C O M M E N TA RY Things aren’t always as they seem: what the randomized trials of red blood cell transfusion tell us about adverse outcomes Mark H. Yazer and Darrell J. Triulzi
O
ver the past few years, a link has been made between the receipt of a red blood cell (RBC) transfusion and a variety of adverse outcomes. These connections tend to have been drawn from nonrandomized studies. Recently, several large, randomized controlled trials (RCTs) of RBC transfusion in a variety of different patient populations have been reported and, through a careful reading of their outcomes, offer interesting repudiations to some of these previously drawn connections. Consider a large retrospective study on mortality after intraoperative RBC transfusion.1 This study analyzed nearly 1 million surgery patients over a 5-year period whose data had been recorded in the American College of Surgeons National Surgical Quality Improvement Project (ACS-NSQIP) database. Outcomes on patients who received between 1 and 4 units of RBCs during their surgery were compared to those who did not receive an intraoperative transfusion. Trauma, pediatric, and cardiac surgery patients on bypass were excluded, as were patients who received more than 4 perioperative RBC units. In an unadjusted analysis the authors demonstrated that receipt of 1 to 4 units of RBCs during surgery was associated with a statistically increased risk of developing numerous adverse events including mortality, wound infection, pulmonary and renal complications, and the development of postoperative sepsis in a dose-dependent manner. The authors then went on to perform a propenABBREVIATIONS: GI = gastrointestinal; ICU(s) = intensive care unit(s); RCT(s) = randomized controlled trial(s). From The Institute for Transfusion Medicine and the Department of Pathology, University of Pittsburgh, Pittsburgh, Pennsylvania. Address reprint requests to: Mark H. Yazer, MD, The Institute for Transfusion Medicine, 3636 Boulevard of the Allies, Pittsburgh, PA 15213; e-mail: [email protected]. Received for publication January 30, 2014; revision received March 8, 2014, and accepted March 14, 2014. doi: 10.1111/trf.12706 © 2014 AABB TRANSFUSION 2014;54:3243-3246.
sity matching analysis whereby patients who received a small quantity of intraoperative RBCs were matched to those who did not receive an intraoperative transfusion and once again demonstrated that a variety of adverse events, including death, wound infections, pulmonary and renal complications, and longer lengths of stay occurred significantly more often among the patients who received intraoperative RBC transfusions. From these analyses the authors concluded that the administration of a small volume of RBCs during surgery is possibly a “discretionary” event that should be avoided as their analysis indicated that it only led to harm.1 It would seem difficult to argue with a study that included nearly 1 million patients. However, although all of the transfusions analyzed in this study were administered during surgery, the authors based their propensity matching analysis on preoperative variables. So while the patients who went on to receive an intraoperative transfusion might have been equally healthy (or sick) upon entering the operating room as those that did not require a transfusion, clearly something happened to some of these patients during their surgery that led them to receive a transfusion. It is unfortunate that the ACS-NSQIP database does not include intraoperative variables such as blood loss, vital signs, pressor requirements, and so forth because having a more detailed understanding of why some patients were transfused during their surgery while others were not is at the root of understanding why the transfused patients did worse than the nontransfused patients. If the transfused patients really did not have more blood loss, longer surgeries, lower blood pressures, and so forth than those who were not transfused, perhaps there is a direct link between RBC transfusion and adverse events. However, as the intraoperative variables were not presented, it is entirely possible that confounding by indication explains these results;2 sicker patients receive more transfusions and sicker patients have worse outcomes. Confounding by indication obscures the apparently clear link between the RBC transfusion and the adverse outcome; that is, the reason why the patients needed the transfusion could have caused the adverse outcome, not the transfusion itself. So is a single RBC transfusion harmful to surgical patients? This study cannot answer Volume 54, December 2014 TRANSFUSION
3243
YAZER AND TRIULZI
this question because, although the number of patients in this study was enormous, the intraoperative patient variables that triggered the transfusion were not presented. Furthermore, the notion that confounding by indication explains these results is supported by a retrospective study of cardiac surgery patients that demonstrated that bleeding itself, not RBC transfusion, was correlated to mortality3 (discussed in detail in Waters and Yazer4). Can the RCTs on RBC transfusion help us tease out a cause-and-effect relationship between RBC transfusion and adverse events? To wit there are several interesting and relevant findings from these studies. Consider one of the first large RCTs on RBC transfusion in intensive care unit (ICU) patients. In the TRICC trial, stable, anemic, and euvolemic ICU patients were randomized to be transfused if their hemoglobin (Hb) decreased below 7.0 g/dL in the restrictive arm or if their Hb decreased below 10.0 g/dL in the liberal arm.5 This randomization scheme produced a significant difference in the mean number of RBC units transfused between the two groups (2.6 ± 4.1 units vs. 5.6 ± 5.3 units; p < 0.01). Based on the above-mentioned large retrospective study, if a dose–response relationship between RBC transfusion and the development of adverse outcomes exists, then surely the group that received on average 5.6 units should have fared worse than those who received half as many? Not so. There was no significant difference in the primary outcome (30-day mortality) between these two groups. In fact the only significant differences in outcomes between these two groups were in two secondary outcomes dealing with organ dysfunction scores; the number of failing organs, length of stay in the ICU and hospital, and other mortality outcomes (all secondary outcomes) were not significantly different between these groups. Furthermore, the incidence of infectious complications including bacteremia and septic shock were not different between these two groups. In fact, the only complication that was different between these patient groups was a higher incidence of cardiogenic pulmonary edema in the liberal group, which was entirely predictable given the high incidence of transfusion-associated circulatory overload in ICU patients. The TRICC study featured ICU patients and the large retrospective study featured surgical patients; could this fact alone explain why the highly transfused ICU patients did not fare worse than those in the restrictive group? In other words, are ICU patients who receive transfusions fundamentally different than patients who are transfused during surgery? Perhaps, but if the negative effects of RBC transfusion are so powerful, one would expect to see more adverse events among more highly transfused patients. The FOCUS study analyzed more than 2000 patients undergoing hip fracture repair surgery.6 Similar to the TRICC trial, patients were randomly assigned to either the restrictive group where transfusions were administered if 3244
TRANSFUSION Volume 54, December 2014
the patients manifested signs and symptoms of anemia or if their Hb fell below 8.0 g/dL or the liberal arm where patients received transfusion if their Hb dropped below 10.0 g/dL. Once again, the patients in the liberal arm received significantly more RBC transfusions than those in the restrictive arm (1866 units vs. 652 units in total, respectively; p < 0.001). In fact nearly 60% of the patients in the restrictive arm did not receive even a single RBC transfusion, while more than 95% of patients in the liberal arm received at least 1 RBC unit. As in the TRICC trial, there were no significant differences in any of the primary outcome measures between the highly transfused liberal group and those in the restrictive group. There were also no differences between groups in a variety of secondary outcomes such as discharge location and patient performance in function and symptom scales. Germane to the discussion of morbidity caused by transfusion, the patients in the liberal group did not have a higher incidence of death, wound infection, new onset of purulent sputum, or longer lengths of stay compared to the patients in the restrictive group. Again, it is perhaps important to recognize that the patients in the FOCUS study received their RBC transfusions in the postoperative period while those in the large retrospective study received theirs during their surgery. However, given that almost 60% of the patients in the restrictive group did not receive any RBC transfusions, the fact that adverse events were not increased among those who were highly transfused in the liberal arm speaks against RBCs as a major cause of morbidity and mortality. A recent study of transfusion thresholds in patients experiencing an upper gastrointestinal (GI) bleed suggests that RBC transfusion is not directly responsible for adverse outcomes.7 In this study the 445 patients in the liberal arm received 1638 RBC units while the 444 patients in the restrictive arm received 671 units (p < 0.001), and almost twice as many patients in the liberal arm received at least one RBC transfusion compared to those in the restrictive arm. Overall the patients in the liberal group had a higher mortality than those in the restrictive arm (9% vs. 5%, respectively; p = 0.02); however, the causes of death included uncontrolled bleeding, “complications of treatment,” and death due to “associated diseases”—thus it is not clear if or how transfusion could have led to the increased mortality in the liberal group, and a causal relationship between increased transfusion and mortality remains speculative.7 There was also a significantly higher rate of protocol violations in the restrictive group, thereby potentially masking smaller observed differences between the groups and perhaps a significant difference in outcomes. Unlike in the propensity matching analysis of the large retrospective study, in this prospective, randomized study of patients with a GI bleed there were no significant differences in bacterial infections, acute kidney injury, or pulmonary complications between those who were highly
LESSONS FROM RCTs ON RBC TRANSFUSION
transfused in the liberal group and those in the restrictive group. Although upper GI bleed patients were not likely included in the large retrospective study, the findings of this study once again cast doubt on the suggestion that a transfusion of 1 RBC unit can cause significant morbidity. Two randomized trials have been performed on patients with cardiac disease. In a RCT of 512 adult patients undergoing on-pump cardiac surgery, the patients in the restrictive transfusion threshold group received a median of 0 RBC transfusions while those in the liberal group had a median of two RBC transfusions (p < 0.001), and approximately half of the patients in the restrictive group were transfused compared to more than 75% of those in the liberal group.8 Once again, there was not a significant difference between these two groups in the primary composite outcome measure that included all-cause 30-day mortality, cardiogenic shock, acute respiratory distress syndrome, or acute renal injury requiring dialysis; there were also no differences in any of the secondary outcome measures including cardiac, pulmonary, or infectious complications between the patients in these two groups. Similarly, in a small randomized pilot study of RBC transfusion thresholds in 110 patients experiencing acute coronary events, 73% of the patients in the restrictive group did not receive any transfusions compared to more than 75% of patients in the liberal group who received one or two RBC transfusions.9 Overall the patients in the liberal group received more than three times the number of RBC transfusions as their counterparts. Although there was no statistical difference in outcomes between the groups, the patients in the liberal group demonstrated a trend toward a lower rate of the composite primary outcome of death or MI or unscheduled revascularization. Furthermore, there were also no significant differences in secondary outcomes between groups including pneumonia or blood stream infection, deep venous thrombosis or pulmonary embolism, congestive heart failure, and other secondary outcomes. Although this was a small pilot trial, and the results need to be interpreted in this light, these data also do not support a powerfully negative effect of RBC transfusion. In total, all of these RCTs have succeeded in creating a group of patients who were highly transfused and a comparison group where many of the patients were not transfused at all. If RBC transfusion really does cause adverse events in a dose-dependent manner, the patients in the liberal groups would have been expected to have demonstrated far higher rates of these events than those in the restrictive arms of these studies. Yet the increase in mortality and morbidity seen in the transfused patients in the large retrospective study was not seen in the liberally transfused patients in most of the RCTs. Thus the following points, although closely related semantically, are distinct and might not have been apparent in the absence of well designed RCTs:
1.
2.
Despite the fact that the patients in the restrictive group received fewer RBC units, their outcomes were not worse than those who received more RBC units. Despite receiving a greater number of RBC transfusions, the patients in the liberal groups did not fare worse in terms of morbidity or mortality than those in the restrictive arms.
These comments do not necessarily apply to the patients in the acute coronary syndrome study where liberal transfusions may be beneficial. However, for the other patients, these collective findings highlight the importance of performing RCTs when confounding and bias could be present. It is true that a large RCT cannot be performed for all aspects of medical practice, nor should one be—clinical judgment and personal experience are important factors to consider when treating patients. In some cases, the magnitude of the impact of the intervention is so powerful that its effect simply could not be explained by confounders alone (think of smoking and lung cancer), but if some smokers lived in houses with high levels of radon gas, another important etiologic agent in lung cancer, then an RCT would be required to determine the relative contributions of each agent in the development of lung cancer. Similarly, with respect to medical decision making, it is commonly argued that an RCT is not required to determine if parachutes save the lives of parachutists; while this might seem obvious, the reason why such a study does not need to be performed is because there are no confounding variables—if the chute opens then the parachutist will live and if it doesn’t then he or she will die. In patients with existing conditions and complex medical or surgical problems, multiple confounders exist and the best way to reduce the bias that they can introduce into the study is to use the most powerful tool that we currently have—randomization. Certainly retrospective and observational studies have their place, mainly as hypothesis generators, but making major practice changes based on their findings might be folly if their results are likely to be caused by bias and confounding and not by the intervention.
CONFLICT OF INTEREST The authors have disclosed no conflicts of interest.
REFERENCES 1. Ferraris VA, Davenport DL, Saha SP, et al. Surgical outcomes and transfusion of minimal amounts of blood in the operating room. Arch Surg 2012;147:49-55. 2. Middelburg RA, van de Watering LM, van der Bom JG. Blood transfusions: good or bad? Confounding by indication, an underestimated problem in clinical transfusion research. Transfusion 2010;50:1181-3. Volume 54, December 2014 TRANSFUSION
3245
YAZER AND TRIULZI
3. Dixon B, Santamaria JD, Reid D, et al. The association of blood transfusion with mortality after cardiac surgery: cause or confounding? (CME). Transfusion 2013;53:19-27. 4. Waters JH, Yazer MH. Bleeding causes harm . . . really?! Transfusion 2013;53:2-4. 5. Hebert PC, Wells G, Blajchman MA, et al. A multicenter, randomized, controlled clinical trial of transfusion require-
gies for acute upper gastrointestinal bleeding. N Engl J Med 2013;368:11-21. 8. Hajjar LA, Vincent JL, Galas FR, et al. Transfusion requirements after cardiac surgery: the TRACS randomized controlled trial. JAMA 2010;304:1559-67. 9. Carson JL, Brooks MM, Abbott JD, et al. Liberal versus
ments in critical care. Transfusion Requirements in Critical
restrictive transfusion thresholds for patients with symp-
Care Investigators, Canadian Critical Care Trials Group. N Engl J Med 1999;340:409-17.
tomatic coronary artery disease. Am Heart J 2013;165: 964-71 e1.
6. Carson JL, Terrin ML, Noveck H, et al. Liberal or restrictive transfusion in high-risk patients after hip surgery. N Engl J Med 2011;365:2453-62.
3246
7. Villanueva C, Colomo A, Bosch A, et al. Transfusion strate-
TRANSFUSION Volume 54, December 2014
O
ver the past few years, a link has been made between the receipt of a red blood cell (RBC) transfusion and a variety of adverse outcomes. These connections tend to have been drawn from nonrandomized studies. Recently, several large, randomized controlled trials (RCTs) of RBC transfusion in a variety of different patient populations have been reported and, through a careful reading of their outcomes, offer interesting repudiations to some of these previously drawn connections. Consider a large retrospective study on mortality after intraoperative RBC transfusion.1 This study analyzed nearly 1 million surgery patients over a 5-year period whose data had been recorded in the American College of Surgeons National Surgical Quality Improvement Project (ACS-NSQIP) database. Outcomes on patients who received between 1 and 4 units of RBCs during their surgery were compared to those who did not receive an intraoperative transfusion. Trauma, pediatric, and cardiac surgery patients on bypass were excluded, as were patients who received more than 4 perioperative RBC units. In an unadjusted analysis the authors demonstrated that receipt of 1 to 4 units of RBCs during surgery was associated with a statistically increased risk of developing numerous adverse events including mortality, wound infection, pulmonary and renal complications, and the development of postoperative sepsis in a dose-dependent manner. The authors then went on to perform a propenABBREVIATIONS: GI = gastrointestinal; ICU(s) = intensive care unit(s); RCT(s) = randomized controlled trial(s). From The Institute for Transfusion Medicine and the Department of Pathology, University of Pittsburgh, Pittsburgh, Pennsylvania. Address reprint requests to: Mark H. Yazer, MD, The Institute for Transfusion Medicine, 3636 Boulevard of the Allies, Pittsburgh, PA 15213; e-mail: [email protected]. Received for publication January 30, 2014; revision received March 8, 2014, and accepted March 14, 2014. doi: 10.1111/trf.12706 © 2014 AABB TRANSFUSION 2014;54:3243-3246.
sity matching analysis whereby patients who received a small quantity of intraoperative RBCs were matched to those who did not receive an intraoperative transfusion and once again demonstrated that a variety of adverse events, including death, wound infections, pulmonary and renal complications, and longer lengths of stay occurred significantly more often among the patients who received intraoperative RBC transfusions. From these analyses the authors concluded that the administration of a small volume of RBCs during surgery is possibly a “discretionary” event that should be avoided as their analysis indicated that it only led to harm.1 It would seem difficult to argue with a study that included nearly 1 million patients. However, although all of the transfusions analyzed in this study were administered during surgery, the authors based their propensity matching analysis on preoperative variables. So while the patients who went on to receive an intraoperative transfusion might have been equally healthy (or sick) upon entering the operating room as those that did not require a transfusion, clearly something happened to some of these patients during their surgery that led them to receive a transfusion. It is unfortunate that the ACS-NSQIP database does not include intraoperative variables such as blood loss, vital signs, pressor requirements, and so forth because having a more detailed understanding of why some patients were transfused during their surgery while others were not is at the root of understanding why the transfused patients did worse than the nontransfused patients. If the transfused patients really did not have more blood loss, longer surgeries, lower blood pressures, and so forth than those who were not transfused, perhaps there is a direct link between RBC transfusion and adverse events. However, as the intraoperative variables were not presented, it is entirely possible that confounding by indication explains these results;2 sicker patients receive more transfusions and sicker patients have worse outcomes. Confounding by indication obscures the apparently clear link between the RBC transfusion and the adverse outcome; that is, the reason why the patients needed the transfusion could have caused the adverse outcome, not the transfusion itself. So is a single RBC transfusion harmful to surgical patients? This study cannot answer Volume 54, December 2014 TRANSFUSION
3243
YAZER AND TRIULZI
this question because, although the number of patients in this study was enormous, the intraoperative patient variables that triggered the transfusion were not presented. Furthermore, the notion that confounding by indication explains these results is supported by a retrospective study of cardiac surgery patients that demonstrated that bleeding itself, not RBC transfusion, was correlated to mortality3 (discussed in detail in Waters and Yazer4). Can the RCTs on RBC transfusion help us tease out a cause-and-effect relationship between RBC transfusion and adverse events? To wit there are several interesting and relevant findings from these studies. Consider one of the first large RCTs on RBC transfusion in intensive care unit (ICU) patients. In the TRICC trial, stable, anemic, and euvolemic ICU patients were randomized to be transfused if their hemoglobin (Hb) decreased below 7.0 g/dL in the restrictive arm or if their Hb decreased below 10.0 g/dL in the liberal arm.5 This randomization scheme produced a significant difference in the mean number of RBC units transfused between the two groups (2.6 ± 4.1 units vs. 5.6 ± 5.3 units; p < 0.01). Based on the above-mentioned large retrospective study, if a dose–response relationship between RBC transfusion and the development of adverse outcomes exists, then surely the group that received on average 5.6 units should have fared worse than those who received half as many? Not so. There was no significant difference in the primary outcome (30-day mortality) between these two groups. In fact the only significant differences in outcomes between these two groups were in two secondary outcomes dealing with organ dysfunction scores; the number of failing organs, length of stay in the ICU and hospital, and other mortality outcomes (all secondary outcomes) were not significantly different between these groups. Furthermore, the incidence of infectious complications including bacteremia and septic shock were not different between these two groups. In fact, the only complication that was different between these patient groups was a higher incidence of cardiogenic pulmonary edema in the liberal group, which was entirely predictable given the high incidence of transfusion-associated circulatory overload in ICU patients. The TRICC study featured ICU patients and the large retrospective study featured surgical patients; could this fact alone explain why the highly transfused ICU patients did not fare worse than those in the restrictive group? In other words, are ICU patients who receive transfusions fundamentally different than patients who are transfused during surgery? Perhaps, but if the negative effects of RBC transfusion are so powerful, one would expect to see more adverse events among more highly transfused patients. The FOCUS study analyzed more than 2000 patients undergoing hip fracture repair surgery.6 Similar to the TRICC trial, patients were randomly assigned to either the restrictive group where transfusions were administered if 3244
TRANSFUSION Volume 54, December 2014
the patients manifested signs and symptoms of anemia or if their Hb fell below 8.0 g/dL or the liberal arm where patients received transfusion if their Hb dropped below 10.0 g/dL. Once again, the patients in the liberal arm received significantly more RBC transfusions than those in the restrictive arm (1866 units vs. 652 units in total, respectively; p < 0.001). In fact nearly 60% of the patients in the restrictive arm did not receive even a single RBC transfusion, while more than 95% of patients in the liberal arm received at least 1 RBC unit. As in the TRICC trial, there were no significant differences in any of the primary outcome measures between the highly transfused liberal group and those in the restrictive group. There were also no differences between groups in a variety of secondary outcomes such as discharge location and patient performance in function and symptom scales. Germane to the discussion of morbidity caused by transfusion, the patients in the liberal group did not have a higher incidence of death, wound infection, new onset of purulent sputum, or longer lengths of stay compared to the patients in the restrictive group. Again, it is perhaps important to recognize that the patients in the FOCUS study received their RBC transfusions in the postoperative period while those in the large retrospective study received theirs during their surgery. However, given that almost 60% of the patients in the restrictive group did not receive any RBC transfusions, the fact that adverse events were not increased among those who were highly transfused in the liberal arm speaks against RBCs as a major cause of morbidity and mortality. A recent study of transfusion thresholds in patients experiencing an upper gastrointestinal (GI) bleed suggests that RBC transfusion is not directly responsible for adverse outcomes.7 In this study the 445 patients in the liberal arm received 1638 RBC units while the 444 patients in the restrictive arm received 671 units (p < 0.001), and almost twice as many patients in the liberal arm received at least one RBC transfusion compared to those in the restrictive arm. Overall the patients in the liberal group had a higher mortality than those in the restrictive arm (9% vs. 5%, respectively; p = 0.02); however, the causes of death included uncontrolled bleeding, “complications of treatment,” and death due to “associated diseases”—thus it is not clear if or how transfusion could have led to the increased mortality in the liberal group, and a causal relationship between increased transfusion and mortality remains speculative.7 There was also a significantly higher rate of protocol violations in the restrictive group, thereby potentially masking smaller observed differences between the groups and perhaps a significant difference in outcomes. Unlike in the propensity matching analysis of the large retrospective study, in this prospective, randomized study of patients with a GI bleed there were no significant differences in bacterial infections, acute kidney injury, or pulmonary complications between those who were highly
LESSONS FROM RCTs ON RBC TRANSFUSION
transfused in the liberal group and those in the restrictive group. Although upper GI bleed patients were not likely included in the large retrospective study, the findings of this study once again cast doubt on the suggestion that a transfusion of 1 RBC unit can cause significant morbidity. Two randomized trials have been performed on patients with cardiac disease. In a RCT of 512 adult patients undergoing on-pump cardiac surgery, the patients in the restrictive transfusion threshold group received a median of 0 RBC transfusions while those in the liberal group had a median of two RBC transfusions (p < 0.001), and approximately half of the patients in the restrictive group were transfused compared to more than 75% of those in the liberal group.8 Once again, there was not a significant difference between these two groups in the primary composite outcome measure that included all-cause 30-day mortality, cardiogenic shock, acute respiratory distress syndrome, or acute renal injury requiring dialysis; there were also no differences in any of the secondary outcome measures including cardiac, pulmonary, or infectious complications between the patients in these two groups. Similarly, in a small randomized pilot study of RBC transfusion thresholds in 110 patients experiencing acute coronary events, 73% of the patients in the restrictive group did not receive any transfusions compared to more than 75% of patients in the liberal group who received one or two RBC transfusions.9 Overall the patients in the liberal group received more than three times the number of RBC transfusions as their counterparts. Although there was no statistical difference in outcomes between the groups, the patients in the liberal group demonstrated a trend toward a lower rate of the composite primary outcome of death or MI or unscheduled revascularization. Furthermore, there were also no significant differences in secondary outcomes between groups including pneumonia or blood stream infection, deep venous thrombosis or pulmonary embolism, congestive heart failure, and other secondary outcomes. Although this was a small pilot trial, and the results need to be interpreted in this light, these data also do not support a powerfully negative effect of RBC transfusion. In total, all of these RCTs have succeeded in creating a group of patients who were highly transfused and a comparison group where many of the patients were not transfused at all. If RBC transfusion really does cause adverse events in a dose-dependent manner, the patients in the liberal groups would have been expected to have demonstrated far higher rates of these events than those in the restrictive arms of these studies. Yet the increase in mortality and morbidity seen in the transfused patients in the large retrospective study was not seen in the liberally transfused patients in most of the RCTs. Thus the following points, although closely related semantically, are distinct and might not have been apparent in the absence of well designed RCTs:
1.
2.
Despite the fact that the patients in the restrictive group received fewer RBC units, their outcomes were not worse than those who received more RBC units. Despite receiving a greater number of RBC transfusions, the patients in the liberal groups did not fare worse in terms of morbidity or mortality than those in the restrictive arms.
These comments do not necessarily apply to the patients in the acute coronary syndrome study where liberal transfusions may be beneficial. However, for the other patients, these collective findings highlight the importance of performing RCTs when confounding and bias could be present. It is true that a large RCT cannot be performed for all aspects of medical practice, nor should one be—clinical judgment and personal experience are important factors to consider when treating patients. In some cases, the magnitude of the impact of the intervention is so powerful that its effect simply could not be explained by confounders alone (think of smoking and lung cancer), but if some smokers lived in houses with high levels of radon gas, another important etiologic agent in lung cancer, then an RCT would be required to determine the relative contributions of each agent in the development of lung cancer. Similarly, with respect to medical decision making, it is commonly argued that an RCT is not required to determine if parachutes save the lives of parachutists; while this might seem obvious, the reason why such a study does not need to be performed is because there are no confounding variables—if the chute opens then the parachutist will live and if it doesn’t then he or she will die. In patients with existing conditions and complex medical or surgical problems, multiple confounders exist and the best way to reduce the bias that they can introduce into the study is to use the most powerful tool that we currently have—randomization. Certainly retrospective and observational studies have their place, mainly as hypothesis generators, but making major practice changes based on their findings might be folly if their results are likely to be caused by bias and confounding and not by the intervention.
CONFLICT OF INTEREST The authors have disclosed no conflicts of interest.
REFERENCES 1. Ferraris VA, Davenport DL, Saha SP, et al. Surgical outcomes and transfusion of minimal amounts of blood in the operating room. Arch Surg 2012;147:49-55. 2. Middelburg RA, van de Watering LM, van der Bom JG. Blood transfusions: good or bad? Confounding by indication, an underestimated problem in clinical transfusion research. Transfusion 2010;50:1181-3. Volume 54, December 2014 TRANSFUSION
3245
YAZER AND TRIULZI
3. Dixon B, Santamaria JD, Reid D, et al. The association of blood transfusion with mortality after cardiac surgery: cause or confounding? (CME). Transfusion 2013;53:19-27. 4. Waters JH, Yazer MH. Bleeding causes harm . . . really?! Transfusion 2013;53:2-4. 5. Hebert PC, Wells G, Blajchman MA, et al. A multicenter, randomized, controlled clinical trial of transfusion require-
gies for acute upper gastrointestinal bleeding. N Engl J Med 2013;368:11-21. 8. Hajjar LA, Vincent JL, Galas FR, et al. Transfusion requirements after cardiac surgery: the TRACS randomized controlled trial. JAMA 2010;304:1559-67. 9. Carson JL, Brooks MM, Abbott JD, et al. Liberal versus
ments in critical care. Transfusion Requirements in Critical
restrictive transfusion thresholds for patients with symp-
Care Investigators, Canadian Critical Care Trials Group. N Engl J Med 1999;340:409-17.
tomatic coronary artery disease. Am Heart J 2013;165: 964-71 e1.
6. Carson JL, Terrin ML, Noveck H, et al. Liberal or restrictive transfusion in high-risk patients after hip surgery. N Engl J Med 2011;365:2453-62.
3246
7. Villanueva C, Colomo A, Bosch A, et al. Transfusion strate-
TRANSFUSION Volume 54, December 2014
