Jairath 2015.

Study characteristics
Methods	Design: RCT, pragmatic, open‐label, feasibility, cluster design trial Setting: 6 university teaching hospitals in the UK Recruitment: September 2012 to March 2013 Maximum follow‐up: 28 days
Participants	6 clusters including 936 participants. This trial did not require participants to meet a haemoglobin threshold for enrolment Participants were 18 years of age or older and were admitted to 1 of the participating hospitals with upper GI bleeding Participants with exsanguinating haemorrhage were excluded. Participants consented to permit data collection and follow‐up Liberal threshold: n = 533; mean (SD) age = 60.4 (20.0) years Restrictive threshold: n = 403; mean (SD) age = 58.0 (20.3) years
Interventions	Liberal group: 10 g/dL threshold Restrictive group: 8 g/dL threshold
Outcomes	Feasibility outcomes included: recruitment rates, adherence to transfusion policy, difference in Hb, RBC exposure, evidence for selection bias Clinical outcomes included: further bleeding, thromboembolic and ischaemic events, number of infections, mortality, serious adverse events, health‐related quality of life
Notes	Trial title: Transfusion in Gastrointestinal Bleeding (TRIGGER) Trial registration (prospective): ISRCTN85757829 Trial funding: NHS Blood and Transplant Research and Development COI statement by investigators: "we declare no competing interests" (p 144)
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"We randomly allocated (using a computer‐generated randomisation sequence) centres to a transfusion policy using a random permuted block of six (three hospitals per policy), without stratification or matching (randomisation done by BCK)"
Allocation concealment (selection bias)	High risk	"All clinicians, patients, and outcome assessors were unmasked to treatment allocation" (Jairath 2015 p 2)
Blinding of participants and personnel (performance bias) Objective outcomes	Low risk	Blinding of personnel for this intervention is not feasible, but in our view, for objective outcomes such as mortality (the primary outcome used within this review), we graded risk of bias as 'low' "TRIGGER was an open‐label cluster randomised trial whose primary outcome was further bleeding. Because of the cluster randomisation, all researchers in a hospital were aware of treatment allocation and so could not perform a blinded assessment. A blinded adjudication committee was also not feasible as it was impossible to compile relevant information to send to the committee in a blinded manner. Therefore, the definition of further bleeding was modified to exclude subjective aspects (such as whether symptoms like vomiting blood were severe enough to indicate the outcome had been met), leaving only objective aspects (the presence versus absence of active bleeding in the upper gastrointestinal tract confirmed by an internal examination)"
Blinding of outcome assessment (detection bias) Objective measures	Low risk	Blinding of mortality (the primary outcome used within this review) is not relevant
Blinding of outcome assessment (detection bias) Subjective measures	Low risk	Quality of life measures were reported. As described above, blinding was not possible for a cluster trial, and it is unclear how assessors could use this information in a selective way
Incomplete outcome data (attrition bias) All outcomes	Low risk	533 were enrolled in the liberal arm, 403 in the restrictive arm. All participants were analysed for feasibility outcomes. 512/533 of one arm and 393/403 were analysed for further bleeding at 28 days. The intraclass correlation coefficient (ICC) was very low (0.0001) for the outcome of mortality; we therefore included data considering the participant as the unit of randomisation and ignored clustering, but we performed a sensitivity analysis excluding this trial to see what effect, if any, it had on the analysis Rates of return for EQ‐5D data at 28 days (by telephone) were low (237/533 and 267/403), but these results were as expected for this patient population
Selective reporting (reporting bias)	Unclear risk	Trial was prospectively registered. No reporting bias was apparent
Other bias	Low risk	No other biases noted