IN Shylock’s soliloquy in The Merchant of Venice ,1in his appeal for lack of bias, Shakespeare invoked a physiologic event (the title of this editorial) that would have been known as well to the 17th-century audience as it is to 21st-century anesthesiologists. Although that Elizabethan trespass was surely less severe than is modern spinal reconstruction, total hip arthroplasty, or sternotomy and cardiopulmonary bypass, the qualitative result is the same: hemorrhage. Countless studies have examined hemorrhage, including many evaluating a substantial variety of therapeutic and pharmacologic maneuvers aimed at reducing hemorrhage and the ensuing complications when hemorrhage is not treated adequately. Despite the many pleas for the elimination of bias, as was Shylock’s, those requests have not been heeded fully. His own plea presaged his being thwarted in obtaining a pound of flesh, because he could not perform the surgery without blood loss2any more than it can be accomplished today.
In this issue of Anesthesiology, Zufferey et al. 3examine a modern therapy aimed at decreasing hemorrhage induced by surgical trauma. They provide a meta-analysis examining an unapproved use (frequently called “off-label” or “investigational”) of aprotinin, a protease inhibitor with antifibrinolytic properties, and tranexamic acid, an inhibitor of plasminogen activation and plasmin, in orthopedic surgery. In doing so, they have made an exceptional effort to eliminate bias, and thus provide a better meta-analysis than is usual. They contacted investigators to verify published data and obtain additional information when that available was insufficient; did not consider trials without a specific transfusion protocol; provided separate analyses with and without clinical trials that were not randomized and double-blind; examined and presented “funnel plots”; and performed repetitive analyses to ensure that one trial did not exert a dominating influence on the results. They concluded that both aprotinin and tranexamic acid are efficacious in reducing the proportion of patients receiving transfusion of allogeneic erythrocytes.
However, Zufferey et al. pooled data from all doses, even when more than one was given in a single study—a post hoc procedure generally not accepted, especially by regulatory authorities. Thus, no individual dose was assessed as efficacious, and we are not presented with an understanding of the lowest efficacious dose. Second, there was not a full assessment of the magnitude of the transfusion decrease or blood loss. Statistical significance does not necessarily equate with clinical importance. Both pharmaceuticals reduced by approximately half the fraction of patients given allogeneic erythrocytes. However, although nearly all of the randomized, double-blind trials with tranexamic acid found this statistical reduction, only two of the similar seven trials with aprotinin did so. Other results related to transfusion also differed between the two compounds. For tranexamic acid, none of the randomized, double-blind trials reduced intraoperative blood loss, although most trials reduced total perioperative blood loss (median 31%) and reported a statistically significant decrease in volume of transfused erythrocytes (median 1.5 units). For aprotinin, of the seven double-blind, randomized trials assessed by Zufferey et al. , most demonstrated a reduction in intraoperative blood loss (median 29%) and a reduction in total perioperative blood loss (median 32%). However, the reduced blood loss did not uniformly result in decreased transfused volume of allogeneic erythrocytes. Only three of the seven trials found a statistically significant reduction of the volume of transfused erythrocytes (median 1.3 units). Although the study with the largest number of patients (280) found a statistically significantly reduced blood loss, the magnitude (intraoperative, approximately 100–140 ml; total perioperative, approximately 300 ml) did not reduce erythrocyte transfusion (not surprisingly in view of the relatively modest reduced blood loss).
A meta-analysis can be no better than the data on which it depends, and the analysis itself is subject to biases introduced in the methodology used and the criteria for trial and endpoint selection. Clinical trials regarding transfusion present special difficulties. Prominent among these are the problems in conducting double-blind trials, selection and assessment of appropriate efficacy endpoints, and ensuring that transfusion protocols are identical for all institutions and physicians and are implemented fully and followed without exception. None of the trials that made up the current meta-analysis reported the degree of adherence to their transfusion protocol. The endpoint selected by Zufferey et al. (proportion of patients not transfused with erythrocytes), although endorsed by regulatory authorities,4may not be the most clinically appropriate. If possible, the endpoint should closely reflect the pharmacologic action. Tranexamic acid and aprotinin seem efficacious in decreasing blood loss in at least some types of orthopedic surgery, but apparently insufficiently so to reliably decrease the volume of erythrocytes transfused, perhaps owing to the limited blood loss associated with those procedures.
Presumably, the major goal of reducing transfusion is the reduction of the associated risk, although surgical visibility, direct tissue damage, and conservation of blood components are also considerations. The risk of transfusion is largely a function of the number of units or donors to which the recipient is exposed. Calculations from the estimates presented by Fiebig and Busch5produce an estimated total risk of transmission of any one of the following viruses—hepatitis A, B, C; HIV; human T-cell lymphotropic virus types I and II; parvovirus B19; West Nile virus—as approximately less than 1:100,000 per unit of blood component, but cannot include the emergence of new pathogens. The risk of fatal hemolytic transfusion reaction after erythrocyte transfusion is thought to be approximately 1:1,000,000 units transfused6; the risk of clinically important transfusion-related acute lung injury after erythrocyte transfusion is unclear; and the risk of symptomatic bacterial infection transmitted by transfusion of erythrocytes seems to be 1:1,000,000 units.5The persistence of donor leukocytes in recipient blood for more than 1 yr after transfusion,7despite leukoreduction of donated blood,8is worrisome and possibly reflective of the uncertain immunomodulatory risks of transfusion.9
It would seem that an endpoint that would better quantify the decrease of risk of transfusion is one that encompasses the risk from all blood components: a comparison of the number of donors to which the recipient is exposed, rather than only avoidance of transfusion of just erythrocytes. The reports of the trials examined by Zufferey et al. regarding the use of aprotinin (approved only for use in cardiac surgery with cardiopulmonary bypass) or tranexamic acid (approved only for use in hemophilia) in orthopedic surgery did not contain sufficient data to enable analysis of this clinically relevant endpoint. Therefore, even among those trials that found a decrease of erythrocyte transfusion, a potential reduction of total allogeneic transfusion–related risk could not be quantified.
Shylock had something to say about safety, as well. Only two phrases after discussing bleeding, he further observed, “… if you poison us, do we not die?”1Modern therapeutics rarely cause lethality, owing in part to careful evaluations by pharmaceutical companies and regulatory authorities, but nearly all have undesirable effects, some of them potentially serious. Determination of safety of pharmaceuticals can be more difficult than demonstration of efficacy, owing to the generally low incidence of important adverse effects, requiring large sample sizes for their elucidation. Zufferey et al. concluded appropriately that the trials they examined had insufficient information to document the safety of use of antifibrinolytics in orthopedic surgery.3Adding to this uncertainty are three retrospective analyses of cardiac surgery databases10–12that have highlighted a question related to aprotinin that had been raised previously13: that of renal damage. This controversial issue has been addressed recently by several authors.14–16No matter the level of sophistication of statistical approaches, applications, and data examination in an effort to reduce bias, elimination of bias is exceedingly unlikely in retrospective evaluations of databases, including when propensity analysis is used. Indeed, the central purpose of a prospective, randomized, blinded clinical trial is to eliminate such bias.
The implications of the analysis of Zufferey et al. extend beyond the immediate aims of the authors. Their report should stimulate anesthesiologists to maintain an interest and understanding of the increasingly complex fields of coagulation and natural and exogenous inhibitors of clotting. We have learned much in the past decade or so, in part owing to the development of recombinant coagulation factors, and the research associated not only with their development, but that permitted by their availability as pure zymogens or enzymes, rather than as a mix of many components. Current knowledge regarding coagulation has been well reviewed in this journal.17As we have learned about the many steps involved and their precise biochemical nature, so has been developed a variety of pharmaceuticals targeted at specific individual steps of the many reactions. A few are in the clinical domain, but many more are likely to be available clinically in the next few years. Their use, either singly or in combination, for the many patients with risk for thromboembolic disease, will pose substantial intraoperative and postoperative challenges.
We have also learned much about the innate systems that limit unneeded coagulation or eliminate clots after their formation. There are numerous native molecules that stimulate or inhibit fibrinolysis. The many interactions between coagulation and fibrinolysis have been reviewed recently.18When coagulation factor or platelet concentrations are reduced (as occurs routinely during surgery: hemodilution—significant blood loss, and appropriate consumption without clotting factor or platelet replacement) or are absent (as in hemophilia), a loosely formed network of fibrin that is poorly resistant to lysis by plasmin is produced.19This has led some to label this condition as “hyperfibrinolysis;” however, it seems likely that it represents in vivo normal fibrinolytic activity that has been initiated consequent to thrombin production and fibrin formation, acting on a poorly structured fibrin network, as has been shown in vitro . Clinically, it is exceedingly difficult to distinguish between poor fibrin formation or excessive plasmin activity as a potential cause of inappropriate or excessive bleeding. It is possible that the efficacy of aprotinin may be related to fibrinolysis during states of poor clot formation when coagulation factors (and perhaps platelets) have been decreased by loss and dilution.
Four hundred years after the writing of The Merchant of Venice , it is still debated whether Shylock was intended to be portrayed as a villain or as a sympathetic victim of bias. Without appropriate information, we cannot ascertain Shakespeare’s intent or the benefit:risk of the use of aprotinin or tranexamic acid in orthopedic surgery. Existing data from previously conducted prospective randomized clinical trials may be adequate to answer questions regarding the safety of aprotinin in cardiac surgery but do not seem adequate to allow any conclusion regarding its safety in orthopedic surgery. In view of the limited efficacy in orthopedic surgery, clinicians would be well advised to consider such off-label use exceedingly carefully until we have appropriate information from well-designed, well-executed, prospective, randomized, double-blind trials to provide the best, least biased results to guide therapy.
Novo Nordisk A/S, Bagsvaerd, Denmark, and Professor Emeritus, University of California, San Francisco, California. rwes@novonordisk.com