Intraoperative blood transfusion of one or two units of packed red blood cells is associated with a fivefold risk of stroke in patients undergoing elective carotid endarterectomy

Abstract

Objective

Transfused blood can disrupt the coagulation cascade. We postulated that packed red blood cell (PRBC) transfusion may be associated with thromboembolic phenomena. We used propensity matching to examine the relationship between intraoperative PRBC transfusion and stroke during carotid endarterectomy (CEA).

Methods

We selected CEA procedures from the American College of Surgeons National Surgical Quality Improvement Program database from 2005–2009. We excluded bilateral, redo, and emergent procedures. We used multivariate logistic regression to identify independent risk factors for stroke. We then calculated a transfusion propensity score to match patients who received one or two units of transfused PRBC intraoperatively with patients of similar risk profiles who had not been transfused.

Results

Our criteria resulted in 12,786 elective CEA patients. Of these, 82 (0.6%) received a one- to two-unit intra-operative transfusion. Thirty-day stroke rates were 1.4% (179/12,704) in the nontransfused group and 6.1% (5/82) in the transfused group (Fisher exact test, P = .007). In forward stepwise multivariable regression of risk factors, only hemiplegia, stroke history, and transient ischemic attacks were predictive of 30-day stroke. We used these same variables to calculate transfusion propensity. We matched 80 transfused patients with 160 controls, thus, creating two groups with very similar risk profiles differing only by their transfusion status. In the matched groups, there was a fivefold increase in the risk of stroke in transfused patients (Fisher exact test, P = .043

Conclusions

Intraoperative transfusion of one to two units of PRBCs is associated with a fivefold increase in stroke risk. This holds true after consideration of stroke risk variables and operative duration as a surrogate for technical difficulty. The increased risk may be related to several effects of transfused blood on the coagulation inflammation cascade.

Stroke, the fourth leading cause of death in the United States behind coronary artery disease and cancer, accounts for approximately one of every 18 deaths occurring annually in the United States. Each year, roughly 795,000 people experience a new or recurrent stroke; on average, 610,000 are shown to be first attacks whereas 185,000 are recurrent.¹

According to the National Institute of Neurological Disorders and Stroke, 2 to 3 million Americans who have survived a stroke are shown to have sustained some form of permanent disability; furthermore, stroke is estimated to cost the United States roughly $73.7 billion per year (2010). Carotid atherosclerosis is a risk factor for embolic stroke. Carotid endarterectomy (CEA), the “gold standard” in the treatment of symptomatic and asymptomatic carotid artery stenosis,2 rarely requires perioperative transfusion. Perioperative blood transfusion has been previously linked with numerous hazards including increased infection rates, malignancies (non-Hodgkin’s lymphoma), transfusion-related acute lung injury, and thrombotic events in various surgical settings,3 however, no data currently exist linking packed red blood cell (PRBC) transfusion to poorer neurologic outcome in CEA. Traditionally, previous patient transfusion experiences have governed the amount of PRBC that is to be typed and cross-matched for certain surgical procedures, with an average of about two units for CEAs.4 Although studies of representative institutions indicate that PRBC is still being ordered excessively nationwide, the amount of PRBC transfused in the perioperative period has decreased significantly because of increased awareness of transfusion-associated risks and increasing cost. In light of (1) stroke is a periprocedural risk following CEA with significant social and financial burden and (2) intraoperative PRBC transfusion, which has been previously linked with hypercoagulable and embolic events, is still practiced within CEA, analysis is warranted looking at the direct relationship between intraoperative PRBC transfusion and perioperative stroke following CEA.

METHODS

We selected CEA procedures from the American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP) dataset from the years 2005 to 2009 (primary procedure Current Procedural Terminology code 35301 and operative diagnosis International Classification of Diseases, 9th Revision code 43x.x or 44x.x). We excluded bilateral and reoperations, as well as patients with any secondary procedures other than radiographic or other tests (Doppler scan, etc). We also excluded emergent cases, patients without ACS/NSQIP documentation of timely discontinuance of blood thinners, patients with known preoperative transfusion, intraoperative transfusion of three or more units, or postoperative transfusion. We also excluded patients with preoperative impaired sensorium, coma, or totally dependent functional status and/or a central nervous system tumor. We also excluded patients without preoperative hematocrit values.

The ACS NSQIP dataset contains data from a prospective, systematic study of patients undergoing major surgical procedures at over 200 hospitals. In participating hospitals, a data coordinator collected over 50 preoperative patient characteristics, intraoperative processes of care, and 22 uniformly defined postoperative adverse occurrences (including death and stroke with deficit) up to 30 days after the operation. Risk and outcome variables were rigorously defined, and coordinators had previously completed in-depth training on all study definitions. Thirty-day follow-up information was obtained through phone calls, letters, and review of clinic/hospital medical records. Additional operations within 30 days of an included case were excluded. Patients less than 18 years old and admissions for trauma were excluded.

Our primary outcome was stroke with neurologic deficit within 30 days of CEA. We compared stroke rates in the transfused and nontransfused populations using Fisher exact test. We performed forward stepwise logistic regression of all the over 50 ACS NSQIP patient and perioperative risk factors vs postoperative stroke with deficit within 30 days of operation. To these independent predictors, we added sex, age, operative duration, and preoperative hematocrit to the final regression model. We used independent risk factors identified for stroke in the first regression in a second logistic regression of whether the patient received a transfusion or not. The resulting estimated probability of transfusion was used as a transfusion propensity score. We then used greedy, nearest neighbor matching (essentially the closest match found) on the logit of the propensity score (within a maximum range of 0.2 SDs) to match transfused patients (one to two units only) with two nontransfused controls (2:1 match) of similar stroke risk. We measured the effectiveness of our propensity matching by calculating the standardized difference between variables in the two groups. We set the threshold for significant differences as greater than 0.1.

RESULTS

Our strict inclusion criteria resulted in 12,786 elective, initial, isolated CEA procedures. Of these, only 82 (0.6%) received a one- to two-unit intraoperative transfusion. The transfusions were almost evenly split, 44 patients received one unit (stroke 3/44, 6.8%, mortality 0/44) and 38 patients received two units (stroke 2/38 5.3%, mortality 2/38 5.3%). Looking at the two groups combined (82 patients), 30-day stroke rates were 1.4% (179/12,704) in the nontransfused group and 6.1% (5/82) in the transfused group (Fisher exact test, P = .007). Thirty-day mortality was 0.6% (70/12,704) in the non-transfused group and 2.4% (2/82) in the transfused group (Fisher exact test, P = .078).

In forward stepwise multivariable regression of the over 50 clinical patient risk factors, only hemiplegia, stroke history (with and without deficit), and transient ischemic attacks were predictive of 30-day stroke. Age, American Society of Anesthesiology class, functional status, serum albumin, and numerous others were not independent predictors of stroke in this patient population. We added transfusion status, sex, age, operative duration, and hematocrit into the regression model along with the stroke history variables, and the results are shown in Table I. With multivariable adjustment, the OR for postoperative stroke was 4.75 (95% CI, 1.75–12.86; P = .002) in transfused vs nontransfused patients.

Table I.

Multivariate ORs for 30-day stroke with deficit after initial elective CEA (n = 12,786)

Variable	Incidence n or mean (%) ± SD	OR (95% CI)	Significance
Transfused 1–2 units vs none	82 (0.6)	4.75 (1.75–12.86)	0.002
Preop hemiplegia/hemiparesis	554 (4.3)	2.35 (1.37–4.04)	0.002
Hx transient ischemic attacks	3204 (25.1)	1.43 (1.05–1.95)	0.023
Hx CVA/stroke w/neurologic deficit	1587 (12.4)	2.00 (1.30–3.08)	0.002
Hx CVA/stroke w/o neurologic deficit	1028 (8.0)	2.22 (1.44–3.43)	0.000
Female sex	5317 (41.6)	1.13 (0.83–1.53)	0.436
Age, years	71.1 ± 9.4	1.00 (0.99–1.02)	0.804
Operative duration vs ≤80 minutes	2674 (20.9)		0.969
81–99 minutes	2448 (19.1)	0.86 (0.54–1.38)	0.537
100–119 minutes	2598 (20.3)	0.90 (0.57–1.42)	0.661
120–147 minutes	2586 (20.2)	0.88 (0.56–1.40)	0.600
>147 minutes	2480 (19.4)	0.96 (0.61–1.51)	0.873
Preop HCT group vs >40%	6248 (48.9)		0.831
≤25	41 (0.3)	1.21 (0.16–9.26)	0.852
26–30	306 (2.4)	0.73 (0.28–1.93)	0.525
31–35	1646 (12.9)	0.89 (0.56–1.41)	0.622
36–40	4545 (35.5)	0.83 (0.59–1.16)	0.283

CEA, Carotid endarterectomy; CI, confidence interval; CVA, cerebrovascular accident; HCT, hematocrit; OR, odds ratio; SD, standard deviation.

We used these same risk variables to calculate the propensity score and were able to closely match 80 transfused patients with 160 controls. The standardized differences between risk variables are less than 0.1 for all variables indicating homogeneous matching on known risk factors (Table II). The relative risk for stroke was 5.0 in the transfused group vs the matched controls (Fisher exact test, P = .043; Table III). This is similar to the adjusted OR from the multivariate regression.

Table II.

Balance of risk factors between propensity-matched cohorts of patients undergoing CEA; transfused one or two units vs nontransfused controls (n = 240)

Risk variable	Nontransfused, No. (%) or mean ± SD	Transfused (1–2 units), No. (%) or mean ± SD	Standardized difference
No. of patients in cohort	160	80
Age, mean years ± SD	75.1 ± 8.1	74.3 ± 8.7	−0.01
Male	83 (51.9)	43 (53.8)	0.04
Hx hemiplegia, No. (%)	4 (2.5)	2 (2.5)	0
Hx TIA, No. (%)	49 (30.6)	25 (31.3)	0.01
Hx CVA w/neurologic deficit, No. (%)	15 (9.4)	8 (10.0)	0.02
Hx CVA w/o neurologic deficit, No. (%)	14 (8.8)	7 (8.8)	0
Preop hematocrit group
>40.0, No. (%)	18 (11.0)	9 (11.0)	0
35.1–40.0, No. (%)	34 (20.7)	17 (20.7)	0
30.1–35.0, No. (%)	44 (26.8)	22 (26.8)	0
25.1–30.0, No. (%)	56 (35.0)	28 (35.0)	0
≤25, No. (%)	8 (4.9)	4 (4.9)	0
Operative duration quintile
≤80 minutes, No. (%)	28 (17.5)	11 (13.8)	−0.1
81–99 minutes, No. (%)	14 (8.8)	9 (11.3)	0.08
100–119 minutes, No. (%)	27 (16.9)	13 (16.3)	−0.02
120–147 minutes, No. (%)	20 (12.5)	10 (12.5)	0
>147 minutes, No. (%)	71 (44.4)	37 (46.3)	0.04

CEA, Carotid endarterectomy; CVA, cerebrovascular accident; SD, standard deviation; TIA, transient ischemic attack.

Table III.

Relative risk for 30-day mortality and stroke in propensity-matched cohorts of patients undergoing CEA; transfused one or two units vs nontransfused controls (n = 240)

	Nontransfused		Transfused 1–2 units
Outcome	No.	%	No.	%	Relative risk (95% CI)	Exact P value
Cohort size	160		80
Mortality	2	1.3%	2	2.5%	2.0 (0.3–13.9)	.602
Stroke	2	1.3%	5	6.3%	5.0 (1.0–25.2)	.043
Stroke or mortalitya	4	2.5%	7	8.8%	3.5 (1.1–11.6)	.045

CEA, Carotid endarterectomy; CI, confidence interval.

^aNo patient was recorded with both 30-day stroke and death, although this is possible in the National Surgical Quality Improvement Program protocol.

DISCUSSION

Blood transfusion is an uncommon event during carotid endarterectomies. We found increased risk of postoperative stroke in patients receiving one or two units of transfused PRBC during initial, elective, and isolated carotid endarterectomies. An almost fivefold risk for stroke-related to one- to two-unit transfusion was determined in both multivariate logistic regression and propensity-matched analysis. We chose our analysis in the one- to two-unit cohort because these quantities are potentially unrelated to acute anemia and may represent more discretionary transfusion after adjusting for preoperative hematocrit.

Ischemic strokes can be associated with blood transfusion in a variety of mechanisms. Not only viscosity changes that result from the increased hematocrit, but multiple biochemical and molecular mechanisms may contribute to this phenomenon. The main objective of PRBC transfusion is to increase tissue oxygenation, thereby reducing tissue oxygen debt. However, most transfusions given to critical patients consist of PRBCs stored >20 days, and the efficacy of the stored blood for increasing oxygenation has been questioned, as stored PRBCs have low levels of 2,3-diphosphoglycerate and adenosine triphosphate and low deformability.⁵ Storage of RBCs results in a series of morphologic and biochemical changes, collectively referred to as RBC storage lesion⁶ that hinder their function after transfusion. During the first 2 weeks of storage, most 2,3-diphosphoglycerate is lost from the RBC. This causes an increase in the oxygen affinity of hemoglobin leading to a diminished capacity of transfused erythrocytes to release oxygen into the tissues. Another characteristic alteration in stored RBCs is a change from their normal biconcave shape to an echinocyte. This results in lower surface-to-volume ratio, spherocytosis, increased cell hemoglobin concentration and viscosity, increased osmotic fragility, and loss of deformability. Red cells clump together in storage.⁷ The longer the blood bags are stored, the higher number of red cell clumps are present and the larger the numbers of red cells in each of these clumps. From some of the newest microcirculatory work in transfusion, it has been shown that banked blood does not increase oxygen delivery to tissues. Indeed, it may be responsible for up to a 400% decrease in tissue oxygen delivery.⁸ Work in patients after coronary artery bypass surgery grafting (CABG) surgery showed that there was no increase in oxygen delivery to the microcirculation with one or two units of blood.⁹ In some critically ill patients, it has been shown that transfusing banked blood actually decreases gut oxygen delivery making the tissues more acidotic.¹⁰ Therefore, the notion that one will increase oxygen delivery to tissues with transfusion has been shown to be not true in randomized trials.¹¹

In a large (n = 16,000), recently published analysis on coronary bypass surgery,¹² transfusion of more than four units of PRBC was the strongest (OR = 5) independent predictor with respect to perioperative stroke; it is not clear from this analysis whether transfusion support was a causative factor vs a predictor. Another recent publication demonstrated a strong relationship between perioperative platelet transfusion and both stroke and death after coronary artery bypass surgery grafting.¹³

This literature as well as our findings supports an association between ischemic events and PRBC transfusion. Blood transfusion is associated with a greater number of ischemic events in patients undergoing percutaneous coronary interventions for acute coronary syndromes (ACS). Rao and colleagues assessed 24,111 patients with ACS in a meta-analysis of three large trials investigating the relation between transfusion and outcomes. The rates of 30-day rates of death (8% vs 3.08%; P < .001), myocardial infarction (MI) (25.16% vs 8.16%; P < .001), and composite death/MI (29.24% vs 10.02%; P < .001) were all significantly higher for patients receiving transfusions, with a significant interaction occurring at transfusion for a nadir hematocrit value >25%.¹¹ In non-ST elevation ACS, regardless of the hemoglobin concentration, transfusion was associated with an increased risk of the composite end point (adjusted OR, 1.54; 95% CI, 1.14–2.09). The reason for the differential effect of transfusion on patients with ST-elevation MI and non-ST elevation ACS may be related to a variety of factors, including local and systemic hemodynamics, and further study will be required to definitively elucidate the appropriate use of blood product transfusion in those with anemia and acute hemorrhage. Finally, resting myocardial oxygen extraction is high (75%) and can be augmented only via increased coronary blood flow. A major contributor to increasing coronary blood flow is nitric oxide–mediated vasodilatation, which has been proven to increase oxygen delivery to the myocardium via the large coronary arteries. Nitric oxide serves a role in oxygen exchange, has a short half-life, and has been shown to be decreased in stored blood.¹⁴ Therefore, stored red blood cells devoid of nitric oxide may act as a vacuum, depleting local levels of nitric oxide within coronary arteries and causing localized vasoconstriction and tissue hypoxia.

Retrospective analysis provided by the Mayo Clinic (Rochester, Minn), involving the evaluation of red blood cell (RBC) transfusion practices over the past 15 years and the influence of changes on neurologic or cardiac morbidity after CEA, indicated that, in general, a lower transfusion trigger was developed within older patients showing increased comorbidity.¹⁵ The frequency of perioperative stroke and MI between two stratified groups of patients (an early-practice group of patients admitted from 1980 to 1985 and a recent-practice group admitted from 1990 to 1995) was not observed to have been significantly different. Moreover, along with a decrease in RBCs during all phases of the perioperative period, timing as to administration of transfusion shifted from intraoperative to postoperative in later years. Modest perioperative anemia was not found to increase the risk of perioperative cerebral and cardiac ischemia in patients during CEA.

Our analysis was limited by the low incidence of transfusion (0.6%) and stroke (1.4%) with CEA. Any analysis of stroke and transfusion will necessarily suffer from these limitations. Our systematically obtained sample of over 12,000 initial, isolated, carotid endarterectomies in the ACS NSQIP certainly is close to the largest study of its kind. It has many more patients than conceivable in any randomized clinical trial of transfusion, which is unlikely to be performed. We do not have information regarding the use of statins and antiplatelet agents, as the ACS NSQIP does not capture them. We do not have data regarding the postoperative hematocrit or hemoglobin and therefore cannot exclude over transfusion in some of these cases. For statistical reasons, we were unable to separate the patients who received one unit from those who received two units of PRBC. The dataset does not have intraoperative blood pressures or hypotensive events. We do not have information about the use of shunt or patch during the procedure. We adjusted for operative duration as a placeholder for technical complexity as well as past stroke history and preoperative hematocrit. It is conceivable that technical difficulty may be related to intraoperative blood transfusion. The dataset lacks information regarding the reason for the transfusion; however, our findings as well as the association between hypotension and stroke, may suggest that these patients be better served by maintaining normotension rather than blood transfusion. We also do not have information regarding the age of the PRBCs and whether or not leukoreduction was performed.

CONCLUSIONS

In summary, we used CEA as a model to examine the effect of PRBC transfusion on neurologic outcome. Our findings indicate that erythrocyte transfusion is associated with an increased risk for stroke, possibly because of the hypercoagulable and inflammatory reactions that they provoke.

The American College of Surgeons National Surgical Quality Improvement Program and the hospitals participating in the ACS NSQIP are the source of the data used herein; they have not been verified and are not responsible for the statistical validity of the data analysis or the conclusions derived by the authors.