Association of red blood cell transfusion and in-hospital mortality in patients admitted to the intensive care unit: a systematic review and meta-analysis

There is no consistent evidence linking RBC transfusion to in-hospital mortality in ICU patients

The initially overall pooled results suggested that allogeneic blood transfusion was associated with an 11% to 85% increased risk for in-hospital mortality. However, after excluding studies with unadjusted estimates only [19],[31] or studies with median overall quality [31],[32],[42], the association became null. Therefore, we could not draw a definitive conclusion. The results of sensitivity analyses suggested that such risk association may be attributable to studies without adjustment for confounders. In studies adjusted for confounders, studies that reached effect size by using the Cox regression model, and studies that implemented detailed information about leukodepleted blood measurement, we found that RBC transfusion did not increase risk of in-hospital death.

There is no evidence indicating that RBC transfusion is associated with increased mortality in patients admitted to ICU for sepsis and shock, trauma and surgical reasons.

Type of patients is an extremely important confounder for survival after RBC transfusion. Different types of patients have different responses to RBCs transfusion. For example, age of transfused RBCs could potentiate the risk association of volume of RBC transfused in trauma patients, but less so in other types of patients [44],[45]. In our subgroup analysis, we separate enrolled studies into four groups according to patients’ population: all ICU admissions (seven studies), sepsis and shock (two studies), surgical (three studies), trauma (three studies), and others (three studies, one on patients with ALI or ARDS, one on patients with intracerebral hemorrhage, and one on patients with anemia and renal dysfunction). As presented in Table 3, the results suggested that there was no association between RBC transfusion and mortality in sepsis and shock, surgical, trauma and `other’ patients. The pooled overall effect estimate of seven studies on all ICU admissions was significant (RR 1.513, 95%CI 1.123 to 2.039). However, in these seven studies, we noticed that one study did not adjust for any confounders, therefore had a high risk of bias [19]. Actually, after excluding this study, we observed a 7% reduction but still significant pooled overall effect estimate (RR 1.406, 95% 1.016 to 1.945). In a large, multiple center study by Vincent et al., admission type of patients were elective surgery (41.4%), medical (32.6%), emergency surgery (16.6%), trauma (7.6%) and other (1.6%) [2]. In our analysis, we were unable to identify medical ICU patients as an independent subgroup. Literature on the safety of RBC transfusion in medical ICU patients was poor. In the SOAP study, medical admission was found to be an independent predictor for risk of 30-day mortality in ICU patients (RR 1.47, 95% 1.24 to 1.74) [35]. Further studies are needed to identify whether medical admissions are less tolerant of RBC transfusion than other patient populations.

There is a lack of evidence on whether leukodepleted RBC transfusion could reduce in-hospital mortality in ICU patients.

Immunomodulating effects of transfused RBCs have been suggested as potential cause of many adverse effects [46]. Leukocytes from donated blood may play a key role in immune system suppression; however, the exact mechanism is unclear [47]. Studies had reported a decrease in mortality rate among patients receiving prestorage leukoreduced blood compared to patients receiving nonfiltered blood [22],[23]; however, a meta-analysis of 12 RCTs comparing leukodepleted and nonleukodepleted RBCs failed to identify an association between leukoreduction and mortality (OR 1.14, 95%CI, 0.89 to 1.45) [48]. Of the eighteen studies included in our review, only seven of them had information on percentage of patients receiving nonfiltered blood. Filtered RBCs were routinely used in two studies, Sakr et al. [21] suggested that RBC transfusion was independently associated with lower risk of mortality (RR 0.96, 95%CI, 0.95 to 0.98) while Engoren et al. [38] found no association. In our subgroup analyses, studies were divided into four groups according to the usage of leukoreduction. In a subset of 11 studies that did not report information on leukoreduction, the pooled effect estimate suggested that RBC transfusion was associated with higher mortality (RR 1.851, 95%CI, 1.229 to 2.786); however, the pooled effect estimates of other subgroups were not significant, regardless of the degree of implementation of leukodepleted blood (Table 3). Studies lacking information on leukoreduction had a higher risk of in-hospital mortality. Because of insufficient information, we cannot tell whether the use of leukodepleted blood could decrease mortality in transfused patients. Another possible explanation is that studies lacking leukoreduction information also had lower qualities, hence indicating high risk of bias. We recommend that further studies should take leukoreduction implementation into consideration.

Study and statistical analysis methods are of considerable importance in explaining the association between RBC transfusion and in-hospital mortality.

To model the association between RBC transfusion and in-hospital mortality, 10 of the 18 included studies used multivariable logistic regression, and reported OR as the outcome measurement. Pooled effect estimate of this subset of studies found that there is a risk association between RBC transfusion and mortality (RR 1.465, 95%CI, 1.049 to 2.045). In contrast, in five studies used Cox regression, there was no independent association between RBC transfusion and mortality (RR 0.901, 95%CI, 0.622 to 1.305). This may be because that selection of statistical model may have effect on the final result. As logistic regression did not take time-dependent variables into consideration, with increasing follow-up time, the logistic regression coefficients become less reliable [49]. For the same transfusion data, treating transfusion and living status as a time-dependent variable or not may result in different conclusions. Our review suggests that the statistical analysis method may cause disagreement between previous studies.

Previous research revealed that older and more severely ill patients were more likely to have comorbidities after RBC transfusion [2],[4]. The quality of a study relies on the completeness of data on possible confounders [34]. Patients’ admission characteristics and disease severity are two major confounders that can have an impact on transfused patients’ risk of death. Multivariable confounder-adjustment analyses and propensity score analyses are implemented in most of the studies to control imbalances between transfused and nontransfused groups. Under propensity matched groups, patients were matched by their probability to receive a transfusion, thus statistically reducing confounding bias. Previous findings revealed that propensity score analysis could produce estimates that were more precise and robust than regular logistic regression while there were seven or fewer events per confounding variables, but might produce unstable estimates when events per confounder were more than eight [50]. Most of our included studies reported multivariable confounder-adjusted estimates (Table 3), and the mostly used confounders in adjustments were patient characteristics at admission including age, gender and admission symptom score; seven studies used propensity score analyses. Subgroup analyses of the seven propensity score matched studies and the other nine no propensity score matched confounder-adjusted studies all showed no association between RBC transfusion and death (RR 1.089, 95%CI, 0.767 to 1.546; RR 1.358, 95%CI, 0.965 to 1.910 respectively), while the other two studies with unadjusted effect estimates only showed an increased risk (RR 3.933, 95%CI 2.107 to 7.343). This suggests that studies that did not fully adjust for confounders may overestimate the negative effects of RBC transfusion. This finding supported the view of Ruttinger et al. [34] that RBC transfusion may be only a surrogate marker for disease severity and is not causally related to mortality. Ruttinger’s study even suggested the pitfalls of propensity score and multivariate analyses; both cannot adjust for unobserved or unknown confounders, and not all factors that may influence transfusion practice had been collected. Ruttinger considered variables reflecting disease severity during ICU stay, and found that the number of RBC transfusion units was associated with increased mortality, and the association attenuated but persisted while only adjusted for admission variables. When controlling for variables of disease severity during ICU stay, this association vanished. Disease severity during ICU stay might be particularly important. We suggest that further research should take this into consideration. In our review, we performed meta-regression analyses to assess whether the mean number of units of blood transfused, mean admission Hb level and mean admission symptom score could explain disagreement between study results. Mean number of units of RBC transfused were reported in twelve studies, mean admission Hb level were reported in nine studies, mean APACHE II score were reported in eight studies, mean SOFA score were reported in four studies [2]-[4],[42], and mean SAPS II [35] and mean APACHE III score [39] were reported in only one study, respectively. Results of meta-regression failed to identify significant association across studies for mean number of units of RBC transfused (coefficient -0.1676, 95%CI, -0.3865 to 0.0504), mean admission Hb level (coefficient -0.1848, 95%CI, -0.4343 to 0.0648) and mean APACHE II score (coefficient 0.0202, 95%CI, -0.0569 to 0.0973). In other words, these variables could not explain the disagreement across studies.

Blood transfusion may decrease risk of mortality in older patients.

The slope estimate was significant -0.0417 (95%CI, -0.0680 to -0.0154) (Figure S8 in Additional file 7), suggesting that with older age, transfused patients had better chances of survival. This is interesting since age has long been known as a predictor for worse outcomes including mortality. We speculate that this may be due to the following reasons. (1) Older patients and younger patients have different etiology of anemia. Older patients are more likely to have nutritional deficiency, chronic diseases and unexplained anemia [51]. Consequently, older patients and younger patients may differ in disease severity and clinical treatment. (2) Hb level may be a potential explanation. Using World Health Organization (WHO) definition of anemia (Hb level <13 g/dL in men and <12 g/dL in women), the corrected annual incidence of anemia increased with age [52], and the study found that even a `higher’ Hb level was associated with risk of death in older people [53]. Older patients may have different `optimal’ Hb concentrations, a `mildly low’ and `low-normal’ Hb level may be well tolerated in young patients, but not in older people [54],[55].

Literature on patient’s age, survival and blood transfusion is controversial in defining anemia in older people. Although we identified a significant association between mean age of patient and survival, there was no strong evidence to support this view. Further research could explore whether blood transfusion is more effective in improving survival in older patients.

The debate on whether it is beneficial or harmful for RBC transfusion: articles tend to report lower risk effect estimates.

The number of reports concerning such issues has grown fast in recent years, as shown in Figure S7 in Additional file 8. The meta-regression analysis produced a significant effect for publication year. Recent studies are more likely to report relativity lower risk ratios, indicating that RBC transfusion might have gradually become safer [56]. Safety concerns in RBC transfusion have emerged during the past two decades along with clinical practice changes [57]. The most important change in clinical practice is the recognition that the decision to transfuse should not be driven by a single Hb threshold, but by the physiologic state of the individual patient, presence of comorbidity, cardiopulmonary physiologic parameters, and evidence of blood loss [56],[58]. In our meta-analysis, seven studies from the years 2002 to 2012 had reported average pretransfused Hb levels, and the mean Hb levels ranged from 6.6 g/dL to 8.6 g/dL. Meta-regression for pretransfusion Hb level failed to identify the significance for risk of mortality (coefficient -0.216, 95%CI -0.859 to 0.428).