Relative impact of red blood cell transfusion and anaemia on 5-year mortality in cardiac surgery

Abstract

OBJECTIVES

The aim was to compare the relative effects of red blood cell (RBC) transfusion and preoperative anaemia on 5-year mortality following open-heart cardiac surgery using structural equation modelling. We hypothesized that patient risk factors associated with RBC transfusion are of larger importance than transfusion itself.

METHODS

This prospective cohort study, part of the Cardiac Surgery Outcome Study at St. Olavs University Hospital, Trondheim, Norway, included open-heart on-pump cardiac surgery patients operated on from 2000 through 2017 (n = 9315). Structural equation modelling, which allows for intervariable correlations, was used to analyse pathway diagrams between known risk factors and observed mortality between 30 days and 5 years postoperatively. Observation times between 30 days and 1 year, and 1–5 years postoperatively were also compared with the main analysis.

RESULTS

In a simplified model, preoperative anaemia had a larger effect on 5-year mortality than RBC transfusion (standardized coefficients: 0.17 vs 0.09). The complete model including multiple risk factors showed that patient risk factors such as age (0.15), anaemia (0.10), pulmonary disease (0.11) and higher creatinine level (0.12) had larger effects than transfusion (0.03). Results from several sensitivity analyses supported the main findings. The models showed good fit.

CONCLUSIONS

Preoperative anaemia had a larger impact on 5-year mortality than RBC transfusion. Differences in 5-year mortality were mainly associated with patient risk factors.

The relationship between red blood cell (RBC) transfusion and long-term mortality among cardiac surgery patients has been explored in several previous studies.

INTRODUCTION

The relationship between red blood cell (RBC) transfusion and long-term mortality among cardiac surgery patients has been explored in several previous studies. Some studies found significant associations [1–4], whereas other studies suggest that there is no such significant effect [5–9]. Although the results are conflicting, there is a consensus that RBC transfusion is independently associated with a negative outcome in these patients. Furthermore, patients with low haematocrit values or preoperative anaemia have a higher mortality risk when given RBC transfusion [10–12]. The conflicting findings from observational studies may be attributed to the lack of proper adjustments and overestimation of the effect of RBC transfusion. The increased risk associated with RBC transfusion seen in earlier studies may therefore be due to residual confounding. Based on previous results from our group, we hypothesized that patient and operative risk factors are more important for mortality than transfusion itself [8, 9].

Traditionally, studies on this subject have employed a factorial design in their survival analysis. A limitation of this method is not permitting adequate analysis of intervariable correlations and more complex models where effects are modified by several variables. Risk factors such as sex, body mass index (BMI) and haemoglobin (Hgb) concentrations are highly correlated with the need to transfuse RBC, as well as with mortality [13]. Therefore, structural equation modelling (SEM) which can analyse pathways between multiple observed variables and the outcome, could address a gap in knowledge regarding transfusion and mortality [14, 15].

SEM has commonly been used in social sciences and psychology where unobserved latent variables can be investigated. Although SEM is an old statistical method, it has gained popularity in recent years because of progress made in software and modelling development [15–17]. To the best of our knowledge, the SEM framework has not previously been employed to investigate cardiac surgery-related mortality, perhaps due to the considerable familiarity with Cox proportional hazards modelling for survival analysis.

The aim of the present study was to investigate the association of RBC transfusion and preoperative anaemia with 5-year mortality among patients who undergo cardiac surgery, while allowing for intervariable correlation. We compared the associated risk of RBC transfusion and preoperative anaemia on 5-year mortality in patients who underwent on-pump open-heart surgery, in SEM models depicting potential paths among known risk factors.

MATERIALS AND METHODS

The Trondheim Heart Surgery database has consecutively registered cardiac surgery patients at St. Olavs University Hospital, Trondheim, Norway, since 1992 as part of local quality assurance work. The database has prospectively registered laboratory results, patient- and procedure-related preoperative characteristics, perioperative events and variables as earlier described [8]. Data regarding mortality and cause of death until 31 December 2018 were obtained from the national Norwegian Cause of Death Registry, which has >99% coverage regarding all Norwegian citizens.

The study was part of the Cardiac Surgery Outcome Study (CaSOS), which has used the database to investigate different complications following adult cardiac surgery and was approved by the Norwegian Data Inspectorate and the Regional Committee for Medical and Health Research Ethics in Middle-Norway (27 June 2007, reference 4.2007.1528). Patients included in the database from April 2008 have provided informed consent, and the need for informed consent was waived up to this date.

Patient population and endpoint

The operative procedure and transfusion thresholds are detailed in the Supplementary Material. All adult on-pump open-heart surgery patients operated on from 2000 through 2017 (n = 10 288) were prospectively included in the database. For patients with multiple entries, only the primary entry was used. Exclusion criteria were short-term mortality, i.e. operative mortality, in-hospital mortality or mortality within 30 postoperative days. Non-Norwegian citizens were excluded from the study because death of foreign citizens is not recorded in the Norwegian Cause of Death Registry. Salvage and emergency procedures have a higher risk of mortality and RBC transfusion, so these patients were excluded. Data were incomplete or missing in 13 patients, leaving data from 9315 patients for analysis (Fig. 1).

Figure 1:

Inclusion and exclusion of patients to the study.

Preoperative anaemia was defined according to the World Health Organization criteria, i.e. Hgb concentration below 12.0 g/dl for women and 13.0 g/dl for men, based on a blood sample drawn shortly before surgery. The study endpoint was 5-year mortality defined as all-cause mortality between 30 days and 5 years postoperatively. The exposure was transfusion of at least 1 unit of RBC intraoperatively or postoperatively. Additionally, the study evaluated the potential effect of RBC transfusion on mortality in shorter intervals, i.e. between 30 days and 1 year postoperatively, and between 1 and 5 years postoperatively.

Study approach

We constructed pathway diagrams in the SEM framework hypothesizing causal relations between risk factors, preoperative anaemia, RBC transfusion and 5-year mortality based on literature, clinical knowledge and available data [15]. We then determined whether the hypothesized relationships fit with the observed data using a series of generally accepted fit indices as detailed in the Supplementary Material [14].

SEM allows for simultaneous estimation of multiple linear regression terms describing the relationships among observed variables, and allows for estimation of complex models [14, 15]. Compared with Cox regression analysis, which only estimates a single time-to-event process, SEM can explore intervariable relationships among multiple patient and operative risk factors both with the need to transfuse RBC and mortality. Another advantage of SEM is the measurement of the indirect effects through intermediary variables, thus estimating both the direct and indirect effect of an independent variable on the dependent variable [16]. However, the model needs to be specified a priori based on empirical data and more than 1 hypothesized model may fit the observed data [14, 17].

The main objective of the study was to compare the direct effect of RBC transfusion and the total effect of preoperative anaemia on 5-year mortality, also with inclusion of other variables that may influence the results. To compare the variables measured using different scales, e.g. Hgb (g/dl) and RBC transfusion (units), the regression coefficients were standardized. This ensured that all variables were measured in standard deviations (SD) and their effects are therefore comparable.

Main structural equation models

We employed a stepwise model-building approach, starting with a simplified model A which included preoperative anaemia, RBC transfusion and 5-year mortality (Fig. 2, model A). In the second step, we further included operative blood loss (100 ml increments) (Fig. 2, model B).

Figure 2:

Pathway diagrams for model A and B. Path diagram notation: boxes indicate observed variables; arrows indicate paths, i.e. the influence of variables on each other; numbers indicate regression coefficients, i.e. the level of correlation along the path. All coefficients are standardized, i.e. numbers corresponded to effects when all variables are measured on the same scale. ^aAnaemia defined as preoperative haemoglobin concentration <12.0 g/dl for women and <13.0 g/dl for men

In the third step, we included risk factors associated with morbidity and mortality in cardiac surgery (model C, Fig. 3). The additional variables were based on clinical knowledge and previous models in CaSOS [8,9]: age, sex, BMI, diabetes, chronic pulmonary disease (use of bronchodilators or forced expiratory volume <75%) and preoperative creatinine level. We hypothesized direct effects of each risk factor on all variables in model B. The model also included history of smoking, cardiopulmonary bypass (CPB) duration (min) and operation category as defined in EuroSCORE II [isolated coronary artery bypass grafting (CABG), non-CABG, 2 surgeries or ≥3 surgeries]. Smoking was only hypothesized to be associated with anaemia and 5-year mortality. CPB duration and EuroSCORE II operation category were assumed to be associated with operative blood loss, RBC transfusion and 5-year mortality (Fig. 3). Left ventricular ejection fraction was measured using different methods over the years and was therefore not included.

Figure 3:

Pathway diagrams for model C. Upper panel: detailed part of model C, focusing on associations of preoperative patient risk factors. Excluded model variables from this panel are CPB time, EuroSCORE II operation category and history of smoking. Lower panel: the pathway diagram represents the full model C. Patient risk factors detailed in the upper panel are visualized as a composite variable. Path diagram notation: boxes indicate observed variables; arrows indicate paths, i.e. the influence of variables on each other; numbers indicate regression coefficients, i.e. the level of correlation along the path. All coefficients are standardized, i.e. numbers corresponded to effects when all variables are measured on the same scale. ^aAge, sex, body mass index, diabetes, pulmonary disease and preoperative creatinine level. ^bDirect effect of preoperative anaemia on RBC transfusion and 5-year mortality. ^cDirect effect of risk factors on preoperative anaemia. ^dDirect effect of risk factors on operative blood loss. ^eDirect effect of risk factors on RBC transfusion. ^fDirect effect of risk factors on 5-year mortality. RBC: red blood cell.

Sensitivity analysis

In the fourth and fifth steps, we analysed the risk associated with RBC transfusion and preoperative anaemia on mortality from 30 days to 1 year postoperatively (model D) and from 1 to 5 years postoperatively (model E), including the same pathways as in model C.

In the sixth step, we analysed the risk associated with RBC transfusion and preoperative anaemia on 5-year mortality in isolated CABG (model F) or multiple procedures (model G). Pathways were as in model C but excluding the variable EuroSCORE II operation category. Finally, in the seventh step, we analysed the risk associated with transfusion of at least 2 units of RBC (model H, RBC-transfused n = 3507) using the same pathways as in model C.

Statistical analysis

All statistical analyses were performed using Stata (Release 16, StataCorp, College Station, TX, USA). Comparisons between transfused and non-transfused patients were performed with the χ² or the t-test. Normality of continuous variables was graphically assessed. Multicollinearity was assessed using variance influence factors. Data are presented as mean and SD or number of patients with percentage. P-values <0.05 were considered significant. We calculated the unstandardized and standardized direct, indirect and total (combined direct and indirect effect) of study variables in all models. Sample size considerations and goodness-of-fit indices are explained in the Supplementary Material.

RESULTS

Patient characteristics

Nine thousand three hundred and fifteen patients were eligible for the study (Fig. 1). Among them, 4214 (45.2%) received at least 1 unit of RBC transfusion. There were 1889 patients (20.3%) who had a Hgb concentration below the threshold for anaemia and RBC transfusion was given to 1552 (82.2%) of these patients. Baseline characteristics, operative variables and postoperative complications are shown in Table 1. Several clinical variables and postoperative events differed between the groups, postoperative renal failure being the most frequent and occurring in 704 (16.7%) of RBC-transfused patients compared to 222 (4.4%) of the non-transfused patients.

Table 1:

Baseline patient characteristics, operative variables and postoperative complications

	Patients without RBC transfusion (n = 5101)	Patients with RBC transfusion (n = 4214)	P-value
Preoperative characteristics
Anaemia	337 (6.6)	1552 (36.8)	<0.001
Male sex	4449 (87.2)	2504 (65.8)	<0.001
Age (years)	64 ± 10.1	70 ± 10.0	<0.001
Body mass index (kg/m2)	27.6 ± 4.0	26.2 ± 4.1	<0.001
Present or previous smoker	2943 (57.7)	2138 (50.7)	<0.001
Congestive heart failurea	500 (12.8)	934 (27.0)	<0.001
NYHA functional classificationa
I	314 (6.2)	238 (5.7)	0.30
II	1721 (33.7)	1033 (24.5)	<0.001
III	2676 (52.5)	2401 (57.0)	<0.001
IV	389 (7.6)	541 (12.8)	<0.001
Previous myocardial infarction	2223 (43.6)	1843 (43.7)	0.88
Previous percutaneous coronary intervention	631 (12.4)	417 (9.9)	<0.001
Diabetes	683 (13.4)	703 (16.7)	<0.001
Hypertension	2645 (51.9)	2412 (57.2)	<0.001
Chronic pulmonary disease	613 (12.0)	824 (19.6)	<0.001
Renal dysfunctionb	82 (1.6)	275 (6.5)	<0.001
Preoperative lab values
Haemoglobin (g/dl)	14.5 ± 1.2	13.0 ± 1.5	<0.001
Creatinine (µmol/l)	87.0 ± 24.9	96.1 ± 68.9	<0.001
Operative and postoperative variables
EuroSCORE II operation category
Isolated coronary artery bypass grafting	3948 (77.4)	2186 (51.9)	<0.001
Single non-coronary artery bypass grafting	543 (10.6)	686 (16.2)	<0.001
2 procedures	556 (10.9)	1163 (27.6)	<0.001
3 or more procedures	54 (1.1)	179 (4.3)	<0.001
Urgent surgeryc	1958 (38.4)	2058 (48.8)	<0.001
Cardiopulmonary bypass time (min)	74 ± 29.7	96 ± 42.7	<0.001
Haemoglobin first postoperative day (g/dl)	10.6 ± 1.1	9.5 ± 0.9	<0.001
Blood loss during surgery (ml)	560 ± 180	730 ± 720	<0.001
Postoperative mediastinal blood loss first 16 h (ml)	537 ± 231	799 ± 693	<0.001
Postoperative complications
Renal failured	222 (4.4)	704 (16.7)	<0.001
Myocardial infarction	150 (2.9)	258 (6.1)	<0.001
Cardiac dysfunctione	151 (3.0)	495 (11.8)	<0.001
Primary intubation more than 24 h or reintubation	24 (0.5)	246 (5.8)	<0.001
Intensive care unit stay >24 h	140 (2.7)	665 (15.8)	<0.001
Death >30 days postoperatively	305 (6.0)	619 (14.7)	<0.001

Variables given as mean with standard deviation or n (%).

^aPatients with missing data: congestive heart failure n = 1939; NYHA functional classification n = 2.

^bCreatinine >140 µmol/l or dialysis.

^cSurgery within 2 weeks.

^dAbsolute increase of creatinine >26 µmol/l, relative increase of >50% or postoperative renal dialysis.

^eUse of >2 inotropics or intra-aortic balloon pump.

NYHA: New York Heart Association; RBC: red blood cell.

In the follow-up period, 619 patients (14.7%) who received RBC transfusion died, compared to 305 patients (6.0%) in the non-transfused group. The mean observation time until death was 2.5 years (SD 1.6 years) for patients who received RBC transfusion and 2.9 years (SD 1.4 years) for patients who did not receive any transfusion (P < 0.001). Between 30 days and 1 year postoperatively, 149 deaths (3.5%) were registered in the transfused group and 39 deaths (0.8%) in the non-transfused group (P < 0.001).

Main structural equation models

All main models had acceptable fit, as described in the Supplementary Material. There were no indications of multicollinearity. Results for all models, as well as the correlation matrix are given in the Supplementary Material.

In the first analysis step (model A, Fig. 2), the direct effects of preoperative anaemia and RBC transfusion on 5-year mortality were 0.14 and 0.09, respectively. These numbers represent the amount of variance in mortality explained by these independent variables, i.e. 14% and 9%. The total effect of anaemia on 5-year mortality when combining the direct and indirect effects was 0.17. Model A indicated that preoperative anaemia was a greater determinant of 5-year mortality than RBC transfusion (i.e. total combined direct and indirect effect: 0.17 vs 0.09).

The results from the step 2 model also including operative blood loss (model B, Fig. 2) were comparable to those of the first step (model A). Total effects for model B were: anaemia 0.17, RBC transfusion 0.08 and operative blood loss 0.08. All path coefficients in model A and B were significant.

In the third step, all study variables were included (model C). Details regarding patient risk factors with direct effects are shown in the upper panel of Fig. 3, and the complete model is shown in the lower panel. Age had the largest total effect on 5-year mortality among cardiac surgery patients (0.15), followed by creatinine level (0.12), pulmonary disease (0.11), anaemia (0.10), diabetes (0.06) and operative blood loss (0.06). BMI was a negative predictor of 5-year mortality, meaning that a higher BMI was associated with increased survival (−0.06). RBC transfusion (0.03) had the weakest total effects on mortality. The coefficients for direct and total effects of all study variables on anaemia, RBC transfusion and 5-year mortality in model C are summarized in Table 2.

Table 2:

Direct and total effects of study variables on anaemia, RBC transfusion and 5-year mortality in model C

Variables	Direct effect coefficientsa	Total effect coefficientsa
Effects on anaemia
Age	0.15*	0.15*
Female sex	0.04*	0.04*
Body mass index	−0.13*	−0.13*
Diabetes	0.12*	0.12*
Pulmonary disease	0.06*	0.06*
Creatinine level	0.18*	0.18*
Smoker	0.01	0.01
Effects on RBC transfusion
Preoperative anaemia	0.29*	0.29*
Operative blood loss	0.09*	0.09*
Age	0.13*	0.17*
Female sex	0.28*	0.30*
Body mass index	−0.12*	−0.15*
Diabetes	0.03*	0.06*
Pulmonary disease	0.03*	0.04*
Creatinine level	0.06*	0.11*
CPB time	0.20*	0.23*
EuroSCORE II operation category	0.04*	0.03*
Smoker	Not applicable	0.003
Effects on 5-year mortality
RBC transfusion	0.03*	0.03*
Preoperative anaemia	0.09*	0.10*
Operative blood loss	0.06*	0.06*
Age	0.13*	0.15*
Female sex	−0.02	−0.01
Body mass index	−0.04*	−0.06*
Diabetes	0.05*	0.06*
Pulmonary disease	0.10*	0.11*
Creatinine level	0.10*	0.12*
CPB time	0.01	0.04*
EuroSCORE II operation category	0.05*	0.04*
Smoker	0.04*	0.04*

^aAll coefficients are standardized, i.e. numbers correspond to effects when all variables are measured in the same scale.

CPB: cardiopulmonary bypass; RBC: red blood cell.

^* P-value <0.05.

Sensitivity analysis

The results from the sensitivity analyses, i.e. fourth to seventh steps of the analysis, were similar for anaemia and RBC transfusion as in the main analysis (model C). Details are given in the Supplementary Material. In model D, model E and model G, RBC transfusion was not significantly associated with mortality (P = 0.10, P = 0.061 and P = 0.44, respectively).

The results for age, BMI, diabetes, pulmonary disease and preoperative creatinine level were similar in model D–H as in model C. Female sex was not significantly associated with mortality in any model.

DISCUSSION

The main finding of the study suggests that when appropriately adjusted for correlations among variables in a SEM model, the effect on 5-year mortality of RBC transfusion became weaker than that of other patient risk factors. In the follow-up period, age was the most important risk factor for 5-year mortality. The results from our models showed that preoperative anaemia, pulmonary disease and increased creatinine levels had equivalent effects in all observation periods. Of the significant variables, RBC transfusion had the smallest effect. Several variables were highly correlated to RBC transfusion, of which preoperative anaemia had the strongest correlation. The results from the present study expand on previously published studies because of the alternative analytical approach.

SEM is a statistical method which has not commonly been used to investigate mortality in this patient group. Cox proportional hazards modelling, which is widely used, allows for adjustment of multiple variables in a regression analysis of a time-related outcome such as mortality. But an inherent problem when using Cox regression to investigate the hazard associated with RBC transfusion is the strong correlation of patient risk factors such as age, sex, BMI and preoperative anaemia with the need to transfuse and with mortality [10, 13]. Cox regression assumes that the independent variables are uncorrelated but tolerates low-to-moderate correlation. If correlation is high as in the present setting, the coefficients and P-values may become imprecise. Therefore, a major advantage of using SEM is that it allows for more complex models based on hypothesized causal relationships and the measurement of direct, indirect and total effects on dependent variables [14, 17].

The findings of an association between RBC transfusion and 5-year mortality in cardiac surgery patients point in the same direction as previous studies [1–4], but the effect was smaller in the present study. Some studies have not found an association and attributed the difference in observed long-term mortality to patient risk factors [5–9]. Preoperative anaemia had 3 times the total effect on long-term mortality compared to RBC transfusion in the present study, both being significant. Although preoperative anaemia is sometimes unavoidable, it may also be a modifiable risk factor. The present study shows that preoperative anaemia may be considered a more harmful risk factor compared to transfusion and efforts should be directed at finding better ways to manage preoperative anaemia. The impact of preoperative anaemia is consistent with previous studies indicating that it is an independent risk factor among cardiac surgery patients [11, 12]. The results suggest that patient risk factors that are both associated with risk of transfusion and mortality, such as anaemia, are the main cause of the difference in long-term mortality seen between transfused and non-transfused patients in cardiac surgery. It is crucial that physicians continue striving to minimize unnecessary transfusion. But an important clinical implication of the present study is that RBC transfusion should not be withheld from patients who could benefit from it, because of a perceived strong harmful effect with respect to mortality.

In previous studies from CaSOS, adjustments were included for more risk factors and no significant association between RBC transfusion and long-term mortality was found [8, 9]. They were omitted in the present study because SEM models are specified a priori based on hypothesized causality and too complex models render acceptable model fit difficult to attain. Randomized controlled trials comparing adverse effects of liberal versus restrictive transfusion thresholds have not shown differences in outcomes between the groups, except for total number of RBC units transfused [18, 19]. These findings indirectly support the notions that there are no long-term effects of RBC transfusion itself and that differences in mortality may be due to patient risk factors. Therefore, the significant association of RBC transfusion with 5-year mortality in the present study may be due to residual confounding.

In our models, higher BMI had a positive effect on survival in cardiac surgery, although it is commonly associated with cardiovascular disease and mortality [20]. The term ‘obesity paradox’ is commonly used to denote this phenomenon, which has been observed previously [21]. Possible explanations include selection bias (obese patients with fewer risk factors are admitted to surgery), an inverse age effect (obesity is more common in younger patients) or less haemodilution among males and those with higher BMI and consequently less transfusion [13, 22].

Female sex is considered a risk factor for mortality following cardiac surgery [13] but was not significantly associated with 5-year mortality in our models. The sex effect could be related to different risk profiles in men and women undergoing cardiac surgery [23]. A previous study in CaSOS explored sex as a risk factor for survival following cardiac surgery and did not find a significant association with long-term mortality [24]. The roles of sex and BMI on RBC transfusion and 5-year mortality were not investigated in the present study.

Several postoperative complications were significantly more common in the RBC-transfused group. Whether these differences could be entirely attributed to transfusion or were also associated with patient-related risk factors, and the impact they have on long-term mortality, was not fully explored. A previous study in CaSOS suggested that postoperative complications and not RBC transfusion, are significantly associated with long-term mortality in isolated CABG patients [8]. SEM could be employed to investigate this relationship between transfusion, postoperative complications and long-term mortality.

Limitations

Important challenges in SEM are correct model specification, omitted or missing variables and small sample sizes. A strength of the present study is the large number of patients in the CaSOS database which allowed for estimation with multiple independent and dependent variables. The results differ from previous studies because SEM allows for analysis of associations among correlated variables, such as risk factors and transfusion. The findings of the present study are only accurate if the causal relationships presented are true. Other hypothesized models between variables may yield different results, and we cannot exclude that inclusion of further variables could have improved the models [17].

A limitation of SEM is that mortality is analysed as a categorical variable (yes/no) without considering the time to death as in Cox regression. Previous studies in CaSOS found no significant association of RBC transfusion with long-term mortality using Cox regression [8, 9]. Results from the sensitivity analysis showed that RBC transfusion was not significant when investigated in the separate periods from 30 days to 1 year or 1 to 5 years postoperatively, even if it was significant in the main analysis without such subdivision. This may be a statistical power issue arising because the total number of deaths was lower in each separate period. Despite the adjustments of risk factors in the SEM models and the relatively long follow-up time, residual confounding due to missing adjustment variables cannot be excluded.

The patient data in the CaSOS database were prospectively collected and have undergone quality assurance controls. There were few missing data, and therefore the database is considered representative of the general population in Middle Norway. The study was limited to a single centre and local transfusion policy may differ from other institutions.

The transfusion policy has undergone minor changes during the study period. The intraoperative transfusion threshold has remained the same, but the postoperative transfusion threshold has varied between 8 and 9 g/dl. The final decision to transfuse was left to the attending physician and this introduces a potential bias. The present study did not evaluate the dose effect, intraoperative or postoperative timing of RBC transfusion or storage time of RBC units.

CONCLUSIONS

Preoperative anaemia was statistically significant and had a greater impact on 5-year mortality compared to RBC transfusion for all observation periods. The results from the SEM models indicated that the risk associated with RBC transfusion is mainly due to patient risk factors.

SUPPLEMENTARY MATERIAL

Supplementary material is available at ICVTS online.

Abbreviations

BMI

Body mass index

CABG

Coronary artery bypass grafting

CaSOS

Cardiac Surgery Outcome Study

CPB

Cardiopulmonary bypass

Hgb

Haemoglobin

RBC

Red blood cell

SD

Standard deviation

SEM

Structural equation modelling

 

Presented at the 34th Annual Meeting of the European Association for Cardio-Thoracic Surgery, Barcelona, Spain, 8–10 October 2020.

Funding

This work was supported by the Faculty of Medicine and Health Sciences at NTNU [15/6075 MKD].

Conflict of interest: none declared.

Author contributions

Long Tran: Conceptualization; Data curation; Formal analysis; Investigation; Validation; Visualization; Writing—original draft. Guri Greiff: Conceptualization; Data curation; Writing—review & editing. Alexander Wahba: Conceptualization; Writing—review & editing. Hilde Pleym: Conceptualization; Data curation; Writing—review & editing. Vibeke Videm: Conceptualization; Formal analysis; Funding acquisition; Investigation; Project administration; Supervision; Validation; Writing—review & editing. 

Reviewer information

Interactive CardioVascular and Thoracic Surgery thanks Mahmoud Salama Al-Shiekh, Ko Bando and the other, anonymous reviewer(s) for their contribution to the peer review process of this article.