Allogeneic blood transfusion remains a cornerstone of therapy in the Intensive Care Unit (ICU), with nearly 50% of all ICU patients in the US and abroad receiving packed red blood cells (pRBCs) during their ICU course 12. At the same time, a large body of evidence specific to the acutely ill hospitalized patients indicates that allogeneic blood is associated with an increased risk of infectious complications 345 and increased hospital length of stay 6. Among the critically ill in particular, exposure to blood has been associated with a substantially increased risk of nosocomial infection 789101112, multiorgan failure 13, and death 14.

In addition to infectious complications, there are concerns about transfusion-related immunomodulation and a shrinking blood supply. Because of these issues, efforts have focused on ways to minimize an individual's exposure to pRBCs while in the ICU. One such effort was the TRICC trial 15, which showed that lowering of the transfusion trigger from 10 g/dL (liberal transfusion strategy) to 7 g/dL (restrictive transfusion strategy) in appropriate ICU patients results in comparable, and possibly even somewhat better, outcomes, with a substantial reduction in blood exposure and utilization. Notably there was a significant reduction in certain acute complications in the restrictive when compared to the liberal transfusion strategy group.

Unfortunately, recent data have indicated that utilization of the lower trigger is quite variable 1216, particularly among patients on mechanical ventilation 17. This indicates that we may be exposing the critically ill patients to potentially avoidable allogeneic blood, and may in turn be increasing their risk and overall number of transfusion-attributable severe acute complications (TSACs). Furthermore, since allogeneic blood carries an economic cost, there may be an opportunity to improve quality of ICU care while at the same time saving healthcare dollars via a restrictive transfusion strategy.

We sought to quantify the potential extent to which reducing the pre-transfision (pre-TF) hemoglobin (Hb) from the current practice levels to 7 g/dL, in appropriate patients, may result in the prevention of transfusion-attributable severe acute complications (TSACs). We additionally calculated the potential pRBC units and cost savings that may be realized with this reduction of overall utilization of blood. To address these questions, we constructed a simulation model based on recently published studies.

No human subjects were enrolled in the study, and, thus, the study was exempt from regulations guiding protection of human subjects. This is a secondary analysis based on previously published and peer-reviewed data. All calculations were performed in Microsoft Excel. Multivariate simulations and sensitivity analyses were performed using Crystal Ball^® software (Decisioneering, Inc., Denver, CO).

In our base-case scenario, and based on the inputs and assumptions described in the Methods section, we estimate that there are 1,020,800 adult subjects in ICUs in the US annually who may be at risk for a late transfusion Table 2). Utilizing current transfusion practices, these subjects receive 3,073,360 pRBC units. With the universal institution of a restrictive practice, the TF burden is reduced by 42% to 1,778,234 u pRBCs, resulting in a net savings of 1,295,126 units of pRBCs. Based on our estimated rate of TSACs in this population, this extent of transfusion avoidance prevents 39,246 additional TSACs annually, or approximately 1 TSAC per every 26 patients at risk for a late TF. Total annual blood management costs under current practice conditions are calculated to be $1,948,509,929, all due to the costs of pRBCs. With the restrictive strategy, the estimated cost savings realized would be $821,109,699 annually.

The base case, lower bound and upper bound values for the input variables in the simulations are shown in Table 5. After 10,000 trials for each outcome, the median values (95% confidence intervals) for annualized TSACs avoided, pRBCs saved and attendant cost savings were 39,246 (2,360 to 43,696), 1,295,126 (753,705 to 2,040,700), and $821,109,699 ($348,827,493 to $1,310,093,076), respectively. The worst case scenario was also examined. In the least effective scenario, where TSACs are infrequent (552 units pRBC/1 TSAC) and the lowest percentage of patients is at risk for late transfusions (13%), the number of TSACs avoided would be 1,052 annually. Conversely, in the most effective scenario, where TSACs are frequent (19 units pRBC/1 TSAC) and the highest percentage of patients is at risk for late transfusion (50%), there is a potential to avoid 84,137 TSACs.

While it is self-evident that the estimate of the proportion of ICU patients at risk for late transfusion contributed to 100% of the variability in the estimate of the number of pRBC units saved, it accounted for 59% of the variability in the estimated dollars saved, with the uncertainty around the cost of 1 unit of blood responsible for the remaining 41% of the variability. As for the variability in the estimates of the TSACs avoided, the single major contributor to this, with the 90% share, was the uncertainty in the estimate of the number of pRBC units associated with 1 TSAC; the remaining 10% of this outcome's variability was due to the uncertainty in the proportion of ICU patients at risk for a late pRBC transfusion.

This model-based simulation quantifies the magnitude to which the established literature suggests that TSACs are a significant burden in critically ill patients. Furthermore, our results indicate that efforts to prevent these TSACs through transfusion avoidance can enhance outcomes. Universal adoption of restrictive transfusion strategy represents one potential tool for transfusion avoidance. Our model underscores that a restrictive strategy may be a cost-saving, and thus dominant, option in an anemia management program, in addition to helping reduce the risk for the complications following pRBC transfusion.

Prior efforts to explore the relationship between transfusion and its negative sequelae in the ICU have focused on two areas: nosocomial infection and acute lung injury. Multiple observational analyses, both retrospective and prospective, have demonstrated a relationship between exposure to pRBCs and nosocomial infection. Taylor and colleagues, for example, in a review of their Project Impact database demonstrated an independent correlation between pRBC transfusion and nosocomial infections 8. Addressing ventilator-associated pneumonia (VAP) specifically, Shorr et al. demonstrated a link between transfusion and this process 11, while Earley and co-workers confirmed a connection between transfusion and VAP in a cohort of trauma patients 20. Recently Taylor et al. prospectively substantiated the nexus between transfusion and hospital-acquired infections 22.

Additionally, adverse pulmonary outcomes following transfusion are now better understood. Transfusion-related acute lung injury (TRALI) has emerged as a serious and underappreciated consequence of transfusion. Although estimated to occur in 1 in 5,000 transfusions, the incidence is likely higher in critically ill subjects where the condition may be difficult to distinguish from either the progression of previously underlying lung injury or other forms of pulmonary edema. Suggesting the existence of deleterious pulmonary complications of pRBC transfusions, Gong et al. reported that transfusion independently increased the risk for developing ARDS by 50% (adjusted odds ratio: 1.52, 95% CI: 1.00–2.31) in ICU patients with predisposing factors for ARDS 23. Similarly, Kahn and colleagues in a group of individuals with subarachnoid hemorrhage noted that pRBC transfusion independently more than doubled the probability of ARDS 24. Most recently another cohort study of nearly 5,000 critically ill patients found an adjusted odds ratio of 2.80 (95% confidence interval 1.90–4.12) of developing ARDS in association with pRBC transfusions and a dose-response relationship 25. Finally, Gajic and coworkers have demonstrated an 18% absolute reduction in the development of acute lung injury in association with instituting a protocol limiting high tidal volume ventilation and inappropriate transfusions 26.

In light of these results, several systematic attempts have been made to limit the risks related to transfusion. Some advocate broader use of leukoreduction. Through leukoreduction, fewer pro-inflammatory mediators are infused into the recipient, and in turn, a less intense pro-inflammatory response arises. Despite theoretical benefits, it is unclear if leukoreduction will meaningfully decrease the rate of TSACs. In a large analysis of the Canadian experience with leukoreduction, post-transfusion fever was reduced, but there was no evident impact on nosocomial infection rates 27. A recent randomized trial in acute trauma patients also failed to show any advantages of leukoreduction 28.

Recent efforts to limit ICU patients' exposure to allogeneic blood prompted a number of treatment studies with the rHuEPO. Thus, a 1999 study by Corwin et al. demonstrated a nearly 50% reduction in blood utilization, as well as a trend toward increased transfusion independence in the rHuEPO group 29. A large Phase IIB follow-on trial by the same investigators found a 10% increase in transfusion avoidance and a 19% decrease in overall blood utilization 16. Unfortunately, likely due to sample size, neither trial demonstrated a reduction in adverse events that may be attributable to exposure to allogeneic blood, or a favorable difference in other important outcomes. A large Phase III study was recently completed and should shed some light on the outstanding questions of how rHuEPO may fit into the blood management scheme in the ICU 30.

Reliance on a restrictive transfusion strategy, on the other hand, has appeared more successful at altering outcomes. As noted earlier, Hebert and colleagues showed that lower levels of hemoglobin are well tolerated in most critically ill patients, and that in some subjects fewer transfusions may be associated with improved outcomes 15. Unfortunately, although some larger institutions appear to have successfully adopted a restrictive transfusion strategy 31, it is noteworthy that recent large European and US observational studies suggest that the transfusion trigger in most ICUs remains on average at approximately 8.5 g/dL (1, 2), and is actually even higher in the subgroup of patients on mechanical ventilation 17.

Why is it important to appreciate the extent of the burden of TSACs? An outcome has to be measured before it can be improved. This is most evident with blood stream infection 12 and ventilator-associated pneumonia 112032, which have enjoyed improved outcomes through interventions aimed at their prevention 3334. Our study, by estimating their clinical and economic burden, sets the stage for focusing similar efforts to quantify and prevent some of the avoidable TSACs. An immediate action that may improve these outcomes is to adhere to the restrictive transfusion strategy whenever clinically feasible.

Our project's purpose was to determine to what extent universal adoption of a restrictive transfusion strategy might enhance efforts at transfusion avoidance and with what potential direct cost savings to the system. In that vein, our results show that a restrictive strategy may result in a substantial benefit to the patient and the system through avoidance of TSACs. For instance, we calculated that across-the-board utilization of a restrictive trigger would result in direct blood cost savings of $821 million annually. Given that ICU-acquired complications have substantial humanistic and economic costs 32, this strategy is likely to be even more attractive when one examines it in the context of ICU outcomes as a whole.

One prior report has attempted to detail the burden of blood-related complications in the ICU patients 35. However, their calculations were based on estimates generated from epidemiologic data dealing with all subjects transfused, and not only those who are critically ill. As such, it is not clear that the results from reports dealing with populations of mixed severity of illness are generalizable to critically ill subjects. For example, TRALI arising in an otherwise mildly ill person is likely to have different consequences than TRALI in an already mechanically ventilated patient. Additionally, Shermock and co-workers only focused on several well-defined and rare transfusion-related complications 35. They addressed viral transmission, bacterial contamination, and other similar processes that most clinicians would a priori associate with de minimus risk. Although possessing clear diagnostic criteria, these complications are likely not the major concerns of the intensivist. Hence, we sought to build and to expand on their efforts by exploring recently defined complications of transfusion that have been specifically described in critically ill patients.

In addition to focusing on outcomes that arise in the ICU, our analysis has several other strengths. We relied on data from recent reports describing the results of transfusion in critically ill patients. Moreover, most of the outcome inputs used in our model represent results of randomized, controlled trials 1516. Thus differences between those transfused and not transfused can be assumed to be directly attributable to added exposure to pRBCs. We also were conservative in our actual assumptions. Rather than assuming all subjects in an ICU might be at risk for a late transfusion, for example, we postulated that about 1/3 of all subjects might fit into this category. Furthermore, to formally evaluate the impact of uncertainty surrounding the point estimates for the model inputs, we conducted several sensitivity analyses to test the robustness of our conclusions. Finally, the model's transparency should aid in its applicability to concrete clinical situations.

Nonetheless, our study has several important limitations. First, despite our attempt to acknowledge and adjust for uncertainty, the data from which we draw our model inputs are limited. Unfortunately there are relatively few well-done studies exploring transfusion epidemiology, potential adverse events related to transfusion in the critically ill, and the significance of adoption of restrictive transfusion strategy in the ICU in the real world. Studies quantifying the actual cost of blood in the ICU setting are scarce as well. For this reason our final sensitivity estimates are relatively broad and imply intrinsic sensitivity to the accuracy of the input variables' values. That is, the most significant influence on the number of TSACs avoided in the least effective scenario is the assumption that it takes 552 units to precipitate 1 TSAC. While this number is based on the 95% confidence intervals reported by Hebert 15, it needs to be viewed in the context of observational studies that suggest a much greater incremental risk associated with exposure to each unit of pRBCs 81112. Second, we assumed that the incidence of our primary outcomes of interest, cardiovascular complications and ARDS/ALI, follows a linear dose-response relationship with pRBC exposure. Although some observational studies suggest this to be the case 111222, to the best of our knowledge, no well-designed prospective study has been carried out to confirm this. In reality, both of these adverse outcomes might not be linearly related to pRBC transfusion amounts, but could alternatively follow a threshold effect. Third we did not add the economic or clinical cost of an actual TSAC into our model due to the paucity of such data in peer-reviewed literature. Fourth, the data on liberal transfusion-associated complications are derived from a single trial 15. This may be problematic in several ways: 1). Risk estimates from a trial may not reflect accurately the risks in real practice, and 2). Given a vast list of exclusion criteria and treating physician's discretion with respect to the decision to enroll patients, the results may not be generalizable to all ICU patients. Since a similar study is not likely to be replicated in subpopulations of patients of interest, we have attempted to overcome these short-comings by performing sensitivity analyses utilizing fairly conservative risk estimates and thus biasing the model against restrictive strategy. However, the possibility remains that by applying the strategy across the board, our model may thus overestimate the true effect that the institution of a sensible restrictive practice appropriate to a multitude of clinical situations would have. Finally, the sources for some of the critical inputs were derived from two different trials with potentially different populations. To assess this, we examined the potential differences between the restrictive strategy group in the TRICC trial and the placebo arm in the Corwin study (Table 6) 1516. Although the placebo group in the Corwin trial (n = 652) was slightly larger than the restrictive arm in the TRICC study (n = 418), they were similar in their mean age, gender distribution and APACHE II scores. As for primary diagnoses, there was an imbalance between the groups in the proportion of patients with trauma, 20% vs. 48.5%, and cardiovascular disease, 18% vs. 5.4%, in TRICC and Corwin, respectively. While a deficit of cardiovascular diagnoses may have biased the Corwin placebo arm in the direction of a restrictive strategy, the excess of trauma patients would operate in the opposite direction, potentially biasing against a restrictive strategy. However, these limitation notwithstanding, to the best of our knowledge, ours is the first study to estimate the plausible range of avoidable costs and complications is restrictive transfusion strategy were to be adopted more uniformly in the ICUs across the US.

Our model estimates that universal adoption of a restrictive transfusion strategy could result in avoidance of nearly 40,000 transfusion-attributable severe acute complications annually among the critically ill in the US. This in turn could result in the savings of nearly $1 billion in costs associated with blood alone to the healthcare system. Furthermore, these findings remain robust across a range of sensitivity analyses. Efforts should be undertaken to support more ubiquitous adoption of a clinically appropriate restrictive transfusion strategy in the ICU.

ICU, intensive care unit; pRBC, packed red blood cells; TF, transfusion; TSAC, transfusion-attributable sever acute complication; ARDS, acute respiratory distress syndrome; rHuEPO, recombinant human erythropoietin; VAP, ventilator-associated pneumonia; TRALI, transfusion-related acute lung injury; CI, confidence interval; ALI, acute lung injury

Dr. Shorr has served as a consultant to Ortho Biotech Clinical Affairs, LLC, manufacturer of epoetin alfa. Dr. Zilberberg is a former employee of and has served as a consultant to Ortho Biotech Clinical Affairs, LLC. She is also a stockholder in Johnson & Johnson, its parent company.

MDZ carried out study design, analysis, data interpretation, drafting the manuscript. AFS carried out study design, data interpretation, drafting the manuscript. Both authors read and approved the final manuscript.