Background
Critically ill patients with sepsis are frequently anemic. The underlying pathophysiology is multifactorial and includes production of inflammatory cytokines that increase hepcidin which reduces iron availability [
1], dilution due to fluid therapy, and blood loss [
2]. Red blood cell (RBC) transfusions to correct anemia can be life-saving but are also associated with a number of potential adverse effects which makes risk–benefit assessment challenging [
3‐
5]. The adverse effects include infections, hemolytic reactions, transfusion-related acute lung injury (TRALI), pulmonary edema due to volume overload (transfusion-associated cardiac overload, TACO) and effects on the immune system with transfusion-related immunomodulation (TRIM) [
6].
The clinical impact of the adverse effects on morbidity and mortality of transfusion of RBCs in sepsis has been investigated in randomized controlled trials (RCT) [
7‐
9]. In the subgroup analyses of septic patients in the TRICC trail [
7] and in the TRISS trail [
8], a liberal RBC transfusion strategy (hemoglobin level > 90–100 g/L) did not confer a benefit as compared to a restrictive strategy (hemoglobin level > 70 g/L). However, in the TRICOP trial, performed in a cohort of septic oncology patients, a liberal transfusion strategy was associated with a lower 90-day mortality [
9]. Furthermore, observational studies have also demonstrated positive effects of RBC transfusions. [
10,
11].
Thus, these data are inconclusive and areas of uncertainty remain. In the RCTs, the time between admission and inclusion was 6 h or longer, which means that the effect of any RBC transfusion given early in the course of sepsis was not studied. Moreover, patients in the restrictive group received on average one unit of blood, meaning that the potential adverse effects of a low dose of RBC transfusion could not be assessed. Furthermore, a higher proportion of patients in the restrictive group discontinued the study which could have biased the results.
In an attempt to address some of these uncertainties, we propensity score-matched patients with severe sepsis or septic shock who received low-grade RBC transfusions any of the first 5 days after intensive care unit (ICU) admission to those who did not receive RBC transfusions and evaluated the effect on mortality and organ failure.
Methods
Data collection and study population
This study was approved by Swedish Ethical Review Authority in Lund, Sweden (registration numbers 2014/916 and 2018/866). All participants were offered an opt-out via advertisement in the local newspaper and the board waived the requirement for written informed consent. The manuscript was prepared according to the STROBE guidelines for observational studies [
12].
All patients ≥ the age of 18 admitted to the 9-bed general ICU at Skåne University Hospital, Lund, Sweden between 2007 and 2018 with severe sepsis or septic shock according to the Sepsis-2 definition were eligible for inclusion [
13]. For patients with multiple admissions with the diagnosis of severe sepsis or septic shock, only the first admission was included in the study. Day 0 started at admission and ended at 06:00. As described above, a condition with massive bleeding can affect outcome and patients who received high-grade RBC transfusion (> 670 ml [= more than 2 units]) during the first 5 days were, therefore, excluded. RBC transfusions were given at the discretion of the treating physician but in local guidelines, it was recommended to keep hemoglobin level above 100 g/L in critically ill patients with severe sepsis or septic shock.
Mortality data were extracted from the Swedish intensive care quality register PASIVA (Otimo Data AB, Kalmar, Sweden). Physiological and laboratory data and pre-existing conditions (age, gender, chronic obstructive pulmonary disease (COPD), renal failure, diabetes), outcome variables (except mortality) and fluid administration data were collected from raw data, i.e., from the electronic master chart system of the hospital (Melior, Cerner, N. Kansas City, MO, USA), or from the patient data management system at the ICU (Intellispace critical care and anaesthesia (ICCA), Philips, Amsterdam, the Netherlands).
Outcome variables
Mortality was assessed at 90 and 180 days after ICU admission. Organ support was assessed by calculating days alive and free (DAF) of organ support for the first 28 days after admission to the ICU [
14]. For patients who died in the ICU, we counted the days without the specified organ support before death. Organ support measures were vasopressors for circulatory failure, mechanical invasive ventilation for respiratory failure and renal replacement therapy (RRT) for kidney failure. To further assess organ failure, the maximum sequential organ failure assessment (SOFA) score during the first 28 days after admission was analyzed. Kidney failure was also evaluated according to the acute kidney injury network (AKIN) scoring system [
15]. The maximal AKIN score the first 10 days after ICU admission was used for analysis.
Statistics
Patients receiving low-grade RBC transfusion during the first 5 days of ICU admission were propensity score matched with non-transfused patients to adjust for differences in baseline variables associated with outcome. The propensity score was calculated with linear logistic regression using a one-to-many macro for SAS [Parsons 2004] with the covariates specified in Table
1. Physiological and laboratory variables used in the propensity score matching were collected within 90 min of admission to the ICU. Note that the hemoglobin value at admission was not included in the matching in the primary analysis. In a secondary sensitivity analysis, the median hemoglobin value day 0 was included in the matching. A greedy matching procedure in both the primary and secondary analyses matched treated to controls at a ratio of 1:1. In a first step, a match was sought with a propensity score that was identical to 8 decimal places to the treated patient. If no match was found, a match would be sought at 7 decimal places and so on. If no match was found at one decimal place, the patient receiving RBC transfusion was excluded from the study. A control could only be used once. A standardized difference of < 10% has previously been suggested to indicate negligible differences in the mean or prevalence of covariates between groups [
16,
17].
Table 1
Patient demographics before and after propensity matching
Pre-existing conditions |
Age, mean (SDb) | 64 (16) | 65 (15) | 0.047 | 0.490 | 65 (15) | 65 (15) | 0.010 | 0.914 |
Male gender, no (%) | 233 (57) | 234 (51) | 0.131 | 0.054 | 130 (55) | 138 (58) | 0.068 | 0.460 |
Blood malignancyc, no (%) | 16 (4.4) | 42 (9.2) | 0.187 | 0.007 | 12 (5.1) | 14 (5.9) | 0.037 | 0.687 |
COPDd, no (%) | 42 (10) (0.305) | 52 (11) | 0.030 | 0.652 | 29 (12) | 28 (12) | 0.013 | 0.888 |
Cirrhosis, no (%) | 17 (4.2) | 12 (2.6) | 0.087 | 0.198 | 11 (4.6) | 8 (3.4) | 0.065 | 0.483 |
Immunosuppressione, no (%) | 30 (7.4) | 54 (12) | 0.148 | 0.031 | 18 (7.6) | 19 (8.0) | 0.016 | 0.864 |
Malignancyf, no (%) | 43 (11) | 73 (16) | 0.156 | 0.023 | 31 (13) | 33 (14) | 0.025 | 0.789 |
Nosocomial infectiong, no (%) | 33 (8.1) | 44 (9.6) | 0.051 | 0.460 | 21 (8.9) | 24 (10) | 0.043 | 0.639 |
Airway infection, no (%) | 112 (28) | 116 (25) | 0.054 | 0.428 | 63 (27) | 63 (27 | 0 | 1.000 |
Surgeryh, no (%) | 64 (16) | 99 (22) | 0.148 | 0.031 | 40 (17) | 48 (20) | 0.087 | 0.346 |
GIi-bleeding, no (%) | 0 (0) | 6 (1.3) | 0.163 | 0.021 | 0 (0) | 0 (0) | 0.000 | 1.000 |
DICj, no (%) | 25 (6.2) | 33 (7.2) | 0.041 | 0.552 | 19 (8.0) | 14 (5.9) | 0.083 | 0.368 |
I.C.k volume effect, no (%) | 2 (0.5) | 4 (0.9) | 0.046 | 0.505 | 2 (0.8) | 2 (0.8) | 0 | 1.000 |
Physiological and laboratory variables at admissionl, mean (SD) |
Heart rate, mean (SD) | 107 (23) | 109 (24) | 0.085 | 0.216 | 106 (23) | 106 (25) | 0.013 | 0.889 |
SBPm, (mmHg) | 108 (30) | 107 (29) | 0.052 | 0.450 | 107 (30) | 108 (31) | 0.044 | 0.634 |
Lactate (mmol/L) | 3.4 (3.2) | 3.5 (2.7) | 0.027 | 0.691 | 3.4 (3.2) | 3.4 (2.7) | 0 | 0.996 |
Norepinephrine (µg/min) | 7.0 (12) | 11.6 (20) | 0.285 | < 0.001 | 9.0 (13) | 7.7 (11) | 0.108 | 0.241 |
Temperature (°Celcius) | 37.4 (1.5) | 37.4 (1.4) | 0.035 | 0.617 | 37.3 (1.5) | 37.3 (1.4) | 0.007 | 0.940 |
PaO2/FiO2 (kPa) | 23 (15) | 22 (16) | 0.040 | 0.588 | 23 (15) | 22 (15) | 0.040 | 0.665 |
Leucocytes (× 109/L) | 16 (18) | 14 (19) | 0.093 | 0.197 | 16 (20) | 15 (20) | 0.038 | 0.683 |
Platelets (× 109/L) | 188 (130) | 182 (135) | 0.044 | 0.517 | 187 (129) | 185 (120) | 0.018 | 0.845 |
pH | 7.13 (1.5) | 7.31 (0.50) | 0.261 | <0.001 | 7.34 (0.12) | 7.34 (0.11) | 0.042 | 0.647 |
Bilirubin (µmol/L) | 24 (35) | 25 (46) | 0.010 | 0.891 | 24 (33) | 26 (50) | 0.053 | 0.561 |
Creatinine (µmol/L) | 172 (126) | 180 (141) | 0.063 | 0.360 | 181(130) | 175 (133) | 0.042 | 0.645 |
PT/INRn | 1.58 (0.85) | 1.62 (0.82) | 0.040 | 0.550 | 1.58 (0.70) | 1.59 (0.78) | 0.005 | 0.956 |
APTTo (sec) | 42 (19) | 45 (17) | 0.178 | 0.009 | 43 (18) | 42 (14) | 0.050 | 0.586 |
Sample size was based on the number of available patients during the study period. Variables were summarized using mean or median with standard deviation or range as distribution measurement. The propensity score matching was performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) prior to any comparison between the groups. Kaplan–Meier survival analysis was performed and is presented in graphs with corresponding stratified log-rank test. In accordance with the previous recommendations [
18] comparisons between the groups after propensity score matching was performed with paired hypothesis testing using SPSS Statistics version 24 (SPSS Inc., Chicago, Ill., USA). A two-sided P value of less than 0.05 was considered to indicate statistical significance. Continuous variables are presented as median (interquartile range) and all categorical variables are presented as numbers (percentage).
Discussion
In the present study, low-grade leukoreduced RBC transfusion in critically ill septic patients was associated with increased mortality, increased kidney, circulatory and respiratory failure as well as higher SOFA-max score.
With the propensity score matching, we aimed to create the RBC and the control groups as similar as possible with respect to severity of illness at ICU admission. Because a low hemoglobin at admission may be a result of an underlying condition which may influence outcome independent of RBC transfusions, we performed a sensitivity analysis in which hemoglobin concentration at admission was included in the propensity score matching. Although the results of this analysis are slightly less robust due to the lower number of matched patients and hence lower power, the results were very similar. Further, pre-matching differences between the groups in baseline variables not included in the propensity score matching were erased after the matching for all variables except for “Gastric reason for admission” (Table
2). Taken together, this supports the robustness in the propensity score matching and in the findings in the main analysis.
The observed hemoglobin level of 95 g/L before transfusion may appear high given that current guidelines suggest a transfusion trigger of 70 g/L in the absence of ongoing ischemia. As mentioned above, the present study was performed in an institution with a liberal tradition of RBC transfusions and before the publication of the TRISS trial a transfusion trigger of 100 g/L was accepted in hemodynamically unstable patients with sepsis. After 2014, a more restrictive approach was implemented and the observed hemoglobin level before transfusion, therefore, reflects this change of practice.
In contrast to our results, two randomized trials comparing liberal (hemoglobin level goal > 90–100 g/L) vs restrictive (> 70 g/L) transfusion strategy in critically ill patients with sepsis or septic shock did not detect a difference in mortality or need for vasopressors, mechanical ventilation or RRT between the treatment strategies [
7,
8]. It should be noted that previous RCTs on different hemoglobin thresholds for RBC transfusions did not study the possible negative effects of RBC transfusions but rather the effects of different thresholds for RBC transfusions [
7‐
9]. This means that also patients in the low threshold group received a significant number of RBC units. In contrary, the present study included a control group with patients who did not receive any RBC transfusions at all during the study period. Thus, when evaluating if RBC transfusions are harmful, independent of hemoglobin levels within a safe interval, observational studies like the present may have some benefits. Our result demonstrated that the majority of RBCs were administered within the first days after ICU admission. Interestingly, the largest RCT on transfusion thresholds in septic shock to date, the TRISS trial, included patients on average 22 h after admission and could not demonstrate a difference between a low and high transfusion trigger [
8]. This raises the possibility that the effect of transfusions on outcome could be time dependent, in the sense that the early RBC transfusions given to septic patients might be the transfusions with highest risk.
As mentioned above, previous observational studies on the effect of RBC transfusions in sepsis have reported a decrease in mortality [
10,
11]. In contrast to our study, these studies did not exclude patients receiving massive transfusions. Because massive transfusions indicate active bleeding, a condition in which the risk–benefit ratio may favor transfusion, it is possible that this difference in study design may explain the difference in results. Moreover, the average transfused patients in both of these studies were transfused at hemoglobin of 75–80 g/L which could contribute to the difference in results.
What is the potential pathophysiological mechanism of the observed increased mortality and morbidity after RBC low-grade transfusions in our study? As mentioned above, known adverse effects of RBC transfusion include TACO, TRALI and TRIM, all of which may cause negative effects in many organs [
6]. Although patients in the RBC group received more fluids the first 5 days, the fluid balance was not different between the groups, which would indicate that fluid overload (TACO) was not the reason for the differences in outcome. The incidence of TRALI is previously estimated to be about one case per 12 000 transfusions [
19]. Given that TRALI most commonly occurs after plasma transfusion and that no episodes of TRALI presentation were reported for included patients we believe that it is unlikely to the main cause of the differences between the groups even if underreported. TRIM represents an interaction of a multitude of immunomodulatory mediators in the RBC transfusion with the immune system, leading to both proinflammatory and immunosuppressive effects [
20]. In the setting of sepsis, such effects may be deleterious and represents a potential mechanism by which RBCs may adversely affect outcome.
Limitations
We recognize the limitations in the present study due to its retrospective nature. Although baseline values were carefully adjusted for severity of illness, the presence of undetected factors of relevance for outcome such as differences in comorbidities cannot be ruled out. Further, the study was done at a single department which limits the external validity. The risk–benefit ratio for transfusions is likely to be dependent on the Hb level at which RBC are transfused. Thus, it is important to emphasize that the results may only be valid for the transfusion level observed in our cohort and may not be generalized to other transfusion triggers.
Although there is evidence for the safety of a restrictive transfusion strategy in sepsis, we still lack knowledge of safety in certain sub-populations such as septic patients with myocardial ischemia, severe hypoxemia or acute hemorrhage. Furthermore, the potentially harmful effects of RBC transfusion on morbidity and long-term mortality warrants further evaluation including, e.g., individualized therapy and methods of preventing or treating anemia without RBC transfusion.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.