In this large cohort of surgical ICU patients, hemoglobin concentrations were less than 9 g/dl on at least one occasion in 57.6% of patients. Lower hemoglobin concentrations were associated with higher morbidity and mortality. In a multivariable analysis, higher hemoglobin concentrations and blood transfusions were independently associated with a lower risk of in-hospital death, especially in patients aged from 66 to 80 years, in patients admitted to the ICU after non-cardiovascular surgery, in patients with higher severity scores, and in patients with severe sepsis.
In this study, we demonstrate that anemia is common in surgical intensive care patients. The cause of anemia in these patients is likely to be multifactorial [4
]. The retrospective design of our study does not allow us to elaborate on the exact cause of the low hemoglobin concentrations. Nevertheless, we found that lower hemoglobin concentrations were associated with poor outcome even after adjustment for possible confounding factors. Our data confirm the results of previous studies in mixed populations of medical and surgical critically ill patients [1
], in surgical patients who declined blood transfusions [22
], and in patients with ischemic heart disease [24
]. We additionally demonstrate a correlation between hemoglobin concentrations and organ dysfunction/failure as assessed by the SOFA scores in these patients.
Blood transfusion has also been thought to increase the risk of death in ICU patients [1
]. Indeed, transfused patients in our study had higher ICU and in-hospital mortality rates; however, after adjustment for possible confounders and severity of illness, blood transfusion was associated with a lower risk of in-hospital death. The discrepancy between our results and those of previous observational studies [1
] may be related to the implementation of leukoreduction in our institution. Hébert and colleagues [15
] reported reduced in-hospital mortality rates after implementation of leukoreduction in a large Canadian multicenter study compared with the control period. van de Watering and colleagues [16
] showed increased survival rates in post-cardiac surgery patients transfused with packed RBCs filtered to remove leukocytes compared with those transfused with blood just treated to remove buffy coats. Another possible explanation may be the different case-mix in our study from those of the previous observational cohort studies [1
], which included mixed medical and surgical ICU patients. Nevertheless, our data support those of the recently published analysis from the SOAP study [2
], in which blood transfusion, mostly with leukoreduced blood, was associated with a lower RR of death.
In-hospital mortality was the primary end point in our study. This was also the primary end point for previous prospective randomized [15
] and observational studies [1
]. Possible deleterious effects of blood transfusions, especially immunosuppression, are expected to occur later in the course of the disease. The relatively short ICU length of stay in our study may, therefore, render the ICU mortality inadequate in this context.
The results of propensity score matching in our study do not exclude beneficial effects of blood transfusion despite similar outcomes between the matched groups. Severely ill patients were not included in this analysis due to the absence of suitable matched pairs. These patients may be more likely to benefit from blood transfusion, a hypothesis supported by the subgroup analysis in our study. The optimal transfusion trigger in ICU patients has been a matter of controversy. Although randomized controlled trials would be the most appropriate means to investigate this issue, observational studies such as ours can provide insight, generate hypotheses, and complement the results of randomized studies. Randomized controlled studies in which subjects are randomized to two different therapeutic strategies, independent of their needs, are at risk of therapeutic misassignment [26
]. Exclusion of subgroups of patients according to study protocol, dropout of others due to declined consent or non-compliance of physicians, and failure of recruitment are all factors that hinder extrapolation of the results of randomized controlled trials to other patient populations with different case mixes. Changes in practice and quality of care over time may be another important factor that necessitates reassessment of current treatment strategies. Although the TRICC study [9
] demonstrated that a restrictive strategy of blood transfusion was as effective as a liberal strategy, leukoreduction was not implemented at the time that study was performed. Whether or not the results of the TRICC study have changed transfusion practice in ICUs is unclear. The mean pre-transfusion hemoglobin concentration in our study was 8.2 g/dl, which is similar to a large multicenter observational study [3
] performed after the results of the TRICC study were published [9
] and the evolution of hemoglobin concentrations in our study was also similar to that reported in this study. This could be explained by the limitations of the TRICC study [9
] that may hinder the adoption of the restrictive transfusion strategy in all ICU patients.
We also identified subgroups of patients that are more likely to benefit from blood transfusion, including patients with higher severity of illness and more organ dysfunction. These data may help in guiding transfusion practice in surgical ICU patients, until the results of relevant randomized trials are available.
To the best of our knowledge, our study is the largest to date investigating the impact of anemia and possible risks of blood transfusion in surgical intensive care patients. However, some limitations should be considered. First, our analysis is retrospective in nature and our results are only hypothesis generating. A randomized controlled trial is warranted to clarify this issue. Second, the multivariable analysis does not take into account unmeasured variables and can not establish a cause-effect relation. The confounding effect of unmeasured variables can not be excluded. Nevertheless, many relevant variables were considered. Third, similar to previous observational [1
] and interventional studies [9
], the impact of blood transfusions given before and after the ICU stay on outcome was not evaluated and the indication for blood transfusion was not identified. Fourth, the indication for blood transfusion was not considered in our analysis and may have been an important confounding factor. However, indication for blood transfusion is usually influenced by hemoglobin concentrations, comorbidities, and severity of illness, all of which are factors that were considered in our analysis. Finally, the results of our study may not be extrapolated to patients with other case mixes, such as medical patients.