Liberal or Restrictive Transfusion after Cardiac Surgery

To the Editor: In their article on red-cell transfusion, Murphy et al. (March 12 issue)1 report substantial nonadherence to treatment in both study groups — one of which had a restrictive threshold for hemoglobin level in red-cell transfusions and one of which had a liberal threshold. Although nonadherence (defined as either the failure to transfuse red cells within 24 hours after a patient’s hemoglobin fell below an assigned threshold or the administration of a transfusion when the hemoglobin level was above the assigned threshold) increases the likelihood of a null result for the primary outcome, it might confound the finding of increased 90-day mortality in the restrictive-threshold group, since patients in that group who received transfusions above the restrictive threshold were assumed to be sicker, whereas patients in the group with a liberal threshold were assumed to be healthier (and hence not receiving transfusions below the liberal threshold). Did the authors observe an increase in mortality in association with nonadherence to treatment in the restrictive-threshold group? Hemant R. Mutneja, M.D.

n engl j med 372;11 nejm.org March 12, 2015 998 T h e ne w e ngl a nd jou r na l o f m e dicine P erioperative anemia is common after cardiac surgery and is associated with significant increases in morbidity and mortality. [1][2][3] The transfusion of allogeneic red cells is the preferred treatment for acute anemia and is also used in patients undergoing cardiac surgery; typically, more than 50% of patients receive a perioperative transfusion, 4,5 which uses a substantial proportion of blood supplies. 6 Observational studies suggest that transfusion is harmful after cardiac surgery; associations have been reported between transfusion and infection, low cardiac output, acute kidney injury, and death. 2,7,8 In contrast, randomized, controlled trials of red-cell transfusion with restrictive thresholds (i.e., transfusions at lower hemoglobin levels) versus more liberal thresholds (transfusions at higher hemoglobin levels) in a range of acute care and surgical settings have shown no significant differences between the two approaches with respect to adverse events or 30-day mortality. 9 These findings, combined with increasing demands on blood services 10 and the costs of storing, handling, and administering red-cell units, 11 have led to an emphasis on restrictive transfusion thresholds in contemporary blood-management guidelines [12][13][14] and in health policy statements. 15,16 Nevertheless, uncertainty about a safe threshold for restrictive red-cell transfusion in cardiac surgery persists and is reflected in the wide range of transfusion rates in cardiac centers in the United Kingdom (25 to 75%) 5 and in the United States (8 to 93%). 4 Uncertainty persists because previous trials comparing liberal and restrictive thresholds in cardiac surgery lacked adequate statistical power, [17][18][19][20][21] and because other trials involved patients who have not undergone cardiac surgery and the results of those trials may not apply to patients with unstable cardiovascular disease. 9,22 To address this uncertainty, we performed the Transfusion Indication Threshold Reduction (TITRe2) trial to test the hypothesis that a restrictive threshold for red-cell transfusion, as compared with a liberal threshold, would reduce postoperative morbidity and health care costs.

Trial Design and Oversight
TITRe2 was a multicenter, parallel-group, randomized, controlled trial conducted at 17 cardi-ac surgery centers in the United Kingdom. Details of the methods have been reported previously. 23 The trial was funded by the National Institute for Health Research (NIHR) Health Technology Assessment program. A National Health Service research ethics committee approved the study, which was conducted in accordance with the principles of the International Conference on Harmonisation-Good Clinical Practice under the oversight of University Hospitals Bristol National Health Service Foundation Trust. The last author vouches for the data and the analyses and for the fidelity of this report to the study protocol (available with the full text of this article at NEJM.org).

Participants
Patients older than 16 years of age who were undergoing nonemergency cardiac surgery were eligible to participate; exclusion criteria 23 are described in Table S1 in the Supplementary Appendix, available at NEJM.org. Participants provided written informed consent before surgery. If the hemoglobin level dropped below 9 g per deciliter (or the hematocrit fell below 27%) at any time after surgery, the participant was randomly assigned to a study group. Thresholds were expressed in terms of hemoglobin level or hematocrit; hereinafter, hemoglobin threshold should be interpreted as a reference to either hemoglobin level or hematocrit.

Randomization
Patients were randomly assigned to either the liberal transfusion-threshold group (threshold hemoglobin level, 9 g per deciliter) or the restrictive transfusion-threshold group (threshold hemoglobin level, 7.5 g per deciliter) by means of a secure Internet-based system that concealed assignments and used cohort minimization to balance assignments according to center and type of surgery. Physicians and nurses were aware of the group assignments. We intended participants to be unaware of the group assignments and tested our success in keeping the study groups blinded by asking the patients if they were aware of the group they were in.

Interventions
Participants in the liberal-threshold group received a transfusion of 1 unit of red cells immediately after randomization. An additional unit was transfused if the patient's hemoglobin level remained below 9 g per deciliter or dropped below 9 g per deciliter again during postoperative hospitalization. In the restrictive-threshold group, 1 unit of red cells was transfused if the hemoglobin level dropped below 7.5 g per deciliter; a further unit was transfused if the level remained below 7.5 g per deciliter or dropped below 7.5 g per deciliter again during postoperative hospitalization. Physicians could contravene the assigned threshold but had to document the reason for the contravention and record the hemoglobin level at the time of the contravention. Similarly, a physician could permanently discontinue adherence to the assigned treatment threshold. This discontinuation did not constitute withdrawal of the participant from the study, and we continued to collect outcome data in accordance with the protocol for all such participants and included them in the analysis population. Other aspects of postoperative care were carried out in accordance with the center's usual practice. Follow-up consisted of contact with the participants by mail or telephone 3 months after randomization to inquire whether a primary outcome event or some other serious adverse event had occurred, to find out about health resources used since discharge, and to ask questions about general health status and participants' awareness of their random assignment.

Outcomes
The primary outcome was a composite of a serious infection (sepsis or wound infection [24][25][26] or an ischemic event (permanent stroke, myocardial infarction, infarction of the gut, or acute kidney injury 27 ) within 3 months after randomization. Definitions and adjudication procedures are described in Table S2 in the Supplementary Appendix. An event was classified as "present" if it was recorded as having occurred, "absent" if it was confirmed that it had not occurred, and "missing" if it was not possible to confirm whether the event had occurred.
Several secondary outcomes were prespecified, including the number of units of red cells and other blood components transfused after randomization; the occurrence of an infection (either sepsis or wound infection, as for the primary outcome, but not including ischemic events); the occurrence of an ischemic event (permanent stroke, myocardial infarction, infarction of the gut, or acute kidney injury, as for the primary outcome, but not including infections); the duration of stay in the intensive care unit (ICU), a high-dependency unit (in which care is less intensive than in an ICU but more intensive than in a hospital ward), or the hospital; and all-cause mortality. The presence of a clinically significant pulmonary complication (defined according to the need for noninvasive ventilation, reintubation, or ventilation or a tracheostomy) was added as a secondary outcome in an amendment to the protocol dated December 2, 2012. All serious adverse events that occurred during follow-up were documented and coded in accordance with the Medical Dictionary for Regulatory Activities, version 14.1; adjudicators of adverse events were unaware of the group assignments.
General health status was assessed at 6 weeks and 3 months after surgery with the use of the EuroQol Group 5-Dimension Self-Report Questionnaire (EQ-5D). 28 The EQ-5D measure of health-related quality of life consists of a descriptive system, which can be converted into a single summary index score (ranging from −0.594 to 1), and a score on a visual-analogue scale (ranging from 0 to 100). For both the index score and the score on a visual-analogue scale, higher scores indicate better quality of life.

Adherence to the Protocol
Nonadherence was defined as either the failure to transfuse red cells within 24 hours after a patient's hemoglobin fell below the assigned threshold or the administration of a transfusion when the hemoglobin level was above the assigned threshold. Multiple instances of nonadherence could occur for one patient. An instance of nonadherence was considered to be severe when it changed the classification of a patient with respect to receipt of any transfusion (i.e., when a patient's hemoglobin level fell below the assigned threshold but the patient did not receive any transfusion or when a patient's hemoglobin level never fell below the assigned threshold but the patient did receive a transfusion).

Cost Analysis
We performed a cost analysis in accordance with guidelines established in the United Kingdom by the National Institute for Health and Care Excellence. 29 Resources used in the hospital and up to 3 months after surgery were documented, valued in 2012-2013 pounds sterling with the use of national sources [30][31][32][33] (wherever possible the valuations were based on actual costs rather than on charges), and converted to U.S. dollars (£1=$1.67).
Resources included blood products and any resources associated with complications (including diagnostic tests), length of hospital stay, and various levels of care up to 3 months after surgery; the costs of the index surgery were not included. Further details of the cost analysis are provided in the Supplementary Appendix.

Statistical Analysis
Basing estimates on previous data 2 and allowing for anticipated nonadherence to the assigned thresholds, 23 we estimated that the frequency of the primary outcome in the group with the restrictive transfusion threshold would be approximately 11% and that the frequency in the group with the liberal threshold would be approximately 17%. We calculated that a sample size of 1468 would be required for the study to have 90% power to detect this difference in a two-sided test, at a 5% level of significance. The target sample size was increased to 2000 to account for uncertainty regarding the rate of nonadherence, since higher-than-expected rates of nonadherence would reduce the power. All the analyses were performed on an inten-tion-to-treat basis according to a prespecified analysis plan. 34 Continuous data are summarized as means and standard deviations or as medians and interquartile ranges if distributions are skewed. All the analyses were based on mixedeffects methods, with adjustment for the type of surgery as a fixed effect and center as a random effect (described as shared-frailty terms in timeto-event models). Binary outcomes were analyzed with the use of logistic regression. Timeto-event outcomes were analyzed with the use of Cox proportional-hazards models; data on duration of stay in the ICU, a high-dependency unit, and the hospital were censored at the time of a patient's death, and data on the time to death were censored at the time of the last follow-up for survivors. EQ-5D scores were analyzed with the use of mixed-effects, mixed-distribution models. 35 We compared the frequency of the primary outcome in prespecified subgroups by estimating the interaction between group assignment and subgroup variable. Sensitivity analyses were performed for the primary outcome (as described in the Supplementary Appendix) and for mortality. A 5% significance level (two-sided) was used in the analysis of the main treatment effects and in subgroup analyses, and a 10% significance level was used in the analysis of interactions between assigned group and time in longitudinal * Plus-minus values are means ±SD. The restrictive transfusion threshold was less than 7.5 g per deciliter, and the liberal transfusion threshold was less than 9 g per deciliter. There were no significant between-group differences at baseline. Details on other preoperative and intraoperative characteristics are provided in Table S3 in the Supplementary Appendix. CABG denotes coronary-artery bypass grafting, and GFR glomerular filtration rate. † Data on body-mass index (the weight in kilograms divided by the square of the height in meters) were missing for one patient in the group with a liberal transfusion threshold. ‡ The European System for Cardiac Operative Risk Evaluation (EuroSCORE) allows the calculation of the risk of death after heart surgery. A score of 0 indicates the lowest level of risk and a score of 44 indicates the highest level of risk; a score of more than 5 is considered to indicate a high risk. Data are missing for 17 patients in the group with a restrictive transfusion threshold and 21 patients in the group with a liberal transfusion threshold. § New York Heart Association (NYHA) classes range from I to IV, with higher classes indicating worse condition. ¶ Canadian Cardiovascular Society (CCS) angina classes range from I to IV, with higher classes indicating greater limitations on physical activity owing to angina. ‖ Data were missing for two persons in this group.

Study Population
Patients were screened for eligibility between July 2009 and February 2013; a total of 3565 consented to take part in the study (Fig. S1 Table 2). The baseline characteristics of patients who consented to participate but did not undergo randomization are shown in Table S4 in the Supplementary Appendix.
At discharge, 15.1% of patients believed that they knew which treatment they had received; 75.6% of those patients (115 patients) were correct (Table S13 in the Supplementary Appendix). At 3 months after surgery, a greater number of patients (27.5%) thought that they knew which treatment they had received, but fewer (56.6%) were correct.

Hemoglobin Levels and Transfusions
After randomization, the mean nadir in the hemoglobin level was approximately 1 g per deciliter lower in the group assigned to the restrictive threshold than in the group assigned to the liberal threshold (Fig. 1); 53.4% of those in the restrictive-threshold group and 92.2% in the liberal-threshold group received one or more transfusions (risk ratio, 0.58; 95% confidence interval [CI], 0.54 to 0.62, P<0.001) ( Table 2, and   Table S5 and Fig. S2A in the Supplementary Appendix). A median of 1 unit of red cells (interquartile range, 0 to 2) was transfused in the restrictive-threshold group, and a median of 2 units (interquartile range, 1 to 3) were transfused in the liberal-threshold group. During the entire index admission, 63.7% of the patients in the restrictive-threshold group and 94.9% of those in the liberal-threshold group received transfusions. The use of other blood products was similar in the two groups ( Table 2, and Table S5 and Fig. S2B in the Supplementary Appendix). The rate of severe nonadherence was 9.7% in the restrictive-threshold group and 6.2% in the liberal-threshold group, and the rates of any nonadherence were 30.0% and 45.2%, respectively ( Table 2).

Outcomes
Outcome data at 3 months after randomization were not obtained for 25 participants (1.2%) (Fig. S1 in the Supplementary Appendix). The numbers of patients with data for each outcome analysis are shown in Table S6 in the Supplementary Appendix; for the primary outcome analysis overall, data were missing for 97 of 2003 participants (4.8%). The primary outcome was observed in 35.1% of the patients in the restrictive-threshold group and 33.0% of the patients in the liberal-threshold group (odds ratio, 1.11; 95% CI, 0.91 to 1.34; P = 0.30) ( Table 3, and Fig.  S3 in the Supplementary Appendix). The majority of primary outcome events in each group occurred before hospital discharge (Table S7 in the Supplementary Appendix). The Supplementary Appendix includes additional details regarding the primary outcome, including the reasons for missing data (Table S6), the distribution of primary-outcome events before and after hospital discharge (Table S7), and a Kaplan-Meier plot of the time from randomization to the primary outcome (Fig. S3).
The duration of patient stay in the ICU or high-dependency unit did not differ significantly between the two groups, and the rates of clinically significant pulmonary complications were also similar (Table 3). There were significantly more deaths in the restrictive-threshold group than in the liberal-threshold group (4.2% vs. 2.6%; hazard ratio, 1.64; 95% CI, 1.00 to 2.67; P = 0.045). days was 2.6% in the restrictive-threshold group and 1.9% in the liberal-threshold group. Kaplan-Meier curves are shown in Figure S4 in the Supplementary Appendix. EQ-5D scores were similar in the two groups ( * Additional details regarding total units of red cells transfused after randomization and other transfusions that were performed are provided in Table S5 and Figure S2 in the Supplementary Appendix. CI denotes confidence interval. † This category includes intraoperative transfusions (which were performed in 184 of 1000 patients [18.4%] in the restrictive-threshold group and in 180 of 1003 patients [17.9%] in the liberal-threshold group) and postoperative transfusions before randomization (which were performed in 98 of 1000 patients [9.8%] in the restrictive-threshold group and in 114 of 1003 patients [11.4%] in the liberal-threshold group). ‡ This category includes transfusions performed during a reoperation after randomization or after treatment had been discontinued. § This estimate is a risk ratio, rather than an odds ratio. The risk ratio reflects the comparison of any transfusion versus no transfusion. This statistic was calculated from an unadjusted logistic-regression model with a log-link function because a model adjusting for cardiac procedure or center would not converge. ¶ This category includes transfusions that were performed before and those that were performed after randomization. ‖ Severe nonadherence changed the classification of a patient as having or not having received a transfusion -that is, a transfusion was not performed in a patient whose hemoglobin level fell below the assigned threshold or a transfusion was performed in a patient whose hemoglobin level was above the assigned threshold. ** Any nonadherence includes instances such as red-cell transfusions performed outside the prescribed 24-hour window or 2 units given consecutively without the hemoglobin level being measured again.

Sensitivity and Subgroup Analyses
When additional acute kidney injury events, identified by means of routinely collected data on creatinine level, were included in the primary outcome, there was a trend toward higher risk in the restrictive-threshold group than in the liberalthreshold group (odds ratio, 1.20; 95% CI, 1.00 to 1.44; P = 0.04) ( Table S11 in the Supplementary Appendix). A similar trend was seen when patients who received a transfusion before randomization were excluded from the primaryoutcome analysis (odds ratio, 1.23; 95% CI, 0.97 to 1.54; P = 0.08). When the primary outcome was restricted to serious events, there was no significant difference between the two groups (odds ratio, 0.99; 95% CI, 0.77 to 1.27; P = 0.94) (Tables S11 and S12 in the Supplementary Appendix). A further sensitivity analysis showed no significant heterogeneity among sites (P = 0.65) and no trend toward a null effect with increases in severe nonadherence (Fig. S5 in the Supplementary Appendix). There was no indication of significant heterogeneity with respect to the primary outcome according to subgroup (Fig. 2).
Other cost components and total mean costs up to 3 months after surgery were similar in the two groups (£10,636 [$17,762] in the restrictive-threshold group and £10,814 [$18,059] in the liberalthreshold group) ( Table S14 in the Supplementary Appendix).

Discussion
In the TITRe2 trial, we tested the hypothesis that the use of a restrictive threshold, as compared with a liberal threshold, for the transfusion of red cells after cardiac surgery in adults would reduce postoperative morbidity and costs. We observed no significant between-group difference with respect to the primary composite outcome. This finding cannot be explained by the possibility that the study did not have adequate power, since the power of the study was greater than that planned because of the higherthan-expected frequency of the outcome. There were also no significant differences in outcome according to hemoglobin threshold in prespecified subgroup analyses, a finding that is inconsistent with the view that the thresholds for hemoglobin in red-cell transfusions should be adjusted according to the patient's level of risk. More patients in the restrictive-threshold group than in the liberal-threshold group died (4.2% vs. 2.6%). There were no significant differences in other secondary outcomes, including total costs, between the two strategies. Our results differ from those of observational analyses of transfusion in patients undergoing cardiac surgery, 36 which have uniformly showed that red-cell transfusion is associated with an increased risk of death and other serious adverse outcomes. The difference is probably due to the fact that observational analyses are confounded by prognostic factors that influence the decision to transfuse red cells. In contrast, our results are consistent with findings of a Cochrane review of randomized, controlled trials involving surgical patients and critically ill patients, 9 in which the clinical outcomes in patients who received transfusions with restrictive thresholds for hemoglobin level were similar to those in patients who received transfusion with liberal thresholds.
A restrictive threshold for transfusion is likely to be favored because it requires the use of fewer units of allogeneic red cells. However, the results of our secondary analyses create new  * This value is an odds ratio. † Since the amount of missing data was greater than 5% (owing primarily to missing data on posthospital discharge), a separate treatment estimate was estimated for infections that occurred before hospital discharge (according to the rules regarding missing data outlined in the statistical analysis plan in the study protocol). For this treatment effect, we estimated an odds ratio of 1.07 (95% CI, 0.85 to 1.36; P = 0.55). ‡ The duration of stay in the intensive care unit (ICU) or high-dependency unit after randomization was 0 days for 63 patients in the restrictive-threshold group and 61 patients in the liberal-threshold group; data were censored for 23 patients in the restrictive-threshold group and 15 patients in the liberal-threshold group. In addition, 37 patients in the restrictive-threshold group and 32 patients in the liberal-threshold group had more than one admission to the ICU or high-dependency unit. § This value is a hazard ratio. ¶ The duration of hospital stay after randomization was 0 days for 4 patients in the restrictive-threshold group and 2 patients in the liberal-threshold group; data were censored for 25 patients in the restrictive-threshold group and 17 patients in the liberal-threshold group. sisted in sensitivity analyses (Table S11 in the  Supplementary Appendix), is a cause for concern. It is not clear in what way anemia that was attributable to the restrictive threshold may have resulted in an increased number of deaths. The difference in hemoglobin level between the groups was modest (1 g per deciliter), and an assessment of causes of death or of severe adverse events that preceded death did not suggest a cause-and-effect relationship, although establishing a cause-and-effect relationship may have been an unrealistic expectation given the small number of deaths that occurred and given a medical setting (cardiac surgery) in which death typically occurs after a sequence of adverse events. A benefit from transfusion with a more liberal hemoglobin threshold was also suggested in two sensitivity analyses of the primary outcome, one in which patients who had received a transfusion before randomization were excluded and one in which additional acute kidney injury events, as determined on the basis of serial data on creatinine levels, were included. These findings seem to support a hypothesis that the use of a more liberal hemoglobin threshold may be beneficial in patients with a hemoglobin level of less than 9 g per deciliter after cardiac surgery. This hypothesis is clinically plausible. The TITRe2 trial differs from previous large trials of transfusion thresholds in that all the participants had cardiovascular disease 38,39 ; in addition, a substantial proportion of participants are likely to have been dependent on oxygen supplementation in the immediate postoperative period. 40,41 Therefore, patients undergoing cardiac surgery are often at the limits of their cardiovascular reserve and may benefit from higher hemoglobin levels. Such patients were excluded The gray vertical lines represent the overall treatment estimate (solid line) and the 95% confidence interval (dashed lines) for the primary outcome as calculated for the entire analysis cohort. The sizes of the circles designating the point estimates reflect the sizes of the subgroups. The restrictive transfusion threshold for hemoglobin was less than 7.5 g per deciliter, and the liberal transfusion threshold was less than 9 g per deciliter. CABG denotes coronary-artery bypass grafting, COPD chronic obstructive pulmonary disease, GFR glomerular filtration rate, and LV left ventricular. from the only contemporary trial we could find that showed restrictive transfusion to be beneficial, a trial that assessed transfusion thresholds in patients with acute upper gastrointestinal bleeding. 42 In contrast, a large trial involving patients with hip fracture, 43 in which 63% of the participants had cardiovascular disease, showed no benefit from restrictive transfusion, and a more recent feasibility trial of transfusion thresholds in patients with unstable coronary disease (myocardial infarction) showed a reduced risk of death among patients who received transfusions at a more liberal hemoglobin threshold. 44 Patients with cardiovascular disease may represent a specific high-risk group for which more liberal transfusion thresholds are to be recommended. This hypothesis should be tested in future pragmatic trials.
The main limitation of our trial was our inability to keep health care staff unaware of the group assignments. However, the use of objective end points and the adjudication of end points by personnel who were unaware of the group assignments protected against detection bias. The nature of nonadherence to protocol differed according to group but affected the overall transfusion rate in only a small percentage of participants. This percentage was similar in the two groups. Another limitation was that prospective data collection failed to identify acute kidney injury events that were apparent on the basis of the routinely collected data on serial creatinine levels. We attribute this discrepancy to differences among centers in the baseline creatinine value used to define acute kidney injury. 45 Finally, we cannot exclude the possibility that other transfusion thresholds, or a wider differential between the two transfusion thresholds, could have altered the results.
In conclusion, the TITRe2 trial compared a restrictive transfusion threshold with a liberal transfusion threshold after cardiac surgery. The restrictive threshold was not superior to the liberal threshold with respect to postoperative morbidity or total costs.