Jairath 2015.

Methods	Cluster‐randomised trial pilot study involving 6 hospitals
Participants	Participants were 18 years of age or older and admitted to 1 of the participating hospitals with upper gastrointestinal bleeding. Participants with exsanguinating haemorrhage were excluded. Participants consented to permit data collection and follow‐up.
Interventions	The following transfusion strategy was encouraged at each participating hospital: Liberal 10 g/dL threshold: n = 533; mean (SD) age = 60.4 (20.0) years Restrictive 8 g/dL threshold: n = 403; mean (SD) age = 58.0 (20.3) years
Outcomes	Feasibility outcomes included recruitment rates, adherence to transfusion policy, difference in haemoglobin concentration, RBC exposure, and evidence for selection bias. Clinical outcomes included further bleeding, thromboembolic and ischaemic events, number of infections, mortality, serious adverse events, and health‐related quality of life.
Notes	This was the only cluster trial in our review. It did not require participants to meet a haemoglobin threshold for enrolment.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	The hospital was randomized, not the individual participant.
Allocation concealment (selection bias)	High risk	The hospital was randomized, so everyone knew which arm the participants were in.
Blinding of participants and personnel (performance bias) All outcomes	High risk	The trial was not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Mortality allowed a judgement of low risk of bias. Assessment of other clinical outcomes was unblinded.
Incomplete outcome data (attrition bias) All outcomes	High risk	There was a high percentage of missing data.
Selective reporting (reporting bias)	Unclear risk	No reporting bias was apparent.
Other bias	High risk	There was differential enrolment by treatment arms.