OBJECTIVES:   <span">A placebo-controlled randomised trial (GRID) compared regorafenib and best supportive care (BSC) to BSC alone in the treatment of metastatic/unresectable gastrointestinal stromal tumours in patients previously treated with imatinib and sunitinib. Regorafenib significantly improved progression-free survival; however, overall survival (OS) did not differ significantly between arms. By 8^thJune 2015, 87.9% of BSC patients switched to regorafenib post-progression; switching occurred at a median of 0.9 months after randomisation. A cost-effectiveness analysis was conducted adjusting for crossover. METHODS:   <span">OS was modelled using patient-level data from GRID. Two methods of crossover adjustment were implemented: Iterative Parameter Estimation (IPE) and Rank-Preserving Structural Failure Time (RPSFT). Hazard ratios (HRs) were estimated using a Weibull model and logrank test respectively for IPE and RPSFT, with stratification and recensoring. Parametric models were fitted to the patient-level data. A partitioned survival model was used to model three health states: progression-free, progressed, and dead, over a lifetime horizon. Utilities were derived from trial data, resource use was based on a physician survey, and management costs taken from published sources using an English NHS payer perspective. The main outcome was cost per QALY gained. The price of regorafenib reflected a patient access scheme. RESULTS:   <span">Prior to crossover adjustment a HR of 0.91 (95%CI: 0.65-1.27) was produced with no median OS gain versus BSC alone. IPE and RPSFT adjustment methods produced HRs of 0.56 (95%CI: 0.39-0.78) and 0.59 (95%CI: 0.42-0.83), respectively with a median OS gain of 7.3 and 6.1 months versus BSC alone. Parametric extrapolation beyond the trial was applied, leading to ICERs of £34,420 and £40,188 per QALY gained for IPE and RPSFT crossover corrections, respectively. CONCLUSIONS:   <span">In this trial with frequent and early crossover, statistical correction made a substantial difference to modelled cost-effectiveness. Failing to apply appropriate correction for crossover can lead to misleading cost-effectiveness estimates.