Skip to main content

Foundational Trials

Villanuev

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Transfusion Strategies for Acute Upper Gastrointestinal Bleeding
  • Year: 2013
  • Journal published in: New England Journal of Medicine
  • Citation: Villanueva C, Colomo A, Bosch A, Concepción M, Hernandez-Gea V, Aracil C, Graupera I, Poca M, Alvarez-Urturi C, Gordillo J, Guarner-Argente C, Santaló M, Muñiz E, Guarner C. Transfusion strategies for acute upper gastrointestinal bleeding. N Engl J Med. 2013;368(1):11-21.

Context & Rationale

  • Background
    Red-cell transfusion is common in acute upper gastrointestinal bleeding (UGIB), but optimal haemoglobin thresholds were uncertain; extrapolation from critical care “restrictive” trials was limited because acute GI bleeding populations were excluded, and there were mechanistic concerns that transfusion might worsen portal pressure and rebleeding in portal-hypertensive haemorrhage.
  • Research Question/Hypothesis
    In severe acute UGIB, does a restrictive red-cell transfusion strategy (threshold 7 g/dL; target 7–9 g/dL) improve outcomes compared with a liberal strategy (threshold 9 g/dL; target 9–11 g/dL)?
  • Why This Matters
    Transfusion thresholds influence mortality, rebleeding, complications (especially cardiopulmonary overload), and blood utilisation; clarifying thresholds is particularly consequential for cirrhosis/variceal bleeding where physiology suggests harm from over-transfusion.

Design & Methods

  • Research Question: In adults with severe acute UGIB, does transfusing at Hb <7 g/dL (target 7–9) versus Hb <9 g/dL (target 9–11) reduce all-cause mortality and further bleeding?
  • Study Type: Randomised, single-centre, pragmatic, open-label trial (Hospital de la Santa Creu i Sant Pau, Barcelona) with concealed allocation; stratified by presence/absence of cirrhosis; intention-to-treat analysis.
  • Population:
    • Adults (>18 years) with haematemesis (or bloody nasogastric aspirate), melaena, or both, confirmed by staff; enrolled June 2003–December 2009; urgent gastroscopy within 6 hours.
    • Key exclusions included: exsanguinating haemorrhage requiring immediate transfusion; recent acute coronary syndrome/stroke/TIA/peripheral ischaemia; recent transfusion; recent trauma/surgery; lower GI bleeding; low-risk presentation (clinical Rockall score 0 with Hb ≥12 g/dL). 1
  • Intervention:
    • Restrictive transfusion strategy: transfuse packed red cells when Hb <7 g/dL; post-transfusion target 7–9 g/dL; 1 unit at a time with reassessment. 1
  • Comparison:
    • Liberal transfusion strategy: transfuse when Hb <9 g/dL; post-transfusion target 9–11 g/dL; 1 unit at a time with reassessment; both arms permitted “off-protocol” transfusion for anaemia symptoms/signs, massive bleeding, or surgery. 1
  • Blinding: Open-label (strategy-based transfusion; blinding not feasible); primary endpoint objective (all-cause mortality).
  • Statistics: 430 patients/group (860 total) required to detect ≥5% absolute mortality difference (assumed 10% mortality in liberal arm) with two-sided α=0.05 and β=0.20 (80% power); analysed by intention-to-treat; time-to-event compared via Kaplan–Meier/log-rank and Cox models adjusted for prespecified baseline risk factors. 12
  • Follow-Up Period: 45 days (primary endpoint within 45 days; survival presented to 6 weeks).

Key Results

This trial was not stopped early. 921 randomised; 32 withdrawn after randomisation; intention-to-treat analysis included 444 (restrictive) and 445 (liberal).

Outcome Restrictive (Hb <7; target 7–9) (n=444) Liberal (Hb <9; target 9–11) (n=445) Effect p value / 95% CI Notes
Death from any cause within 45 days 23 (5%) 41 (9%) HR 0.55 95% CI 0.33 to 0.92; P=0.02 Primary outcome; time-to-event analysis
Further bleeding (overall) 45/444 (10%) 71/445 (16%) HR 0.62 95% CI 0.43 to 0.91; P=0.01 Definition required haemodynamic instability or Hb drop ≥2 g/dL within 6 hours
Adverse events (any) 179 (40%) 214 (48%) HR 0.73 95% CI 0.56 to 0.95; P=0.02 Overall complications lower with restrictive strategy
Transfusion reactions 14 (3%) 38 (9%) HR 0.35 95% CI 0.19 to 0.65; P=0.001 Included TACO and allergic reactions
Transfusion-associated circulatory overload (TACO) 2 (<1%) 16 (4%) HR 0.06 95% CI 0.01 to 0.45; P=0.001 Major driver of cardiopulmonary harm signal
Cardiac complications (ACS, pulmonary oedema, arrhythmias) 49 (11%) 70 (16%) HR 0.64 95% CI 0.43 to 0.97; P=0.04 Myocardial infarction: 0 vs 8 (2%) (supplement)
Length of hospital stay 9.6 ± 8.7 days 11.5 ± 12.8 days Difference −1.9 days P=0.01 Mean ± SD
Blood utilisation (RBC units per patient) 1.5 ± 2.3 3.7 ± 3.8 Difference −2.2 units P<0.001 No transfusion: 225/444 (51%) vs 61/445 (14%)
  • Between-group separation was clinically meaningful: discharge Hb 9.2 ± 1.2 g/dL (restrictive) vs 10.1 ± 1.0 g/dL (liberal); mean RBC units 1.5 ± 2.3 vs 3.7 ± 3.8; no-transfusion proportion 51% vs 14%.
  • Mortality benefit appeared concentrated in cirrhosis Child–Pugh A/B: 5/113 (4%) vs 13/109 (12%), HR 0.30; 95% CI 0.11 to 0.85; P=0.02; no clear benefit in Child–Pugh C: 10/26 (38%) vs 12/29 (41%), HR 1.04; 95% CI 0.45 to 2.37; P=0.91.
  • Portal-pressure physiology aligned with outcomes: in variceal bleeding, hepatic venous pressure gradient increased in the liberal arm (20.5 ± 3.1 to 21.4 ± 4.3 mmHg; P=0.03) but not in the restrictive arm.

Internal Validity

  • Randomisation and allocation concealment: Computer-generated random numbers; sealed, consecutively numbered, opaque envelopes; stratified by cirrhosis; blocks of four. 1
  • Attrition / post-randomisation exclusions: 921 randomised; 32 withdrawn after randomisation; analysed 444 vs 445 by intention-to-treat.
  • Performance/detection bias: Open-label strategy could influence co-interventions and thresholds for declaring “further bleeding” or complications; mitigated by standardised definitions and objective primary endpoint (all-cause mortality). 12
  • Protocol adherence: Major protocol violation occurred more often with restrictive strategy (39/444 [9%] vs 15/445 [3%]; P<0.001); nonetheless, deviations were <10% in each group and analyses were intention-to-treat.
  • Baseline comparability / illness severity: Groups were similar at baseline (e.g., mean complete Rockall 5.3 ± 2.0 vs 5.4 ± 1.7; shock at admission 28% vs 31% in supplement); population enriched for higher-risk bleeding (low-risk Rockall 0 with Hb ≥12 excluded). 2
  • Timing: Rapid enrolment and endoscopy: randomisation ~19 ± 7 vs 18 ± 6 minutes from admission; gastroscopy ~5 ± 6 vs 5 ± 7 hours. 2
  • Dose/separation of variable of interest: Clear haemoglobin exposure difference across admission (lowest Hb 7.3 ± 1.4 vs 8.0 ± 1.5 g/dL; discharge Hb 9.2 ± 1.2 vs 10.1 ± 1.0; P<0.001 for both) and RBC exposure (1.5 ± 2.3 vs 3.7 ± 3.8 units; P<0.001).
  • Adjunctive therapies: Protocolised acute management reduced contamination: emergency endoscopy within 6 hours; high-dose IV omeprazole for peptic ulcer; vasoactive therapy (somatostatin) and antibiotics for suspected portal hypertension; rescue pathways specified (repeat endoscopy/surgery for non-variceal; balloon tamponade/TIPS for variceal). 1
  • Outcome assessment/statistical rigour: Time-to-event methods with prespecified Cox adjustment factors (mortality model adjusted for age, in-hospital bleeding, cirrhosis, Rockall score; further bleeding model included shock at admission and baseline Hb). 2

Conclusion on Internal Validity: Overall internal validity appears moderate-to-strong: allocation concealment and objective primary endpoint support causal inference, with the main limitation being open-label delivery and higher (though still <10%) protocol deviations in the restrictive arm.

External Validity

  • Population representativeness: Adults with “severe” UGIB (median-to-high Rockall; 31% cirrhosis; 49% peptic ulcer; 21% oesophageal varices) managed in a specialist bleeding unit with rapid access to endoscopy and (for variceal bleeding) portal pressure measurements in a subset.
  • Key exclusions affecting generalisability: Exsanguinating bleeding requiring immediate transfusion; recent acute coronary syndrome/stroke/TIA/peripheral ischaemia; low-risk presentations; and prior recent transfusion—limiting inference to haemodynamically stabilisable bleeding without very recent high-risk ischaemia. 1
  • Systems/setting dependence: Benefits likely depend on timely endoscopic and pharmacological haemostasis (e.g., urgent endoscopy, vasoactive agents, PPI) and availability of rescue interventions; applicability may be attenuated where such pathways are delayed or unavailable.

Conclusion on External Validity: Findings are highly generalisable to hospitalised, non-exsanguinating UGIB managed with modern haemostatic pathways; generalisability is limited for massive haemorrhage and patients with very recent major ischaemic events.

Strengths & Limitations

  • Strengths: Large randomised dataset for UGIB; concealed allocation with stratification for cirrhosis; clinically meaningful separation in haemoglobin and transfusion exposure; objective primary endpoint; prespecified subgroup analyses; mechanistic portal-pressure substudy consistent with clinical outcomes.
  • Limitations: Single-centre and open-label; post-randomisation withdrawals (32) and more protocol violations in restrictive arm (9% vs 3%); exclusions limit applicability to exsanguinating bleeding and very recent high-risk cardiovascular events; some secondary outcomes could be influenced by care decisions in an unblinded trial.

Interpretation & Why It Matters

  • Practice-changing transfusion trigger
    A restrictive threshold (transfuse at Hb <7 g/dL; target 7–9) improved 45-day survival (95% vs 91%; HR 0.55) and reduced further bleeding (10% vs 16%) while substantially reducing RBC exposure (mean 1.5 vs 3.7 units; 51% received none).
  • Mechanism aligns with cirrhosis physiology
    In variceal bleeding, liberal transfusion was associated with increased portal-pressure gradient and higher rebleeding/rescue therapy use, supporting a portal-hypertensive mechanism for harm from over-transfusion.
  • Safety signal favours restriction
    Liberal strategy increased transfusion reactions (9% vs 3%) and TACO (4% vs 0.5%), consistent with avoidable iatrogenic cardiopulmonary complications.

Controversies & Subsequent Evidence

  • Unblinded design vs objective endpoint: Concerns that open-label management could influence rebleeding/rescue decisions and complication ascertainment were raised in published correspondence, although mortality (primary outcome) is less susceptible to bias. 3
  • Cardiovascular disease uncertainty: Exclusion of very recent acute coronary syndrome/stroke/TIA and limited data in high-risk ischaemic subgroups meant the “one-size-fits-all” Hb <7 threshold remained debated, driving guideline nuance (e.g., higher thresholds in known cardiovascular disease). 78
  • Subgroup interpretation (cancer/cirrhosis severity): Benefit was clear in Child–Pugh A/B cirrhosis but not in Child–Pugh C (HR 1.04; wide CI), raising questions about competing risks and whether transfusion strategy meaningfully alters outcomes in advanced liver failure.
  • Subsequent trial evidence: The TRIGGER cluster-randomised feasibility trial demonstrated that implementing hospital-level transfusion policies in UGIB is feasible but highlighted practice variability and the need for adequately powered effectiveness trials. 5
  • Meta-analytic synthesis: A large systematic review/meta-analysis of RCTs (including re-analysed cluster data) found restrictive strategies reduced mortality and rebleeding overall, with no clear increase in ischaemic events. 4
  • Guideline uptake: Contemporary UGIB guidance widely endorses restrictive RBC transfusion around Hb ≤7 g/dL in patients without cardiovascular disease, with more liberal thresholds (≈8 g/dL) when cardiovascular disease is present. 789

Summary

  • Randomised trial in severe acute UGIB comparing transfusion at Hb <7 g/dL (target 7–9) versus Hb <9 g/dL (target 9–11).
  • Restrictive strategy improved 45-day survival: 5% vs 9% mortality; HR 0.55; 95% CI 0.33 to 0.92; P=0.02.
  • Restrictive strategy reduced further bleeding: 10% vs 16%; HR 0.62; 95% CI 0.43 to 0.91; P=0.01.
  • Marked reduction in blood exposure: mean 1.5 ± 2.3 vs 3.7 ± 3.8 units; 51% vs 14% received no transfusion.
  • Harms favoured restriction: fewer adverse events (40% vs 48%) and substantially fewer transfusion reactions/TACO.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • Where cardiovascular disease is present, multiple guidelines endorse a more liberal transfusion trigger (≈8 g/dL) than in the Villanueva trial’s liberal arm (9 g/dL), reflecting ongoing concern about myocardial ischaemia risk in high-risk subgroups.

Overall Takeaway

In severe hospitalised acute UGIB managed with rapid endoscopy and contemporary haemostatic care, a restrictive transfusion strategy (Hb <7 g/dL; target 7–9) improved survival, reduced rebleeding, and decreased transfusion-related harms while substantially reducing blood use. The trial’s mechanistic coherence (portal-pressure rise with liberal transfusion) and consistent downstream effects on rescue therapies helped cement restrictive transfusion as a cornerstone of modern UGIB management, with guideline nuance for patients at high cardiovascular risk.

Overall Summary

  • Restrictive transfusion (Hb <7 g/dL; target 7–9) reduced 45-day mortality (5% vs 9%) and further bleeding (10% vs 16%) in severe acute UGIB.
  • Blood exposure fell markedly (mean 1.5 vs 3.7 units; 51% vs 14% received no transfusion) with fewer transfusion complications (including TACO).
  • Benefit was clearest in cirrhosis Child–Pugh A/B and aligned with portal-pressure physiology, supporting harm from over-transfusion in portal hypertension.

Bibliography