Health Care–Associated Infection After Red Blood Cell Transfusion: A Systematic Review and Meta-analysis

Importance  The association between red blood cell (RBC) transfusion strategies and health care–associated infection is not fully understood.

Objective  To evaluate whether RBC transfusion thresholds are associated with the risk of infection and whether risk is independent of leukocyte reduction.

Data Sources  MEDLINE, EMBASE, Web of Science Core Collection, Cochrane Central Register of Controlled Trials, Cochrane Database of Sytematic Reviews, ClinicalTrials.gov, International Clinical Trials Registry, and the International Standard Randomized Controlled Trial Number register were searched through January 22, 2014.

Study Selection  Randomized clinical trials with restrictive vs liberal RBC transfusion strategies.

Data Extraction and Synthesis  Twenty randomized trials with 8598 patients met eligibility criteria, of which 17 trials (n = 7456 patients) contained sufficient information for meta-analyses. DerSimonian and Laird random-effects models were used to report pooled risk ratios. Absolute risks of infection were calculated using the profile likelihood random-effects method.

Main Outcomes and Measures  Incidence of health care–associated infection such as pneumonia, mediastinitis, wound infection, and sepsis.

Results  The pooled risk of all serious infections was 10.6% (95% CI, 5.6%-15.9%) in the restrictive group and 12.7% (95% CI, 7.0%-18.7%) in the liberal group. The risk ratio (RR) for the association between transfusion strategies and infection (serious infections and selected infections, combined) was 0.92 (95% CI, 0.82-1.04) with little heterogeneity (I² = 6.3%; τ² = .0041). The RR for the association between transfusion strategies and serious infection was 0.84 (95% CI, 0.73-0.96; I² = 0%, τ² <.0001). The number needed to treat (NNT) with restrictive strategies to prevent serious infection was 48 (95% CI, 36-71). The risk of infection remained reduced with a restrictive strategy, even with leukocyte reduction (RR, 0.83 [95% CI, 0.69-0.99]). For trials with a restrictive hemoglobin threshold of <7.0 g/dL, the RR was 0.86 (95% CI, 0.72-1.02). With stratification by patient type, the RR for serious infection was 0.72 (95% CI, 0.53-0.97) in patients undergoing orthopedic surgery and 0.51 (95% CI, 0.28-0.95) in patients presenting with sepsis. There were no significant differences in the incidence of infection by RBC threshold for patients with cardiac disease, the critically ill, those with acute upper gastrointestinal bleeding, or for infants with low birth weight.

Conclusions and Relevance  Among hospitalized patients, a restrictive RBC transfusion strategy compared with a liberal transfusion strategy was not associated with a reduced risk of health care–associated infection overall, although it was associated with a reduced risk of serious infection. Implementing restrictive strategies may have the potential to lower the incidence of serious health care–associated infection.

Efforts to prevent health care–associated infection are among the priorities for the US Department of Health and Human Services¹ with the focus on common activities in the hospital setting, such as using bundles (defined as groups of effective practices) and checklists to prevent central line–associated bloodstream infection,² encouraging hand hygiene,³ and avoiding the use of urinary catheters.⁴ The estimated annual direct medical costs of health care–associated infections to US hospitals ranges from $28 billion to $45 billion, with about 1 in every 20 inpatients developing an infection related to their hospital care.⁵

Transfusion of red blood cells (RBCs) is a common inpatient therapy, with approximately 14 million units transfused in 2011 in the United States, 84.8% of which were leukocyte reduced.⁶ Although direct transmission of infectious agents via allogeneic RBC transfusion is quite low in developed countries, transfusion is associated with immunomodulation, which may affect infection risk.^7-9 Although elevation of inflammatory markers seen in transfusion-related immunomodulation may be ameliorated through leukocyte reduction, other approaches to blood management include lowering the hemoglobin thresholds at which RBC transfusions are indicated, so-called restrictive threshold strategies. Patient blood management (PBM) has been described as an evidence-based, multidisciplinary approach to optimizing the care of patients who might need a transfusion.⁶ Such PBM interventions have been tested in randomized trials, including several recent trials.^10,11 We conducted a systematic review and meta-analysis of the randomized trials that incorporated 2 comparator groups—restrictive vs liberal RBC transfusion strategies—to evaluate their association with the incidence of health care–associated infection. In addition, we studied whether this association persisted when restricted to patients who received leukocyte-reduced RBC units.

We followed methods defined in the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines statement.¹² Eligibility criteria included the following requirements: (1) randomized trial, (2) use of 2 comparator groups in which 1 group received a restrictive RBC transfusion strategy and the other group received a liberal RBC transfusion strategy, and (3) infectious outcomes after randomization were reported. There were no restrictions regarding patient age or conditions.

The search sources included MEDLINE, EMBASE, Web of Science Core Collection, Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, ClinicalTrials.gov, International Clinical Trials Registry, and the International Standard Randomized Controlled Trial Number register. The search strategy began with specifications as outlined in the Cochrane review of transfusion thresholds for guiding allogeneic RBC transfusions¹³; specific search terms are listed (eTable 1 in the Supplement). No restrictions were placed on language or type of publication. Boolean search terms were used to capture the comparators relevant to transfusion thresholds, strategies, triggers, protocols, practices, policies, criteria, or standards. The date of the last search was January 22, 2014. A medical librarian with experience in searching for systematic reviews assisted.

Study selection was conducted through independent review. Two teams of independent reviewers (J. R. and D. D.; L. K. and M. A. M. R.) examined abstracts for eligibility based on study design (randomized trial) and comparators (RBC transfusion strategies). To divide the work effort, the first team reviewed entries prior to February 6, 2013, and the second team reviewed entries from February 6, 2013, to January 22, 2014. For publications with team member disagreement, the full-text articles were obtained and eligibility was determined through consensus. Trials that did not contain infectious outcomes were excluded, as well as studies with duplicate samples, ineligible comparators, and studies that described the methods of the trials but did not report results.

A data extraction form was developed prior to manuscript review. It contained descriptive information regarding the study, as well as data regarding enrollment, randomization, hemoglobin thresholds, RBC transfusion use, quality (concealment of randomization, blinding, loss to follow-up, and protocol violations), funding, and infectious outcomes. Concealment of randomization was defined as (1) a direct statement of concealment, (2) use of centralized randomization at a coordinating center in trials with multiple sites, or (3) randomization by an independent statistician. Extraction of data was conducted independently by 2 investigators, with subsequent discussion and resolution of discrepancies by consensus.

For each study, risk ratios were calculated comparing the cumulative risk of infection in the restrictive RBC transfusion group (numerator) to the cumulative risk of infection in the liberal RBC transfusion group (denominator). The DerSimonian and Laird random-effects model was used for pooling the risk ratios across studies. For 1 study,¹⁴ a hazard ratio was reported and it was used to approximate the risk ratio of infection; a sensitivity analysis was completed with and without the inclusion of this study. Absolute risks of infection for the restrictive group and for the liberal group were pooled using the profile likelihood random-effects method with variance stabilized using the enhanced Freeman-Tukey arcsine transformation.^15,16 The profile likelihood method has demonstrated better performance than the DerSimonian and Laird method when there is heterogeneity.¹⁶ Heterogeneity was assessed using 3 measures—between-study variance (τ²), Cochran Q test, and the inconsistency index (I²). Number needed to treat (NNT) was calculated using the pooled effects and the median risk of infection in the liberal RBC transfusion group. The NNT reflected the number of patients required to undergo the restrictive transfusion strategy in order to prevent an additional patient from developing an infection compared with the baseline risk (liberal RBC transfusion strategy). Publication bias was evaluated through visual inspection of the funnel plot and through the Harbord test of small-study effects and the Peters test of funnel asymmetry. Sensitivity analyses were performed relevant to study quality assessments (concealed randomization, blinding, withdrawals, and protocol violations). Stata/MP (StataCorp), version 13.0, was used for the analyses. The 2-tailed α level was .05.

The pooled risk ratio for all studies was calculated. In addition, results were stratified by (1) trials that reported all serious infections combined and (2) trials that reported specific types of infections. This was done because the underlying hypothesis was that allogeneic RBC transfusion exerts systemwide effects on immunity.^17-19 When only 1 type of infection is reported (eg, wound infections), individuals with infections at other sites, such as pneumonia or bloodstream infection, contaminate the comparison group representing those patients without (wound) infection. If there is a true systemwide effect, this reporting bias tends to drive the risk ratio toward the null.

Meta-analyses were stratified by clinical setting (eg, cardiac surgery and critical care) because variation based on underlying conditions was a possibility.¹³ In addition, we evaluated whether leukocyte reduction of the RBC units could account for risk differences; results were stratified by the use of leukocyte-reduced RBC units vs unknown or partial use of leukocyte-reduced RBC units.

We evaluated whether the risk of infection varied by the degree of hemoglobin threshold restriction (ie, <7 g/dL in the restrictive transfusion group). Random-effects meta-regression was conducted to investigate the association between the difference in hemoglobin thresholds (threshold in liberal group − threshold in restrictive group) and the risk ratios (dependent variable), with the hypothesis that a wider difference in hemoglobin thresholds between the 2 groups would be expected to yield risk ratios farther from the null. Residual maximum likelihood was used to estimate the between-study component of variance.

There were 2267 publications retrieved from the search and 2189 were excluded because they were not randomized trials, had inappropriate comparators, or both (Figure 1). Of the 78 potential studies, 58 were excluded because they did not report infectious outcomes, were duplicate samples, contained ineligible comparators, were papers describing methods only, or were not trials. There were 20 trials with 8598 patients that met eligibility criteria and were included in this systematic review (Table 1).^{10,11,14,20-37} Sufficient information was available to pool data from 17 randomized trials for meta-analyses.^{10,11,14,20-33,35}Table 2 lists the infection outcomes that were reported in each study.

Five trials were conducted within the United States, 3 in the Netherlands, 2 in Canada, 2 in Denmark, 1 in the United Kingdom, 1 in Brazil, 1 in Spain, and 5 were conducted in facilities within multiple countries (Canada, United States, Belgium, United Kingdom, Australia, and Ireland). Patient enrollment spanned from 1994 in the Transfusion Requirements in Critical Care (TRICC) trial to 2012 in the Carson and coauthors¹⁰ trial of symptomatic coronary artery disease and the de Gast-Bakker and coauthors²⁵ trial of pediatric cardiac patients. In trials with adult participants, the hemoglobin threshold ranged from 6.4 g/dL to 9.7 g/dL in the restrictive group, and the hemoglobin threshold in the liberal group ranged from 9.0 g/dL to 11.3 g/dL. Baseline hemoglobin levels were quite comparable (eTable 2 in the Supplement). There were considerable percentages of patients receiving transfusions after randomization, even in the restrictive groups (cardiac, 27%-60%; critically ill, 46%-67%; gastrointestinal, 49%; low birth weight, 89%; orthopedics, 34%-45%; sepsis, 56%). Patients in the restrictive group received fewer RBC units than in the liberal group with the exception of 2 trials.^14,32 In 1 trial, the study was reported in abstract form and this information was not stated.¹⁴ In the other trial, RBC transfusion use was greater in the restrictive than liberal group in 1 of the 3 participating hospitals.³² The authors published a subsequent paper in which they analyzed the data by RBC transfusion use rather than by random assignment.³⁸ For purposes of our analyses, we used the trial data as randomly assigned, which is conservative in that 2 fewer infections were reported in the restrictive group in the subsequent report.³⁸

Concealed randomization occurred in 14 of the 20 trials (eTable 3 in the Supplement). Some of the trials included blinding of treatment assignments to patients,^21,30,31 surgical staff,²² investigators who performed the outcome assessments,^10,21,29-33 statisticians,^27,35 or independent data and safety monitors.^22,27,35 The percentage of patients who withdrew or were excluded after randomization was low in most trials and ranged from 0% to 17%. The frequency of protocol violations or nonadherence to assigned treatment varied across studies. In a trial of patients undergoing cardiac surgery, nonadherence to protocol occurred in most of the patients in the liberal group (59%) and in 16% of patients in the restrictive group.²⁴ In a randomized trial of patients presenting with acute myocardial infarction, the protocol allowed suspension by the treating physician but, once resolved, transfusion was then given according to the a priori strategies; the number of suspensions was not listed.²³

Of the 20 randomized trials, results from 1 study could not be pooled in the meta-analysis because the unit of observation was operations, not individuals.³⁶ In this study, the rate of fever or infection was 1.7% (95% CI, 0.5%-3.8%) in the restrictive group and it was 2.3% (95% CI, 0.9%-4.7%) in the liberal group. In 2 other trials, the comparators did not include specific hemoglobin thresholds in the protocol but, rather, restriction of transfusion during a certain period (before surgery in 1 trial and after childbirth in another trial).^34,37 In both studies, there was no significant difference in the risk of infection between the randomized groups.

There was one pediatric trial²⁷ in which there was a separate analysis for patients with sepsis.³⁵ In patients with sepsis, there was a significant reduction in the risk of nosocomial infection in those randomized to the restrictive transfusion strategy compared with the liberal strategy, with a risk ratio of 0.51 (95% CI, 0.28-0.95; P = .033).

In the 17 trials with 7456 patients, the overall pooled risk ratio for the association between transfusion thresholds (restrictive vs liberal) was 0.92 (95% CI, 0.82-1.04; P = .206) as shown in the forest plot in Figure 2. Heterogeneity was not significant (Cochran Q test, P = .380; I² = 6.3%; τ² = .0041). Publication bias was not evident through inspection of the funnel plot or by the Harbord test (P = .465) or Peters test (P = .822).

For trials in which all serious infections were combined as the outcome, the pooled risk ratio was 0.84 (95% CI, 0.73-0.96; P = .012) with no significant heterogeneity (Cochran Q test, P = .577; I² = 0%; τ²<.0001). Absolute risks of infection were also pooled. For those patients receiving the restrictive transfusion strategies, the summary risk of infection (all serious infections) was 10.6% (95% CI, 5.6%-15.9%) and, for patients receiving the liberal transfusion strategies, the summary risk of infection was 12.7% (95% CI, 7.0%-18.7%). The NNT with a restrictive strategy in order to prevent all serious infections was 48 (95% CI, 36-71) and the number of avoided infections per 1000 patients was 21 (95% CI, 14-28).

Trials restricted to those with concealed randomization yielded a pooled risk ratio (RR) for all serious infections of 0.83 (95% CI, 0.72-0.95; P = .009; Cochran Q test, P = .581; I² = 0.0%; τ² < .0001). When the studies were limited to those in which there were <5% withdrawals, the pooled RR was 0.84 (95% CI, 0.73-0.97; P = .019; Cochran Q test, P = .563; I² = 0%; τ²<.0001). Restriction to studies with <5% protocol violations yielded a pooled RR of 0.84 (95% CI, 0.68-1.03; P = .095; Cochran Q test, P = .348; I² = 9.0%; τ²=.0044).

Of the 17 trials shown in Figure 2, 7 trials exclusively used RBC units that were leukocyte-reduced. Restriction of the meta-analysis to these 7 trials yielded a pooled RR of 0.83 (95% CI, 0.69-0.99; P = .044; I² = 0%; τ²<.0001). Thus, the risk of infection remained reduced with a restrictive strategy, even with leukocyte reduction.

We also conducted the meta-analysis with stratification by clinical setting, which is shown in Figure 3. There was no statistically significant difference in the risk of infection for cardiac patients (RR, 1.30 [95% CI, 0.85-1.97]; P = .229), individuals who were critically ill (RR, 0.83 [95% CI, 0.65-1.04]; P = .104), patients with acute upper gastrointestinal bleeding (RR, 0.90 [95% CI, 0.69-1.17]; P = .412) and low birth weight infants (RR, 1.06 [95% CI, 0.85-1.31]; P = .627). However, there was a significant difference in the risk of serious infection in patients undergoing hip replacement or knee replacement surgery who were randomized to the restrictive strategy (RR, 0.72 [95% CI, 0.53-0.97]; P = .034) compared with those in the liberal strategy groups; this excludes those trials in which urinary tract infections were reported.

Three of the randomized trials used a low hemoglobin threshold of <7.0 g/dL in the restrictive group whereas the remaining trials used higher thresholds. The pooled RR for these 3 trials was 0.86 (95% CI, 0.72-1.02; P = .078) with no significant heterogeneity (Cochran Q test, P = .892, I² = 0%, τ²<.0001). The forest plot of these trials is shown in eFigure 1 in the Supplement.

There were 15 trials with data necessary for calculating differences in hemoglobin thresholds between liberal and restrictive groups. The remaining trials had thresholds that varied by patient age, time since surgery, blood sampling methods, or respiratory support. The relationship between the difference in thresholds and risk ratios of infection is shown in eFigure 2 in the Supplement, generated from meta-regression. The RR was 0.91 (95% CI, 0.61-1.36) when 15 trials were included and 0.90 (95% CI, 0.55-1.48) when restricted to the 7 trials in which all serious infections were included.

Among hospitalized patients, a restrictive RBC transfusion strategy compared with a liberal transfusion strategy was not associated with a reduced risk of health care–associated infection overall, although it was associated with a reduced risk of serious infection. The best evidence is provided by those trials that minimized reporting bias by including all infections and that were not confounded by leukoreduction (ie, all patients received leukocyte-reduced RBC units).

Leukocyte reduction of RBCs has been shown to decrease the risk of health care–associated infection⁷ and because only 85% of RBC units are leukocyte-reduced in the United States, adoption of universal leukocyte reduction may be an important first step to infection prevention. However, even with leukocyte-reduced RBC units, adherence to a restrictive transfusion strategy is an important second step in preventing health care–associated infections. The elevated risk for health care–associated infection with the liberal use of leukocyte-reduced RBC units may stem from the persistence of transfusion-related immunomodulation despite reduction of leukocytes. The exact components in allogeneic transfusions that mediate immunomodulation are not fully known and may not all be derived from leukocytes. Additionally, it was not clear if leukocyte reduction occurred before or after storage of the RBC units in all of the studies.

The results showed little statistical heterogeneity overall, although stratification by patient type did suggest clinical heterogeneity. For individuals undergoing hip or knee arthroplasty and for patients with sepsis, a restrictive RBC transfusion strategy yielded significantly reduced risks of serious infection than more liberal strategies, a finding that could lead to decreased morbidity as well as considerable potential cost savings if the restrictive strategy was uniformly applied nationwide. There was, however, 1 orthopedic trial¹⁴ in older, frail individuals that suggested similar infection rates and this may be a population that deserves further study. This study used a slightly different outcome—time to first treatment-requiring infection as indicated by a positive urine culture or suspected infection.

We found that transfusion strategies yielded similar infection risks in patients with cardiac disease. The pooled point estimate was elevated at 1.30 (95% CI, 0.85-1.97) with little statistical heterogeneity. In some of the cardiac trials, the differences between RBC transfusion use in the study groups were not particularly great. For example, in the Bracey and coauthors²⁰ trial, 60% of patients in the restrictive group and 64% of patients in the liberal group received a transfusion. In 1 cardiac trial,²⁴ nonadherence to assignment occurred in a large proportion of the participants, making it difficult to evaluate the comparisons presented. Further investigation into the effectiveness of RBC transfusion strategies in cardiac patients may benefit from complete ascertainment of infectious outcomes. In a recent study at 10 centers in the United States and Canada, 5158 cardiac patients were followed-up with uniform collection of infection data and 5.8% were found to have major infections after cardiac surgery.³⁹ The most prevalent types of infection in this cohort were pneumonia, Clostridium difficile infection, and bloodstream infection. There was a 29% increase in the risk of major infection with each RBC unit transfused.³⁹ In the Transfusion Requirements After Cardiac Surgery (TRACS) randomized trial²¹ included in our systematic review (the largest randomized trial of transfusion strategies in cardiac patients to date), the risk of infection after cardiac surgery increased 20% with every RBC unit received (P = .007). The infectious complications measured in this trial were septic shock, mediastinitis, and pneumonia.

We found only 2 trials in critically ill patients, 1 in adults and another in children.^26,27 Each of these trials encompassed heterogeneous groups of patients, presenting with cardiac disease, pulmonary disease, sepsis, and multiple coexisting conditions. Both risk ratios were slightly below the null but, when pooled, the results were not statistically significant. Although no overall effects were seen, it may be valuable to assess whether there are benefits of a restrictive strategy within particular subgroups of critically ill patients because 1 trial reported lower health care–associated infections in patients with sepsis and restrictive thresholds.³⁵ Hébert and coauthors,²⁶ in particular, noted heterogeneous results in patients who were critically ill.

The results of this review give further support to a recent systematic review and clinical practice guideline put forth by the AABB (formerly the American Association of Blood Banks).⁴⁰ This guideline recommends adherence to a restrictive transfusion strategy for the majority of hospitalized patients and lists specific hemoglobin-based recommendations for varied patient populations; yet only 27% of hospitals that responded to the 2011 National Blood Collection and Utilization Survey reported implementing restrictive use of transfusions postoperatively.⁶ Additionally, only 31% of responding hospitals reported having a blood management program.⁶

There are limitations to this systematic review. First, reporting of infectious outcomes varied across studies. In some trials, all infections were listed whereas in others only specific types of infections were reported. Second, strategies were employed with varying transfusion triggers, although half of the trials used a threshold of <8.0 g/dL in the restrictive group. Third, the decision to transfuse is a judgment that depends on clinical signs and symptoms and not solely on a laboratory value. This was explicitly incorporated into the protocol for several of the trials in this review.^{10,22,29,33,34} Therefore, the proportions of patients actually transfused varied across the trials. Finally, the data were sparse for certain patient groups, such as the trial in pediatric cardiac intensive care with 60 patients.²² Additional trials may provide greater insight into the risks and benefits of the 2 strategies.

We recommend that future trials of blood management techniques uniformly measure health care–associated infection as patient outcomes and that summary numbers of patients with infection be reported so that hypotheses relevant to transfusion-related immunomodulation can be evaluated. These should include definitions as outlined by the US Centers for Disease Control and Prevention.⁴¹ Recent evidence regarding the use of procalcitonin for the identification of bacterial infections may prove to be valuable in this regard.⁴²

Among hospitalized patients, a restrictive RBC transfusion strategy compared with a liberal transfusion strategy was not associated with a reduced risk of health care–associated infection overall, although it was associated with a reduced risk of serious infection. Implementing restrictive strategies may have the potential to lower the incidence of serious health care–associated infection.

Corresponding Author: Mary A. M. Rogers, PhD, Patient Safety Enhancement Program, Department of Internal Medicine, University of Michigan, 016-422W NCRC, Ann Arbor, MI 48109-2800 (maryroge@umich.edu).

Author Contributions: Dr Rogers had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Rohde, Dimcheff, Saint, Rogers.

Acquisition, analysis, or interpretation of data: Rohde, Dimcheff, Blumberg, Saint, Langa, Kuhn, Hickner, Rogers.

Drafting of the manuscript: Rohde, Blumberg, Kuhn, Rogers.

Critical revision of the manuscript for important intellectual content: Rohde, Dimcheff, Blumberg, Saint, Langa, Hickner, Rogers.

Statistical analysis: Langa, Rogers.

Obtained funding: Rogers.

Administrative, technical, or material support: Blumberg, Kuhn.

Study supervision: Blumberg, Saint, Rogers.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Saint reports giving expert testimony on legal cases focusing on medical malpractice; receiving payment from the National Institutes of Health, Veterans Affairs, and the Blue Cross Blue Shield of Michigan Foundation to his institution; receiving payment for lectures from many academic meetings, group-purchasing organizations, state hospital associations, and nonprofit organizations; and receiving royalties for edited or authored books from Lippincott Williams & Wilkins, McGraw-Hill, and Wiley-Blackwell; and serving on the board for Doximity and Jvion. No other discosures were reported.

Funding/Support: This work was funded by grant 5R21HL093129-02 from National Heart, Lung, and Blood Institute.

Role of the Sponsor: The sponsor had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Correction: This article was corrected on October 15, 2014, to fix errors in the data.