The value of a clinical prediction rule for allogeneic transfusion in cardiac surgery

Postoperative bleeding is a common, almost expected, complication in cardiac surgery. The mean blood loss is approximately 1000 mL1 and, as a result, as many as 40 percent of patients require perioperative RBC transfusions.2 Open heart surgery with cardiopulmonary bypass is associated with a complex coagulopathy, which predisposes patients to significant bleeding both intraoperatively and postoperatively. In response, a great amount of effort and time has been expended to reduce the amount of bleeding and the need for allogeneic transfusions associated with cardiac surgery. Pharmacologic approaches with antifibrinolytics, aprotinin, and DDAVP have been employed intraoperatively to reduce the amount of bleeding. The first two agents are routinely used in cardiac surgery although many surgeons feel that aprotinin offers greater efficacy in patients at higher risk of bleeding. Recent reports, however, have raised concerns of an increase in adverse effects in patients treated with aprotinin.3,4 Erythropoietin, preoperative autologous donation, acute normovolemic hemodilution, and cell salvage have also been employed to reduce the risk of allogeneic blood transfusion in cardiac surgery though these techniques have not been clearly shown to be cost-effective.5,6 Identification of patients at high risk of bleeding could be an important complementary approach to these blood conservation strategies, such that expensive and/or time-consuming blood conservation strategies might be reserved for those patients who will require transfusions. Unfortunately, bleeding in cardiac surgery is not predictable, and the ability to determine which patients will require transfusion is limited. Clinical prediction rules have been published in the medical literature for more than 20 years. They are clinical tools designed to aid in medical decision making and/or improve clinical practice. Clinical prediction rules use clinical information (e.g., demographic data, clinical signs, and laboratory tests) to predict the probability of an outcome or suggest a diagnostic or therapeutic course of action.7 In the latter case, we believe that the accurate prediction of specific outcomes will lead to improved decision making. Prediction rules have been developed for a number of different clinical scenarios including outcomes following myocardial infarction and surgery and diagnosis of ankle fractures and deep vein thrombosis. In some instances, they have had an impact on clinical practice, and in others, they have had little impact. The critical issues are whether health care practitioners embrace the prediction rules and incorporate then into their clinical practice and whether this use of prediction rules leads to a change in clinical outcomes. In this issue of TRANSFUSION, Alghamdi and colleagues8 have developed a clinical prediction rule, the Transfusion Risk Understanding Scoring Tool (TRUST), to determine which cardiac surgery patients will require allogeneic RBC transfusion. With patient data from two prospectively collected databases of cardiac surgery patients at two tertiary-care hospitals in Canada, a clinical prediction rule was developed and internally validated with the first database and then externally validated in the second database. By use of mutivariable logistic regression and stepwise model building, eight variables (hemoglobin, weight, sex, age, nonelective surgery, serum creatinine, repeat cardiac surgery, and complex surgery) were included and a final model for the prediction rule was then selected based on overall predictive accuracy and simplicity. The prediction rule assigns 1 point for each variable that exceeds the cutoff threshold values. A score of 0 predicts a probability of 0 to 20 percent for RBC transfusion. The probability increases 20 percent for each additional point accumulated with scores of 4 to 8 having the highest probability of transfusion (80-100%). Methodologic criteria for the development and evaluation of clinical prediction rules have been published.7,9 These standards propose an ideal method that is often not met in published predictions rules; however, they serve as an excellent guide to assess individual tools. The selection of the predictive variables is a critical step in the development of any prediction rule. The variables chosen for evaluation in the development of TRUST were clear, clinically sensible, and reproducible. They were determined by reviewing previous studies and included only if those variables were easily available before surgery. Appropriate multivariable analysis and model-building techniques were then used to select the variables and build the final model for the prediction rule. Although two variables with significant associations to perioperative blood transfusion were not included in the final model, this limitation does not weaken the prediction rule. Their possible inclusion was tested during the model building and was not found to be beneficial in increasing the predictive ability of TRUST. As a result, the scores derived from this prediction rule should be reliable, because none of the predictive variables is derived from subjective clinical impressions or interpretations, and the prediction rule has been simplified as much as possible. Ideally, the data for the variables used in the development and validation of a prediction rule would be prospectively collected specifically for this purpose and not taken from preexisting databases.7 This process ensures standardization of the data collected in the development and validation stages and mirrors the subsequent use of the prediction rule in clinical practice, which will help ensure its reproducibility in this setting. Given the simple demographic and laboratory data used as predictive variables in the TRUST, however, the use of existing databases is not likely to have affected the reliability of this prediction rule. The critical issues in the success of a prediction rule are the outcomes selected, the predictive capability, and the sensibility of the tool. If a prediction rule fails in these areas, then it will not be adopted by clinicians for use in their daily practice. The TRUST tool succeeds in many aspects of sensibility. As previously described, the predictive variables are clear, clinically appropriate, and reproducible. The calculation of the score is a simple addition of the number of predictive variables present. The subsequent stratification of the probability for transfusion is simple with patients separated into five quintiles of 20 percent. The predictive capacity of the tool may limit its acceptance by clinicians, however. The clinical benefit of a prediction rule in cardiac surgery is the identification of those patients who will not require RBC transfusion and therefore the use of expensive and time-consuming techniques of blood conservation can be avoided. With this prediction rule, however, even patients who have the lowest preoperative TRUST score of 0 are still classified as having up to a 20 percent risk of receiving a RBC transfusion. The negative predictive value of a score of 1 or greater (i.e., the probability of patients with a score of 0 not receiving a RBC transfusion) is only 88 percent. Therefore, patients with a score of 0, who represent only 11 percent of all cardiac surgery cases, still have a 12 percent probability of receiving a transfusion. The use of a cutoff score of 2 or greater increases the number of patients below the threshold to 29 percent, but the negative predictive value decreases to 80 percent (i.e., 20% of patients with a score of 0 or 1 will require a transfusion). Given the practice of offering preoperative autologous donation to patients with transfusion risks greater than 10 percent,10 many physicians and surgeons may not feel comfortable withholding preoperative autologous donation even in those patients determined to be in the lowest risk category for perioperative transfusions. Since preoperative autologous donation has not been demonstrated to be a cost-effective tool to reduce allogeneic transfusions in cardiac surgery,5,6 use of the TRUST prediction rule may be reasonable and could reduce the high wastage rate for autologous RBC units. This step would likely require a shift in the established criteria for autologous blood collection in many institutions, however. The TRUST prediction rule could also be of value in choosing intraoperative pharmacologic therapy to reduce bleeding. This application of TRUST, however, may be limited as the predicted outcome is blood transfusion and not bleeding. Additionally, antifibrinolytics and/or aprotinin are routinely used in many centers for cardiac surgery, and we are still waiting for evidence to demonstrate that aprotinin is superior to other potentially less harmful antifibrinolytic therapy.11 Alghamdi and colleagues have provided us with a well developed and validated clinical prediction rule for an important clinical outcome: perioperative transfusion in cardiac surgery. Although the rationale for the avoidance of RBC transfusions may have changed focus over the past decade,12 there is still value in the ability to predict this important clinical endpoint. The introduction of blood conservation techniques to reduce the need for allogeneic blood transfusion was largely driven by the risk of HIV and hepatitis C. Although these infectious risks have dramatically declined with improved donor screening and testing, new concerns of increased morbidity13 and mortality14 associated with RBC transfusions have arisen in cardiac surgery and other patient groups. The need for a clinical prediction rule to identify those patients at risk of transfusion is particularly important when the techniques to reduce blood transfusions are expensive, time-consuming, or associated with adverse events. The definitive test, however, for any clinical prediction rule is whether it is adopted into clinical practice and whether it makes a difference in clinical outcomes.7 For the TRUST to fulfill these criteria, a change in the philosophy of blood conservation or improvements in our current techniques for blood conservation in cardiac surgery may be required. As a research tool, TRUST will certainly be a valuable asset to identify patients who are at highest risk for blood transfusion and enroll them in future clinical trials evaluating strategies for blood conservation.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.