In this multi-center, retrospective observational study, the RBC transfusion practices in nine adult medical-surgical ICUs were examined. We identified a total of 4632 unique RBC transfusions events in 2287 mixed medical and surgical ICU admissions. Patients within this cohort received approximately two transfusions per ICU admission, with 1 RBC unit transfused per event, and had a mean pre-transfusion hemoglobin level of 73.4 g/L. Among 4487 RBC transfusion events, 61% were associated with a pre-transfusion hemoglobin value at or above 70 g/L. Over the 33-month study period, this was estimated to cost over $1.8M CAD to the healthcare system. Factors such as being male, 75 years of age or older, admission to the ICU from the operating room, and an ICU admission diagnosis of gastrointestinal bleeding or trauma were positively associated with RBC transfusions events with a hemoglobin value greater or equal to 70 g/L. Moreover, having a pre-transfusion hemoglobin value at or above 70 g/L was also associated with an increase in ICU mortality. There was no impact on overall hospital mortality.
We specifically sought to examine RBC transfusions among critically ill patients for whom a restrictive transfusion strategy is supported by the most current, high-quality evidence [16,17,18,19,20,21]. In doing so, we acknowledged that a restrictive pre-transfusion hemoglobin threshold will not apply in all clinical situations. For instance, in some hemorrhagic or ischemic events, clinical judgment should subvert laboratory value-based thresholds. To account for such reasonable clinical exclusions, we excluded from analysis a considerable number of patients (i.e., greater than 50% of ICU admissions with a RBC transfusion event) for whom a restrictive transfusion strategy has not been proven safe, nor superior, to a liberal transfusion strategy (e.g., chronic anemia, active blood loss, acute coronary syndrome, myocardial infarction, and neurological or traumatic brain injury) . With our final cohort, we, therefore, aimed to decrease the potential for misclassifying a RBC transfusion event—for which a restrictive transfusion approach would not be appropriate—and allowed a conservative evaluation of RBC transfusion practices in the ICU.
Reduced exposure to RBCs, through the adoption of a restrictive pre-transfusion hemoglobin threshold (i.e., less than 70 g/L), has been increasingly recognized in transfusion guidelines as the best practice for most stable, non-bleeding adult patients in the ICU [7, 15, 27, 28]. Within our cohort, we found that over half of all examined RBC transfusions may have occurred outside of the recommended best practice. While these findings suggest potential over-transfusion in the ICU, there have been minor improvements in transfusion practices over time. The reason for this is unknown but is likely the result of increasing awareness and attention to the subject. We also found that over 40% of RBC transfusions examined in our study were associated with a pre-transfusion hemoglobin value between 70 g/L and 79 g/L. Visual inspection of the stratified patient characteristics (Additional file 2) did not suggest differences between transfusions associated with a hemoglobin value below 70 g/L and 70–79 g/L. Interestingly, previous qualitative studies that have examined factors influencing physician transfusion behaviors found the greatest uncertainty among physicians when deciding whether to transfuse patients with a hemoglobin level within this borderline range [29, 30]. Such uncertainty may similarly be reflected in our findings, and targeted efforts to better inform physician decision-making for transfusing such patients may be warranted.
Similar studies examining the appropriateness of RBC transfusion practices have been described in the literature. Previous retrospective audits, for example, have primarily focused on characterizing mean pre-transfusion hemoglobin levels and found them to range substantially between 71 g/L and 91 g/L for most non-hemorrhagic, ICU patient populations [31,32,33,34]. One of the larger studies conducted in the USA, a longitudinal analysis RBC transfusion practice between 1997 and 2007, reported a similar mean pre-transfusion hemoglobin level to our patient cohort after their 10 year follow-up period (significantly decreasing from 79 ± 1.3 to 73 ± 1.3 g/L) . In addition, Netzer et al.  observed a significant decrease in the proportion of patients who were transfused at a hemoglobin level of less than 70 g/L. In contrast, studies that have examined average transfusion volumes have reported higher numbers of RBC units per transfusion event compared to our study results, ranging from 2 to 4.3 RBC units [32, 35]. While it is difficult to reconcile specific reasons for observed differences between these previous studies and our present findings, some differences are likely attributable to variations in the patient case-mix, ICU structure or culture, or even the time since the publication (and acceptance) of seminal and relevant literature and guidelines.
We conducted additional analytical investigations that represent novel contributions to the existing literature and offer considerations for policy and practice, as well as future research. The costing analysis, for instance, enabled the valuation of not only RBC transfusions across the ICUs, but also of the opportunity cost of potentially inappropriate practices over the 33-month study period. Presenting such costing outcomes provides another perspective for clinical experts to reflect on as the stewards of healthcare resources. This information can also help healthcare system decision-makers implement interventions whose implementation costs could still yield a reasonable return on investment and, as such, improve the overall efficiency of care delivered.
We also identified several factors associated with RBC transfusions with a hemoglobin level above 70 g/L. In particular, ICU admission diagnoses of gastrointestinal bleeding and trauma were associated with at least a 30–50% increase in the odds of these transfusions. Despite the evidence supporting the use of a restrictive RBC transfusion strategy for both of these patient groups , as well as our exclusion of transfusions among patients with active bleeding, the propensity for hemorrhagic outcomes with such conditions may account for the increased odds in the observed cases [36, 37]. Due to the overlap in 95% confidence intervals across point estimates, we did not find any one (or few) factor(s) that markedly increased the odds of transfusion over other factors. This suggests that there may be several drivers, both characterized and uncharacterized (e.g., level of physician experience, cultural influences) in our present work that act in concert and underlie the observed proportion of overuse.
In addition, the downstream impacts of pre-transfusion hemoglobin values in our transfused patient cohort differed by ICU and hospital. With respect to mortality, we found that the odds in the ICU were increased for transfused patients with a pre-transfusion hemoglobin of 70 g/L or more, but there was no difference in hospital deaths. Previous meta-analyses of RCTs comparing restrictive versus liberal hemoglobin thresholds found that there was no difference in ICU mortality between groups, yet the risk ratio for hospital mortality was lower among those randomized to the restrictive threshold . While these differences in mortality outcomes may be due to unmeasured confounding in our observational data, our present findings still do not indicate an increased risk of harm for transfused patients with a lower pre-transfusion hemoglobin value (i.e., below 70 g/L).
There are limitations to the present study worth noting. While we were able to capitalize on population-based administrative and clinical data sources, we may have excluded possible confounders and/or introduced other sources of bias due to the retrospective observational design of the study. Moreover, we did not have adequate recording in the data of the exact reason for a given transfusion across all ICUs. Therefore, patients may have been misclassified and actually had hemorrhage or ischemia. While our conservative inclusion criteria were applied in attempt to mitigate such misclassification, supplementing secondary data sources with medical record audits may additionally aid in overcoming misclassification.
Given the de-identified nature of the data, we were unable to examine the association of factors such as hospital type (i.e., teaching hospital, community hospital) and level of physician experience in our logistic regression analyses. We were also unable to identify the consequences directly associated with a potentially inappropriate transfusion (i.e., transfusion-related complications) due to the retrospective design of our study. Lastly, the present analysis focused on RBC transfusion events in medical-surgical ICUs in one Canadian province; the generalizability of our findings to other Canadian and/or international critical care contexts is unclear. Future studies designed with more detailed prospective data collection and with inter-provincial or international cohort comparisons could, therefore, expand understanding in these areas.