The impact of crossover on OS was modeled using 2 randomization-based methods: rank preserving structural failure time (RPSFT) and iterative parameter estimation (IPE), both considered best choice among correction analytics. Hazard ratios (HRs) and 95% CIs were derived using the Cox model. Results: Pts were randomized (2:1) to REG (n = 133) or PBO (n = 66). Data maturity in terms of deaths of randomized pts was 23%, 70%, and 81%. At data cut-off for the final OS analysis, a total of 162 deaths occurred: REG 109 (82%) and PBO 53 (80%); 7 pts remained on treatment. Compared to the 2012 OS analysis, continued crossover treatment further diluted the treatment effect, evidenced by the increasing HR in the ITT analysis at later times (Table). Conclusions: IPE correction provides more stable estimated median times and HRs than RPSFT. These exploratory analyses modeling the impact of crossover to active drug suggest that REG has a greater OS benefit than noted in the ITT analysis. Such sensitivity analytics are needed to understand fully the benefits of active drugs in crossover trials. Clinical trial information: NCT01271712. [Table: see text]