Clinical evidence from two phase 3 trials supporting statistical adjustment methods to assess confounding impact of treatment crossover on overall survival (OS).

Abstract

e22509 Background: In controlled trials that allow patients to crossover from placebo (PBO) to active treatment (ACT) after progression, OS in PBO patients is confounded by ACT. Statistical methods to adjust OS for crossover rely on an assumption of common treatment effect (i.e. PBO to ACT crossover after progression has the same impact as ACT started at randomization). This exploratory analysis of two phase 3 trials (PBO-controlled and allowing crossover) with regorafenib in GIST and sorafenib in thyroid cancer provides evidence to support this assumption. Methods: Target lesions were assessed at baseline and at regular intervals post baseline. Progression-free survival (PFS) was assessed during double-blind (DB) treatment and secondary PFS (SPFS) in PBO patients during open-label (OL) treatment. Changes in target lesion diameter over time were estimated by a parametric model. Early tumor growth rate (TGR), defined as percent change from baseline/month in the sum of target lesion diameters, was calculated as the model curve slope at earliest time on the TGR curve. Results: SPFS during OL in PBO patients is closer to DB PFS in ACT patients than to DB PFS in PBO patients (Table). Early TGR during OL in PBO patients is closer to DB early TGR in ACT patients than to DB early TGR in PBO patients. Conclusions: Currently available statistical methods to adjust for the impact of treatment crossover are supported by clinical results from two independent randomized, controlled phase 3 studies. These results provide validity to crossover correction analyses. Clinical trial information: NCT01271712; NCT00984282. [Table: see text]