This article arises from a report that I recently read in that well-known medical journal, The Daily Telegraph. It was a small paragraph tucked in a corner of page 2, describing a study published in Circulation which apparently demonstrated that blood transfusion had multiple adverse effects in surgical patients, including increased infection rates and prolonged lengths of stay. As an anaesthetist who plods through an elective orthopaedic list every week, acute surgical blood loss and its replacement both intra- and post-operatively occupies my mind a great deal, so I decided to read the original paper and assess its findings against the background of the general literature. As usual, I entered a complex minefield, and, as you will see, was unable to reach any firm conclusions based on the current evidence. Even worse, the evidence seems to be open to interpretation according to personal prejudices, with the same evidence being given an entirely different interpretation by different review writers. My most depressing discovery is that, in spite of the very long history of blood transfusion as a life-saving therapy, we really have no idea whether transfusion is good or bad for patients. The sensible unbiased science has not been done.
The research paper described in the Telegraph did not look at a general surgical population, but at patients undergoing cardiac surgery, a very discreet and unusual group1. The design was a retrospective cohort study, which means in practice that the records of patients over a defined period of time are recalled from hospital databases, and compared according to specific criteria.
|Figure 1. Kaplan-Meier “survival” curve showing the cumulative proportion of patients discharged from ICU/HDU and from hospital over time according to whether patients had had an RBC transfusion. Patients who died in hospital were censored at death. Figure 1 is taken from reference 1, from American Heart Association journal, Circulation|
In this case, the cohorts were patients undergoing routine cardiac surgery, and they were separated into two groups depending on whether or not they received a blood transfusion in the postoperative period (intra-operative transfusion was excluded because of the potential confounding effects of cardiac bypass). The records of more than 8500 patients, who underwent cardiac surgery over an 8-year period, were examined-an impressive figure to be analysed in a single centre. The findings are impressive too, and in rather an alarming way. The two primary outcome measures, infection and ischaemia, were substantially more common in patients who received a blood transfusion, irrespective of lowest recorded haematocrit. In other words, if two post-cardiac surgery patients had a similar haematocrit, and one was transfused while the other was not, the transfused patient was more likely to experience severe infection or an ischaemic event (cardiac, neurological or renal) during their hospital stay. Even worse, transfused patients had an increased length of stay (figure 1) and a substantially increased post-discharge mortality, which rose with each passing year (figure 2).
|Figure 2. Kaplan-Meier survival curve showing the cumulative proportion of patients who died over time according to whether patients had had an RBC transfusion. Vertical dotted lines separate the epochs of follow-up time for which hazard ratios were estimated (ie, 0 to 30 days, 31 days to 1 year, and after 1 year). Figure 2 is taken from reference 1, from American Heart Association journal, Circulation|
At this point you are probably standing in the theatre sluice pouring that unit of blood you were about to give down the drain. Before you commit that ridiculously expensive crime, we should examine the paper more closely, and in the context of other studies. Because this is a retrospective study based entirely on collection of figures from hospital data bases, we know a lot less about the patients than you might think from the impressive array of data presented in the paper. Pre-operative co-morbidities are relatively easy to collect and are well-documented in this study. The bottom line appears to be that pre-operative medical status does not have an impact on transfusion-related adverse outcomes. For example, if two cardiac surgery patients had diabetes, the transfused patient was at greater risk of an adverse outcome than his non-transfused cohort colleague. The authors interpret this to mean that physiological condition does not influence outcome, i.e. the transfused patients were not sicker in the first place. What the study does not tell us is what happened to the patient during their post-operative stay, which is a quite separate issue from their pre-operative status. There is an implicit presumption in this study that a falling haematocrit is the only transfusion trigger. This may be reasonable practice but is not common practice, and human nature is likely to confound the results here. It might be for example that patients with post-operative ischaemia were given a blood transfusion whether their haematocrit appeared to justify it or not, because they were hypotensive, had poor urine output or simply had a degree of ST segment depression that worried the staff and made transfusion more likely. Early signs of systemic infection might also prompt the on-call medical staff to administer a blood transfusion. I am not condoning blood transfusion as part of the management of this clinical scenario. What I am saying is that there is no information given in this paper as to when and why blood transfusions were given. Blood may have been given as part of the management of post-operative complications such as ischaemia and infection, and these types of complication do indeed prolong hospital stay and are a source of increased long-term mortality. A prospective cohort study, using data collection specifically geared to analyse the patients' post-operative course, is the only way to properly answer this question. In my opinion, the study in its present form is inherently flawed in its conclusions.
This brings us to the question of study design in general when looking at the effects of blood transfusion. For any clinical topic under investigation, the most robust form of test is generally held to be the randomised controlled trial (RCT). A study of this nature is inherently prospective of course, and involves dividing patients into untreated (control) and treated groups. Unfortunately it would probably be downright unethical to deliberately assign patients to either receive a blood transfusion or not, without regard to the clinical circumstances. The inevitable result of this difficulty is that modified study designs, with major flaws, have been used to investigate the potentially harmful effects of blood transfusion. The most common design is similar to the one described above, where two groups of patients, one of which received blood and the other did not, are analysed in either a prospective or, more commonly, a retrospective study. Such a study is effectively observational, even though it may be billed as an RCT. The statistical semantics are very important here. A proper randomised controlled trial has two groups which are comparable in every other way except for the intervention. This is not the case for studies of blood transfusion. A very careful analysis by Vamvakas2 of all the then available RCTs and meta-analyses on the harmful consequences of blood transfusion clearly demonstrated huge variability in patient characteristics between RCTs included in the same meta-analysis. More damaging, the patients were not matched for illness severity, and more gravely ill patients predominated in the transfused groups. In other words, normal rigorous statistical rules were not applied. The result was that transfused patients appear in all of these studies to have a greater infection risk, and a greater incidence of cancer several years later. Vamvakas' reanalysis of the data showed no clear difference between transfused and non-transfused patients with regard to infection or cancer rates (figures 3 and 4).