Introduction Transfusion thresholds for upper gastrointestinal bleeding (UGIB) are controversial. Observational studies suggest associations between liberal red blood cell (RBC) transfusion and adverse outcome, and a recent trial reported increased mortality following liberal transfusion.

Methods Pragmatic cluster randomised trial to evaluate the feasibility and safety of implementing a restrictive (transfusion when haemoglobin (Hb) <8g dL) versus liberal (transfusion when Hb <10 g/dL) RBC transfusion policy for UGIB. Hospitals were randomised to a policy which was implemented through a multi-faceted educational intervention targeting all staff caring for patients with UGIB. All adult patients were eligible to participate, regardless of co-morbidity; the only exclusion criterion was exsanguinating haemorrhage. Feasibility and exploratory clinical outcomes were recorded up to day 28.

Results 936 patients were enrolled in 6 hospitals. The consent rate for follow up was higher in the liberal arm (62% vs. 55%; P = 0.04). There were some baseline imbalances, however the Rockall and Blatchford scores were identical between arms as was the prevalence of ischaemic heart disease (IHD) (14% liberal arm vs. 15% restrictive arm). Protocol adherence was 96% in the restrictive arm vs. 83% in the liberal arm. In patients with a Hb <12 g/dL, the Hb at discharge was 10.7 g/dL (sd 1.2) in the liberal arm vs. 10.1 g/dL (sd 1.3) in the restrictive arm (P = 0.05). In the restrictive arm there was a 13% absolute reduction in the proportion of patients receiving RBCs (95% CI for difference in% -35 to 11, P = 0.23) with a mean of 0.8 fewer RBC units transfused (95% CI: -1.9 –to 0.3, P = 0.12). Clinical outcomes were better in the restrictive arm, although differences were not statistically significant (28-day mortality, 7% liberal vs. 5% restrictive, adjusted difference in% = -1.3, 95% CI: -8.0 to 5.5, P = 0.63; further bleeding, 9 vs. 5%, P = 0.29; serious adverse events, 22% vs. 18%, P = 0.48). In the subgroup with IHD, there was a large observed difference for mortality (12% restrictive arm (n = 6) vs. 3% liberal arm (n = 2); interaction P = 0.11).

Conclusion The protocol was feasible and generated clinically important differences in the level of anaemia and RBC exposure. There was a consistent trend towards fewer complications in the restrictive arm, apart from the increased mortality observed in patients with IHD, which could in part be explained by imbalances in baseline risk. A large trial is required to clarify the risk-benefit balance before advocating restrictive RBC transfusion for all patients with UGIB.

Disclosure of Interest None Declared.