Aliment Pharmacol Ther 2010; 32: 215–224
Background Acute upper gastrointestinal bleeding (AUGIB) accounts for 14% of RBC units transfused in the UK. In exsanguinating AUGIB the value of RBC transfusion is self evident, but in less severe bleeding its value is less obvious.
Aim To examine the relationship between early RBC transfusion, re-bleeding and mortality following AUGIB, which is one of the most common indications for red blood cell (RBC) transfusion.
Method Data were collected on 4441 AUGIB patients presenting to UK hospitals. The relationship between early RBC transfusion, re-bleeding and death was examined using logistic regression.
Results 44% were transfused RBCs within 12 hours of admission. In patients transfused with an initial haemoglobin of <8 g/dl, re-bleeding occurred in 23% and mortality was 13% compared with a re-bleeding rate of 15%, and mortality of 13% in those not transfused. In patients transfused with haemoglobin >8 g/dl, re-bleeding occurred in 24% and mortality was 11% compared with a re-bleeding rate of 6.7%, and mortality of 4.3% in those not transfused. After adjusting for Rockall score and initial haemoglobin, early transfusion was associated with a two-fold increased risk of re-bleeding (Odds ratio 2.26, 95% CI 1.76–2.90) and a 28% increase in mortality (Odds ratio 1.28, 95% CI 0.94–1.74).
Conclusions Early RBC transfusion in AUGIB was associated with a two-fold increased risk of re-bleeding and an increase in mortality, although the latter was not statistically significant. Although these findings could be due to residual confounding, they indicate that a randomized comparison of restrictive and liberal transfusion policies in AUGIB is urgently required.
Acute upper gastrointestinal bleeding (AUGIB) continues to be a common reason for hospital admission with an annual incidence in the UK of 100–180/105.1–3 It is also one of the most common indications for red blood cell (RBC) transfusion, accounting for 14% of all RBC units transfused in the UK.4 Whilst RBC transfusion may be life saving in patients with massive exsanguinating gastrointestinal bleeding, the evidence that it is beneficial in less severe bleeding is lacking. Two small trials have failed to demonstrate benefit and one suggested potential harm with increases in re-bleeding, further transfusion and mortality in those receiving early RBC transfusion.5, 6 Furthermore, in AUGIB secondary to portal hypertension, there is evidence that RBC transfusion to levels of >8 g/dL has an adverse effect to the extent that the consensus guidelines strongly recommend maintaining a haemoglobin of around 8 g/dL.7
There has been extensive work on transfusion triggers in anaemia, but none of these has looked specifically at acute gastrointestinal bleeding. Observational studies and randomized controlled trials of RBC transfusion in other critically ill or bleeding patients have shown little evidence of benefit, but some evidence of harm.8–13 A randomized controlled trial of the effects of transfusion on critical care outcomes led to the widespread implementation of restrictive transfusion strategies in critical care.8 A systematic review of the published randomized trials of restrictive vs. liberal red cell transfusions found significant reductions in blood use in the restrictive groups, without an increase in the rate of adverse clinical outcomes.14 Indeed, there was a 24% reduction in adverse cardiac events in the restrictive transfusion group compared with the liberal transfusion group, which was of marginal statistical significance. Mortality was also lower (20%) in the restrictive vs. the liberal transfusion group, although this was not statistically significant (P = 0.07). We report the results of a prospective observational study that examined the hypothesis that RBC transfusion within 12 h of presentation with AUGIB is independently associated with increased re-bleeding and mortality.
All National Health Service (NHS) hospitals accepting acute admissions in the UK (n = 257) were invited to participate in the study. Of these, 223 agreed and 212 (82%) hospitals submitted data. A list of the hospitals and clinicians contributing data is available at http://www.bsg.org.uk/clinical/general/uk-upper-gi-bleeding-audit.html.
Data were collected prospectively on all adults (16 years or over) who presented with AUGIB between 1 May and 30 June 2007 (Figure 1). For this analysis, all patients included had undergone an upper gastrointestinal endoscopy for AUGIB either when it was the presenting condition or when it developed in in-patients. Definitions are provided in Table 1.
|Acute upper gastrointestinal bleeding (AUGIB)|
|Haematemesis, the passage of melaena and/or firm clinical or laboratory evidence of acute blood loss from the UGI tract. Patients presenting with iron deficiency anaemia without evidence of AUGIB were not included.|
|Early red blood cell (RBC) transfusion|
|RBC transfusion within 12 h of presentation with AUGIB.|
|Further haematemesis, passage of fresh melaena, continuing or recurring hypotension and tachycardia +/− fall in haemoglobin after the first endoscopy|
|All cause mortality|
|Death occurring within the hospital admission up to 30 days post index AUGIB|
A clinical lead in each hospital coordinated a team of case-identifiers and data-collectors. Cases were identified from hospital admission units, endoscopy departments, blood transfusion laboratories and all adult wards by the case-identifiers and preliminary data registered on a purpose designed website. For each patient, following discharge or death, an online questionnaire was then completed by the designated data collectors. Information on demography, clinical (including medical co-morbidity and risk factors for AUGIB), laboratory, resuscitation, transfusion, endoscopy (including endoscopic therapy, re-bleeding), radiology, surgery, length of stay and mortality was extracted from the hospital records. Each hospital was given a unique login and password. At no time did the study group have access to patient records or any patient-identifying data. Compulsory fields on the questionnaire ensured a minimum dataset for completed eligible cases, which included the variables required to calculate the Rockall score.15 This risk score has been internally and externally validated as a predictor of mortality and re-bleeding in patients with AUGIB.16–20 and is widely used in the UK.
Data were exported into SPSS (Chicago, Illinois, USA) for cleaning and analysis, and when necessary, hospital clinical leads were contacted to clarify incomplete data. Duplicate cases based on hospital, admission date and time, year of birth and full blood count values were removed from the dataset. Pre- and post-endoscopy Rockall scores based on age, haemodynamic status at presentation, co-morbidity and endoscopic findings (using data from the first inpatient endoscopy following presentation) were calculated for all patients in the analyses presented.
For this analysis, we excluded patients who received no endoscopy or where data on transfusion within 12 h or death were incomplete. Initial univariable analyses were then conducted to compare risk factors for re-bleeding and mortality between those receiving and not receiving early RBC transfusion. Re-bleeding was defined as any bleeding occurring after the first endoscopy. Categorical variables are presented as % (number) with 95% confidence intervals (95% CI) as necessary and differences were compared by the Chi square test. Normally distributed continuous variables were assessed for normality using a standardized normal probability plot (pnorm command Stata 10, College Station, Texas, USA) and presented as mean (s.d.), compared by an unpaired T test. Continuous variables with a skewed distribution are presented as median (interquartile range), tested by the Mann–Whitney test. We examined the proportion of transfused and non-transfused patients who died, stratified for initial haemoglobin concentration and haemodynamic status.
Univariable and then multivariable logistic regression models examined the relationship between early transfusion, re-bleeding and death. Multivariable models were formed both using individual potential confounders alone or in combination and using the calculated Rockall score instead of its individual components. Confounders were retained when their addition produced a 10% change in the reported effect size. Additional, stratified analyses were conducted to examine for interaction between early transfusion and other factors. All analyses were conducted in Stata SE version 10 (StataCorp, College Station, Texas, USA) with the exception of the calculation of confidence intervals of proportions calculated using CIA version 2 (University of Southampton).
A total of 8939 potential cases were registered online. Of these, 1199 did not meet entry criteria and were excluded by the local hospital clinical leads. Of the 7740 cases remaining, 1190 (14%) were submitted without sufficient data for analysis and were excluded by the project group. Of the remaining 6750 cases, 5004 underwent endoscopy and complete information concerning RBC transfusion was available in 4441 (Figure 1). There were no important differences between the demographic characteristics of the patients included and those excluded for reasons of incomplete data. Patients with missing RBC transfusion data (n = 563) had a slightly lower median age (67 years, IQR 47–78) vs. 69 years, (IQR 52–80); P = 0.005) and a higher proportion of female patients (44% vs. 39% female, P = 0.007)) compared with those with complete transfusion data (data not shown).
Of the patient study group, 1974/4441 (44%) received RBC transfusion within 12 h of presentation. The characteristics of patients who received an early RBC transfusion and those who did not receive early RBC transfusion are shown in Table 2. Patients receiving early transfusion were a mean of 4 years older and were more likely already to be hospital in-patients. Predictably, they were also more likely to show signs of haemodynamic instability, have a lower initial haemoglobin level and when endoscoped, be found to have peptic ulcers, oesophageal varices or major stigmata of recent bleeding.
|Variable||Early RBC transfusion n = 1974% (n)||No early RBC transfusion n = 2467% (n)||P|
|New admission||80 (1576)||84 (2082)||<0.001|
|In-patient||19 (375)||15 (360)|
|Unknown||1 (23)||1 (25)|
|Mean (s.d.)||67.9 (16.51)||63.4 (19.19)||<0.001|
|Female||39 (762)||39 (948)||0.536|
|Normal||46 (914)||68 (1679)||<0.001|
|Tachycardia only||26 (507)||24 (595)|
|BP < 100 > 70||23 (450)||7 (168)|
|BP < 70 > 50||3 (64)||0 (9)|
|BP < 50||1 (17)||0 (0)|
|Not recorded||1. (22)||1 (16)|
|First recorded haemoglobin concentration g/dL|
|>10.1||15 (292)||74 (1830)||<0.001|
|8.1–10.0||27 (527)||15 (381)|
|7.1–8.0||19 (376)||3 (71)|
|≤7.0||33 (649)||1.2 (41)|
|Not recorded||7 (130)||6 (144)|
|Mean Hb (s.d.)||8.0 (2.16)||12 (2.54)||<0.001|
|IHD||23 (456)||17 (425)||<0.001|
|Cardiac failure||7 (139)||5 (119)||0.002|
|Respiratory disease||11 (217)||11 (263)||0.697|
|Cirrhosis||15 (295)||8 (192)||<0.001|
|Renal disease||11 (208)||6 (159)||<0.001|
|Stroke||6 (117)||8 (189)||0.027|
|Pre endoscopy Rockall score|
|0–1||19 (375)||39 (959)||<0.001|
|2–3||34 (669)||32 (787)|
|4–5||37 (737)||26 (652)|
|6–7||10 (193)||3 (69)|
|Rockall > 3||47 (930/1974)||29 (721/2467)||<0.001|
|Aspirin||35 (657)||29 (693)||<0.001|
|NSAID||14 (261)||12 (293)||0.170|
|Warfarin||10 (199)||6 (138)||<0.001|
|Median time to endoscopy hours (IQR)||19 (8–43)||25 (15–59)||<0.001|
|Peptic ulcer||44 (862)||31 (750)||<0.001|
|Varices||16 (320)||7 (177)||<0.001|
|MSRH||49 (957)||21 (517)||<0.001|
|Complete Rockall score*|
|0–2||14 (279)||36 (873)||<0.001|
|3–5||42 (831)||46 (1132)|
|6–8||38 (747)||17 (418)|
|>8||6 (116)||1 (33)|
|Rockall > 5||44 (863/1973)||18 (451/2456)||<0.001|
|Initial urea (mmol/L)|
|0–6.5||19 (370)||38 (942)||<0.001|
|6.6–8.0||9 (174)||9 (223)|
|8.1–10||11 (208)||11 (273)|
|10.1–25||41 (807)||28 (690)|
|>25||10 (205)||4 (96)|
|Unrecorded||11 (209)||10 (243)|
Re-bleeding occurred in 15% (644/4441). Re-bleeding was more commonly recorded in in-patients (22%) than new admissions (13%, P < 0.001). As shown in Figure 2, re-bleeding was over twice as common in those receiving an early RBC transfusion as in those not transfused at all levels of complete Rockall scores except for those with scores >8. Complete Rockall scores could not be calculated in 12 cases because of incomplete endoscopy data.
Re-bleeding and haemoglobin concentration at presentation
Table 3 shows the relationship between re-bleeding and initial haemoglobin concentration (that first recorded after presentation), early RBC transfusion and the presence of haemodynamic instability (defined as pulse rate 100 beats/min or systolic blood pressure <100 mmHg at presentation) or haemodynamic stability (pulse rate <100 beats/min and systolic blood pressure >100 mmHg at presentation).
|First recorded haemoglobin concentration (g/dL)||All patients n = 4380*||Haemodynamically unstable n = 1786||Haemodynamically stable n = 2560|
|RBC transfusion n = 1950||No RBC transfusion n = 2430||RBC transfusion n = 757||No RBC transfusion n = 1029||RBC transfusion n = 903||No RBC transfusion n = 1657|
|Rate (95% CI)||30% (22–38)||9.4% (5–14)||28% (18–38)||17% (5–28)||33% (19–46)||6.4% (1–11)|
|Rate (95% CI)||28% (20–36)||0%||36% (26–47)||0%||16% (5–26)||–|
|Rate (95% CI)||24% (20–27)||15% (4–27)||31% (26–37)||23% (5–54)||16% (11–20)||12% (3–31)|
|Rate (95% CI)||21% (17–25)||15% (8–26)||28% (21–34)||27% (11–50)||14% (9–20)||10% (3–22)|
|Rate (95% CI)||22% (19–26)||11% (8–15)||28% (22–34)||11% (5–17)||17% (13–22)||12% (8–16)|
|Rate (95% CI)||27% (22–32)||5.8% (5–7)||28% (21–35)||8.7% (6–11)||25% (17–34)||4.5% (3–6)|
At all levels of initial haemoglobin, those receiving an early RBC transfusion had higher rates of re-bleeding than those not having early transfusion. In those not transfused, the re-bleeding rates declined as the initial haemoglobin rose, but in the transfused, the re-bleeding rates varied little according to initial haemoglobin.
When stratified by the presence or not of haemodynamic instability, re-bleeding rates were predictably higher in those with evidence of haemodynamic instability compared with the stable patients. In both, strata re-bleeding rates were higher in the transfused patients irrespective of the initial haemoglobin.
Using a haemoglobin level of <8 gm/dL as the threshold level for transfusion as advocated in guidelines for gastrointestinal bleeding, in patients with an initial haemoglobin of ≤8 g/dL, re-bleeding occurred in 23% (234/1015, 95% CI 21–26%) of early transfused patients compared with a re-bleeding rate of 15% (17/111, 95% CI 8.6–22%) in those not transfused. In patients with an initial haemoglobin >8 g/dL, re-bleeding occurred in 24% (192/812, 95% CI 21–27%) of early transfused patients compared with a re-bleeding rate of 6.7% (147/2196, 95% CI 5.7–7.8%) in those not transfused.
Table 4 shows the unadjusted and adjusted odds ratios for re-bleeding after early RBC transfusion. Using the complete Rockall score to adjust for case mix resulted in a reduction in the overall odds ratio from 4.05 (95% CI 3.36–4.87) to 2.81 (95% CI 2.32–3.42) and then adjusting for initial haemoglobin resulted in a further small reduction to 2.26 (95% CI 1.76–2.90). The increased odds ratios for re-bleeding were greater in the new admissions and in men, but were unaffected by excluding patients with variceal bleeding or by excluding those taking low-dose aspirin on admission.
|Patient group (n*)||Unadjusted re-bleeding odds ratio (95% CI)||Adjusted by Rockall score (95% CI)||Adjusted by Rockall score and haemoglobin concentration (95% CI)|
|In-patients (708)||2.28 (1.57–3.30)||1.70 (1.16–2.52)||1.35 (0.84–2.16)|
|New admissions (3518)||4.68 (3.76–5.83)||3.27 (2.60–4.11)||2.77 (2.07–3.73)|
|Female (1646)||3.45 (2.54–4.69)||2.54 (1.85–3.50)||2.00 (1.34–3.00)|
|Male (2629)||4.42 (3.50–5.58)||2.96 (2.31–3.79)||2.33 (1.69–3.20)|
|Excluding patients with varices (3820)||4.03 (3.29–4.93)||2.89 (2.34–3.57)||2.15 (1.63–2.83)|
|Excluding patients on aspirin (2949)||4.36 (3.50–5.44)||2.86 (2.26–3.61)||2.41 (1.77–3.27)|
|Total (4276)||4.05 (3.36–4.87)||2.81 (2.32–3.42)||2.26 (1.76–2.90)|
Unadjusted mortality for all patients was 7.8% (346/4441). Mortality was significantly higher among in-patients (20%) compared with new admissions (5.4%, P < 0.001). As shown in Figure 2, mortality was greater in patients receiving early RBC transfusion at all levels of the complete Rockall score with the increases being relatively greater with less serious episodes of AUGIB (lower Rockall scores).
Table 5 shows the relationship between mortality and initial haemoglobin concentration, early RBC transfusion and haemodynamic status. As might be expected, mortality was greatest in those with an initial haemoglobin of <5 g/dL (20%) and lowest in those with levels of >10 g/dL (5%), but mortality rates were similar for intermediate levels of initial haemoglobin. In patients with an initial haemoglobin of ≤8 g/dL, mortality was 13% (130/1025, 95% CI 11–15%) in early transfused patients compared with mortality of 13% (14/112, 95% CI 7.0–20%) in those not transfused. In patients with an initial haemoglobin >8 g/dL, mortality was 11% (91/819, 95% CI 9.4–13%) in transfused patients compared with a mortality of 4.3% (94/2208, 95% CI 3.5–5.2%) in those not transfused.
|First recorded haemoglobin concentration (g/dL)||All patients n = 4441*||Haemodynamically unstable n = 1810||Haemodynamically stable n = 2593|
|RBC transfusion n = 1974||No RBC transfusion n = 2467||RBC transfusion n = 1038||No RBC transfusion n = 772||RBC transfusion n = 1679||No RBC transfusion n = 914|
|Mortality (95% CI)||6.9% (3–11)||4.9% (1–8)||8.9% (3–15)||9.5% (1–18)||4.4% (0–10)||3.0% (0–6)|
|Mortality (95% CI)||19% (13–27)||50% (10–91)||26% (18–37)||50% (10–91)||8.7% (3–20)||_|
|Mortality (95% CI)||12% (9–15)||7.7% (0–16)||16% (12–21)||15% (4–42)||7.4% (4–11)||4.0% (0–12)|
|Mortality (95% CI)||12% (9–16)||14% (8–24)||15% (11–21)||18% (7–39)||8.2% (4–12)||12% (3–21)|
|Mortality (95% CI)||11% (8–14)||7.1% (5–10)||15% (11–21)||8.5% (3–14)||6.7% (4–10)||6.1% (3–9)|
|Mortality (95% CI)||12% (9–16)||3.7% (3–5)||16% (11–22)||5.9% (4–9)||5.3% (1–9)||2.6% (2–4)|
Table 6 shows the unadjusted and adjusted odds ratios for mortality after early RBC transfusion. Using the complete Rockall score to adjust for case mix resulted in a reduction in the overall odds ratio from 2.71 (95% CI 2.14–3.42) to 1.50 (95% CI 1.17–1.92) and then adjusting for initial haemoglobin resulted in a further reduction to 1.28 (95% CI 0.94–1.74). The addition of urea level and of the individual components of the Rockall Score to this model either individually or in combination did not alter the estimate of the OR for early transfusion by the pre-specified 10% and so these variables were rejected from the model.
|Patient group (n*)||Unadjusted mortality odds ratio (95% CI)||Adjusted by Rockall score (95% CI)||Adjusted by Rockall score and haemoglobin concentration (95% CI)|
|In-patients (722)||1.70 (1.17–2.46)||1.21 (0.84–1.78)||1.33 (0.83–2.13)|
|New admissions (3596)||3.35 (2.45–4.59)||1.78 (1.27–2.49)||1.40 (0.92–2.13)|
|Female (1714)||2.31 (1.63–3.29)||1.50 (1.03–2.16)||1.29 (0.82–2.03)|
|Male (2727)||3.06 (2.23–4.17)||1.50 (1.07–2.11)||1.31 (0.86–2.02)|
|Excluding patients with varices (3944)||2.70 (2.08–3.50)||1.52 (1.15–2.01)||1.26 (0.89–1.79)|
|Excluding patients on aspirin (3036)||2.82 (2.13–3.73)||1.41 (1.04–1.91)||1.10 (0.75–1.61)|
|Total (4370)||2.71 (2.14–3.42)||1.50 (1.17–1.92)||1.28 (0.94–1.74)|
We have found that patients with AUGIB having a RBC transfusion with 12 h of admission have a two-fold increased rate of further bleeding and an increased mortality, although the latter was not statistically significant in the fully adjusted model. These observations may be due to an adverse effect of blood transfusion upon haemostasis; they could alternatively be due to differences in case mix between early transfused and nontransfused patients that persisted despite corrections using Rockall scores and haemoglobin concentrations.
The Rockall score was developed to adjust for differences in case mix between hospitals in a previous UK audit of AUGIB management.15 Subsequently, it has been widely validated and found to be an excellent predictor of mortality, and a reasonable predictor of the risk of re-bleeding.16–20 It has the advantage of being simple to calculate and is widely used in the UK. In contrast to the Glasgow Blatchford score and some other scores, it does not incorporate initial haemoglobin level.21 We have therefore included initial haemoglobin in our multivariable analysis. As expected, adjusting for Rockall score had a significant impact on the risk estimates for re-bleeding and for mortality as shown in Tables 4 and 6, but addition of initial haemoglobin to the multivariable model had only a modest effect. While we cannot exclude the possibility that our findings can partly be accounted for by residual confounding, this seems unlikely to be sufficient to explain an effect as large as a two-fold increase.
The term ‘re-bleeding’ is a widely used end point in publications relating to AUGIB. In reality, it is impossible to distinguish with total accuracy gastrointestinal bleeding that ceases and restarts and bleeding that slows and then accelerates or is more continuous. Nevertheless, we can think of no reason why any difficulty in detecting or recording of re-bleeding should have introduced bias in our analysis. The doctors extracting data from case records were given clear criteria and definitions for re-bleeding, and while it was clear that one of the objectives of the audit was to examine the use of blood products, no pre-specified hypotheses were mentioned. A further strength of the dataset is that data were collected prospectively from a wide range of UK acute hospitals with over three-quarters submitting their data. Given this wide coverage, there were inevitably some missing data and we cannot be sure that all AUGIB patients were enrolled. Patients not having an endoscopy were also excluded from this analysis. These patients fell into two groups: patients who were critically ill and either had surgery or died before endoscopy and low risk patients who were discharged to have endoscopy as an out-patient. It is difficult to conceive how any of these exclusions would be a source of bias in our analysis. Current guidelines for the management of acute bleeding from varices recommend maintaining a haemoglobin level of approximately 8 g/dL and we therefore used this level in the dichotomized analysis.7 Using this figure, 19% (381/1973) patients receiving an early RBC transfusion had an initial haemoglobin of >8 g/dL and were recorded as being haemodynamically stable. We cannot judge whether these early transfusions were inappropriate or unnecessary, but a recently published international consensus statement suggested that RBC transfusion was mandatory only when the haemoglobin was 7 g/dL or less.22
While our findings might appear surprising, blood transfusion has a number of effects beyond replacement of red cells. Adverse effects of blood transfusion such as an increased risk of post-operative infection, acute respiratory distress syndrome, multi-organ failure and mortality have been found, predominantly in large observational studies of blood transfusion and as a non-statistically significant outcome in a randomized controlled clinical trial in critical care.8, 23–25 These adverse effects require confirmation in randomized controlled trials, but have been suggested to be due to the effect of leucocytes and biological response modifiers released from leucocytes during the storage of red cell concentrates.23 It should be noted that red cell concentrates have been leucocyte-depleted in the UK since 1999, and data were not collected on the storage of red cell concentrates transfused to the patients in this study.
A small trial in AUGIB which compared liberal transfusion with transfusion only if the haemoglobin concentration fell below 8 g/dL found that the percentage of re-bleeding was markedly lower in the restricted transfusion group.5 There was evidence of a hypercoaguable response in AUGIB, which was reversed by RBC transfusion. There are no clear mechanisms to explain the increased risk of re-bleeding associated with blood transfusion found in this previous trial and our study. Loss of one blood volume and replacement with stored red cell concentrates reduce coagulation factors by approximately 70%, but are not generally associated with a bleeding diathesis.26 Coagulopathy in massive blood loss is proportional to the volume of blood loss and is multi-factorial, contributed to by activation of the coagulation and fibrinolytic cascades, haemodilution, hypothermia, impaired hepatic synthesis of coagulation factors, release of tissue factor into circulation, endothelial cell damage activating intravascular coagulation and DIC, acidosis from haemorrhagic shock, hypovolaemia and transfusions, and impaired clearance of activated coagulation factors resulting from shock.27 In addition, the patients in this study with AUGIB associated with chronic liver disease and portal hypertension would be likely to have pre-existing abnormal coagulation and thrombocytopaenia due to poor hepatic synthetic function and hypersplenism respectively. Paradoxically, red cell transfusion has been shown to restore anaemia-induced prolonged bleeding times to normal in normal volunteers, and to reduce blood loss in patients after cardiac surgery, although this effect may depend on storage of red cells for no longer than 14 days.28 Red cell concentrates are currently stored for up to 35 days in the UK, although most are transfused well before 35 days; as mentioned, data on the storage of red cell concentrates were not collected in this study.
This is an observational study and it is still possible that the two fold increase in re-bleeding found after early RBC transfusion reflects residual confounding i.e. the decision to give an early RBC transfusion reflects a clinical assessment of the risk of re-bleeding for which adjustment with the Rockall score and haemoglobin level has been insufficient. While this possibility maybe plausible at lower levels of haemoglobin, at higher levels, it seems most unlikely that the decision to transfuse early, often made by relatively inexperienced doctors, reflects a clinical assessment of risk not captured by the Rockall score and of sufficient magnitude to be associated with a two-fold increase. A definitive answer to this requires a randomized trial. In several other conditions, randomized trials comparing restrictive with liberal transfusion policies have shown some benefits favouring restrictive transfusion.14 In AUGIB, two small trials have supported restrictive transfusion.6, 7 A third larger trial, as yet not fully reported, has found that in patients with AUGIB and portal hypertension a restrictive transfusion policy in which 38% were not transfused was associated with significantly increased survival and fewer side-effects compared with a liberal policy in which only 9% were not transfused.29 Our study strongly suggests that many patients presenting with AUGIB and a haemoglobin of >8 g/dL do not benefit from early RBC transfusion, and we believe a large trial of restrictive and liberal transfusion policies is now required.
Declaration of personal interests: SH was the study lead, designed and conducted the study, and wrote the manuscript. RL was involved in the study design, wrote the manuscript and is a guarantor. KP had the original idea for the study and was involved in study design and reporting. TC provided statistical input to the study. ST was involved in study design and reporting. MM had the original idea for the study, was involved in study design, wrote the manuscript and is a guarantor. Declaration of funding interests: The study lead (SH) was funded by NHS Blood and Transplant and the British Society of Gastroenterology. Representatives from each of these groups (MM and KP) were involved in the design and reporting of the study. Participating hospitals did not receive any financial support.