Pavlakis 2016.

Study characteristics
Methods	International double‐blind placebo‐controlled phase 2 RCT Kaplan‐Meier estimate of median follow‐up was 17.1 months (95% CI 14.6 to 19.4)
Participants	Diagnosis: metastatic or locally recurrent OGJ or stomach adenocarcinoma or undifferentiated histology Previous treatment status: refractory to ≤ 2 lines chemotherapy including prior fluoropyrimidine and platinum. Whether participants had received fluoropyrimidine and platinum concurrently was not described. Number of participants: 147 enrolled between 7th November 2012 and 25th February 2014 Mean age: 63 (31 to 81) years in Arm A and 62 (32‐85) years in Arm B Number of male participants: 118 ECOG PS status: 62 PS0 and 85 PS1 Location of primary cancer: 56 OGJ and 91 gastric Study locations: 53 centres in 4 countries worldwide (Australia, New Zealand, Canada and South Korea) Other relevant key eligibility criterion was measurable disease according to RECIST v1.1 by CT within 21 days before random assignment. Participants with poorly controlled hypertension, prior anti‐VEGF therapy and uncontrolled CNS disease were excluded. Of 152 randomly allocated participants, 4 in regorafenib arm and 2 in placebo arm discontinued treatment due to withdraw of consent.
Interventions	Arm A (97 participants): oral regorafenib 160 mg once daily on Day 1‐21 of 28‐day cycle Arm B (50 participants): oral placebo once daily on Day 1‐21 of 28‐day cycle Study treatment was continued until progression or prohibitive toxicity.
Outcomes	Primary outcome: PFS Secondary outcome: ORR, clinical benefit status at 2 months (defined as receiving treatment and remaining progression‐free during Week 6‐10), OS, adverse events, QoL
Notes	Participants receiving placebo who were subsequently unblinded because of documented PD were able to receive open‐label regorafenib provided they continued to meet protocol=specified safety and PS criteria. 29 pxs (58%) receiving placebo chose regorafenib Rx after PD. Supported by unrestricted research grant from Bayer HealthCare Pharmaceuticals to the Australasian Gastro‐Intestinal Trials Group to conduct the study independently and by National Health and Medical Research Council (NHMRC) Program Grant ACTRN12612000239864
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation was performed using a permuted‐blocks method and an independent statistician monitored random assignment system.
Allocation concealment (selection bias)	Low risk	Random assignment was performed centrally via a web‐based system, maintaining allocation concealment.
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Double‐blind placebo‐controlled trial: regorafenib and matching placebo were provided by manufacturer to sites, labelled with unique kit numbers, which were assigned to participants on random assignment.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Double‐blind placebo‐controlled trial: responses were assessed locally using RECIST v1.1, blinded central review of all RECIST source documents was undertaken for all participants for date of progression and 13% of participants were randomly selected for blinded central review by 2 independent radiologists.
Imbalance in baseline characteristics across arms (confounding)	Low risk	Randomisation was stratified by lines of prior chemotherapy for advanced disease (one v two) and geographic region (Australia, New Zealand or Canada vs South Korea). The groups were balanced at baseline including for median plasma VEGF (≤ 0.14 vs > 0.14 pg/mL).
Incomplete outcome data (attrition bias) for outcomes other than QoL	Low risk	Efficacy and safety analyses was performed on all the randomised participants excluding those not meeting inclusion criteria. 3 participants in Arm A (regorafenib) were excluded based on absence of measurable disease (2) and more than 2 lines of previous chemotherapy and 2 participants in Arm B (placebo) were excluded as their primary tumours were distal oesophageal in origin.
Incomplete outcome data (attrition bias) for QoL outcome	High risk	96% and 67% of the participants who consented to participate in the QoL sub‐study completed a baseline QoL assessment, and at least one post‐baseline QoL assessment.
Selective reporting (reporting bias)	Low risk	All the relevant predefined outcomes were reported
Other bias	Low risk	No obvious potential source of bias