current as of November 9, 2010. Online article and related content 

  http://jama.ama-assn.org/cgi/content/full/304/14/1568

 . 2010;304(14):1568-1575 (doi:10.1001/jama.2010.1406) JAMA

 Elliott Bennett-Guerrero; Yue Zhao; Sean M. O'Brien; et al.  

Artery Bypass Graft SurgeryVariation in Use of Blood Transfusion in Coronary

Supplementary material http://jama.ama-assn.org/cgi/content/full/304/14/1568/DC1

eTable

Correction Contact me if this article is corrected.

Citations Contact me when this article is cited.

This article has been cited 1 time.

Topic collections

Contact me when new articles are published in these topic areas.Hematology, Other; Cardiovascular Intervention Surgical Physiology; Blood/ Coagulation; Hematology/ Hematologic Malignancies;System; Surgery; Surgical Interventions; Cardiovascular/ Cardiothoracic Surgery; Revascularization; Medical Practice; Medical Practice, Other; Cardiovascular

the same issueRelated Articles published in

. 2010;304(14):1610.JAMAAryeh S. Shander et al. Blood Transfusion as a Quality Indicator in Cardiac Surgery 

. 2010;304(14):1559.JAMALudhmila A. Hajjar et al. Controlled TrialTransfusion Requirements After Cardiac Surgery: The TRACS Randomized

http://pubs.ama-assn.org/misc/permissions.dtlpermissions@ama-assn.orgPermissions 

http://jama.com/subscribeSubscribe

reprints@ama-assn.orgReprints/E-prints 

http://jamaarchives.com/alertsEmail Alerts

at Duke University on November 9, 2010 www.jama.comDownloaded from

ORIGINAL CONTRIBUTION

Variation in Use of Blood Transfusionin Coronary Artery Bypass Graft SurgeryElliott Bennett-Guerrero, MDYue Zhao, PhDSean M. O’Brien, PhDT. B. Ferguson Jr, MDEric D. Peterson, MD, MPHJames S. Gammie, MDHoward K. Song, MD, PhD

PATIENTS WHO UNDERGO CAR-diac surgery receive a signifi-cant proportion of the 14 mil-lion units of allogeneic red

blood cells (RBCs) transfused annu-ally in the United States.1 Numerous ob-servational studies in patients who un-derwent cardiac surgery have shown anassociation between RBC transfusionand adverse outcome, including mor-bidity, mortality, resource utilization,and quality of life.2-9 To date, no largerandomized trials of transfusion thresh-olds have been conducted in cardiacsurgery to our knowledge to addressthis issue.

Almost 20 years ago, the study byGoodnough et al10 demonstrated thatthere was significant practice variabil-ity in transfusion practices at 18 US cen-ters. However, this study and subse-quent studies11-14 were limited in sizeand did not adjust for hospital or pa-tient factors. Since these earlier stud-ies, the Society of Thoracic Surgeons(STS) and Society of Cardiovascular An-esthesiologists published transfusionrecommendations in 2007.15 How-ever, the degree to which guidelineshave resulted in consensus in commu-nity transfusion practice is unknown.Therefore, the primary goal of our study

was to assess use of RBC, fresh-frozenplasma, and platelet transfusions incoronary artery bypass graft (CABG)surgery in contemporary practice. Ouranalyses specifically addressed the de-gree to which transfusion practices var-ied among US hospitals, after adjust-ing for patient characteristics.

METHODSData Source

The STS Adult Cardiac Surgery Data-base (ACSD) was established in 1989 toreport outcomes following cardiotho-racic surgical procedures.16-20 The data-base captures clinical information fromthe majority of US cardiac surgical pro-cedures. A recent analysis demon-

strated that more than 80% of patientsundergoing CABG operations in theUnited States in 2007 were representedin the STS database.21 Sites enter pa-tient data using uniform definitions(available at http://www.sts.org) and cer-

See also pp 1559 and 1610.

Author Affiliations: Divisions of Perioperative Clini-cal Research (Dr Bennett-Guerrero), Biostatistics (DrsZhao and O’Brien), and Cardiology (Dr Peterson), DukeClinical Research Institute, Duke University MedicalCenter, Durham, North Carolina; Department of Car-diovascular Sciences, East Carolina Heart Institute,Greenville, North Carolina (Dr Ferguson); Division ofCardiac Surgery, University of Maryland Medical Cen-ter, Baltimore (Dr Gammie); and Division of Cardio-thoracic Surgery, Oregon Health and Science Univer-sity, Portland (Dr Song).Corresponding Author: Elliott Bennett-Guerrero, MD,Division of Perioperative Clinical Research, Duke Clini-cal Research Institute, Duke University Medical Cen-ter, PO Box 3094, Durham, NC 27710 (elliott.bennettguerrero@duke.edu).

Context Perioperative blood transfusions are costly and have safety concerns. As aresult, there have been multiple initiatives to reduce transfusion use. However, thedegree to which perioperative transfusion rates vary among hospitals is unknown.

Objective To assess hospital-level variation in use of allogeneic red blood cell (RBC),fresh-frozen plasma, and platelet transfusions in patients undergoing coronary arterybypass graft (CABG) surgery.

Design, Setting, and Patients An observational cohort of 102 470 patients un-dergoing primary isolated CABG surgery with cardiopulmonary bypass during calen-dar year 2008 at 798 sites in the United States, contributing data to the Society ofThoracic Surgeons Adult Cardiac Surgery Database.

Main Outcome Measures Perioperative (intraoperative and postoperative) trans-fusion of RBCs, fresh-frozen plasma, and platelets.

Results At hospitals performing at least 100 on-pump CABG operations (82 446 casesat 408 sites), the rates of blood transfusion ranged from 7.8% to 92.8% for RBCs,0% to 97.5% for fresh-frozen plasma, and 0.4% to 90.4% for platelets. Multivari-able analysis including data from all 798 sites (102 470 cases) revealed that after ad-justment for patient-level risk factors, hospital transfusion rates varied by geographiclocation (P=.007), academic status (P=.03), and hospital volume (P� .001). How-ever, these 3 hospital characteristics combined only explained 11.1% of the variationin hospital risk-adjusted RBC usage. Case mix explained 20.1% of the variation be-tween hospitals in RBC usage.

Conclusion Wide variability occurred in the rates of transfusion of RBCs and otherblood products, independent of case mix, among patients undergoing CABG surgerywith cardiopulmonary bypass in US hospitals in an adult cardiac surgical database.JAMA. 2010;304(14):1568-1575 www.jama.com

1568 JAMA, October 13, 2010—Vol 304, No. 14 (Reprinted) ©2010 American Medical Association. All rights reserved.

at Duke University on November 9, 2010 www.jama.comDownloaded from

tified software systems. This informa-tion is sent semiannually to the STS DataWarehouse and Analysis Center at theDuke Clinical Research Institute,Durham, North Carolina. A series of dataquality checks are performed before asite’s data are aggregated into the na-tional sample. Although participation inthe STS database is voluntary, data com-pleteness is high, with overall preopera-tive risk factors missing in fewer than 5%of submitted cases.22

Because the data used in analyses ofthe STS ACSD represent a limited dataset (no direct patient identifiers) thatwas originally collected for nonre-search purposes, and the investigatorsdo not know the identity of individualpatients, the analysis of these data wasdeclared by the Duke University HealthSystem Institutional Review Board to beresearch not involving human sub-jects and is therefore considered ex-empt (Duke University Health SystemProtocol 00005876).

Patient Population

Hospital variation in the frequency ofblood product administration was ana-lyzed in a contemporary sample of iso-lated primary CABG operations usingcardiopulmonary bypass (CPB) per-formed at hospitals participating in theSTS ACSD between January 1, 2008,and December 31, 2008. The time framefor this analysis was chosen to repre-sent the most contemporary data avail-able and to minimize the effect of po-tential changes over time. Hospitals(n=798) contributing at least 1 adultcardiac case per month during 2008were included. Unless stated other-wise, all analyses included all 798 sites(102 470 cases). To increase the ho-mogeneity of the study population, weonly included patients undergoing pri-mary cardiac surgery and excluded pa-tients who previously underwent me-dian sternotomy. Additional exclusioncriteria included (1) combination ofCABG surgery with valve or other ma-jor surgical interventions; (2) off-pump CABG surgery; (3) age youngerthan 18 years; (4) emergent status, elec-tive and urgent status were allowed; (5)

preoperative cardiogenic shock or needfor cardiopulmonary resuscitationwithin 1 hour before surgery; and (6)presence of infective endocarditis pre-operatively. In addition, we excluded122 patients with incomplete data forperioperative blood usage.

Variable Definitions

Blood and Blood Products. The STS da-tabase collects the number of units ofpacked RBCs, platelets, or fresh-frozen plasma administered to the pa-tient intraoperatively and postopera-tively during hospitalization. Themeasurement of packed RBCs does notinclude preoperatively donated or in-traoperatively cell-savaged autolo-gous blood. All blood and blood prod-uct values reported herein reflect thesum of each of the products adminis-tered intraoperatively and postopera-tively.

AcademicvsNonacademicHospitals.Academic status was defined as hospi-tals that have residency programs, ac-cording to the STS participant infor-mation database.

Geographic Region. Hospitals weregrouped into 9 regions using catego-ries defined by the US Census Bureauas follows: New England (Maine, Mas-sachusetts, Vermont, New Hamp-shire, Rhode Island, and Connecti-cut), Mid-Atlantic (New Jersey, NewYork, and Pennsylvania), South Atlan-tic (Delaware, District of Columbia,West Virginia, Virginia, Maryland,North Carolina, South Carolina, Geor-gia, and Florida), Great Lakes (Illi-nois, Indiana, Michigan, Ohio, and Wis-consin), Pacific (Alaska, Hawaii,Oregon, Washington, and California),Mountain (Arizona, Colorado, Idaho,Montana, Nevada, New Mexico, Utah,and Wyoming), Plains (North Da-kota, South Dakota, Nebraska, Kan-sas, Minnesota, Iowa, and Missouri),West South Central (Oklahoma, Loui-siana, Arkansas, and Texas), and EastSouth Central (Mississippi, Alabama,Tennessee, and Kentucky).

Hospital CABG Surgery Volumes in2008. The annual hospital volume forprimary isolated CABG surgery dur-

ing 2008 was categorized into 4 groups(quartiles), with an approximately equalnumber of patients in each group. Thecategories were quartile 1 (�115 cases),quartile 2 (115-183 cases), quartile 3(184-299 cases), and quartile 4 (�300cases).

Statistical Analyses. Baseline char-acteristics were summarized as per-centage or median (interquartilerange) as appropriate and comparedfor patients receiving vs not receivingany RBCs in the intraoperative orpostoperative period. To quantifybetween-hospital variation in bloodusage, we calculated the percentage ofpatients undergoing primary isolatedCABG surgery at each hospital whoreceived any RBCs, any fresh-frozenplasma, and any platelets in the intra-operative or postoperative period. Todisplay the results graphically, we plot-ted hospital-specific percentages ofpatients receiving blood productsagainst hospital-specific numbers ofeligible cases. We superimposed linesrepresenting 99.9% binomial predic-tion limits23 (FIGURE 1). The binomialprediction limits indicate the rangeof results that would normally occuras a result of random statistical varia-tion for a hospital whose true fre-quency of using blood products isequal to the mean for all hospitals.

Additional analyses focused on de-termining the amount of hospital-level variation in blood product usagethat is due to true signal variation, asopposed to random statistical varia-tion (ie, noise), and exploring factorsthat might explain the signal varia-tion. Separate analyses were per-formed for RBCs, fresh-frozen plasma,and platelets. A series of 2-level hier-archical logistic regression models withhospital-specific random interceptswere fit to the patient-level data. In eachmodel, the end point was a patient-level binary variable coded as 1 if thepatient received the blood product andas 0 otherwise. Model 1 contained onlyhospital-specific random intercepts andno covariates. This model was used forestimating the distribution of true hos-pital usage rates of blood products af-

USE OF BLOOD TRANSFUSION IN CABG SURGERY

©2010 American Medical Association. All rights reserved. (Reprinted) JAMA, October 13, 2010—Vol 304, No. 14 1569

at Duke University on November 9, 2010 www.jama.comDownloaded from

ter subtracting out the effect of ran-dom sampling variation. Results weresummarized in tabular form by pre-

senting selected percentiles of the hos-pital distribution. These percentileswere calculated from the estimated

mean and variance of the random ef-fects parameters (assumed to be nor-mally distributed on the log-odds scale)and were transformed from the scale oflog-odds to the scale of probabilities tofacilitate interpretation. The same es-timates were used to construct a his-togram depicting the estimated distri-bution of true hospital-specific usagerates (FIGURE 2).

Model 2 contained hospital-specificrandom intercepts plus patient-levelcovariates, which included age, femalesex, race, left ventricular ejection frac-tion, body surface area, serum creati-nine, date of surgery (dichotomized intofirst vs second 6-month period), pro-cedure, need for dialysis, atrial fibril-lation, hypertension, immunosuppres-sive treatment, percutaneous coronaryintervention less than 6 hours beforesurgery, presence of intra-aortic bal-loon pump or administration of ino-tropes, peripheral vascular disease,unstable angina (no myocardial infarc-tion �7 days), left main disease, aorticstenosis, aortic insufficiency, mitralinsufficiency, tricuspid insufficiency,chronic lung disease, cerebrovasculardisease or cerebrovascular accident, dia-betes, number of diseased coronary ves-sels, myocardial infarction, acuity sta-tus, congestive heart failure, New YorkHeart Association class, preoperativehematocrit, and use of medications,including warfarin, aspirin, adenosinediphosphate inhibitor within 5 days, orglycoprotein IIb/IIIa inhibitor.24 Model2 was used for quantifying between-hospital signal variation after subtract-ing out differences due to patient-levelrisk factors. A hospital’s risk-adjustedRBC usage rate was calculated as1/{1�exp[−(���j)]}, where the con-stant � was chosen to reflect the base-line probability of receiving RBCs foran “average” patient and �j denotes thej-th hospital’s random intercept param-eter. Percentiles were based on the esti-mated mean and variance of the �j’s.This method was also used for fresh-frozen plasma and platelets.

Model 3 contained the factors inmodel 2, plus 3 hospital-specific fac-tors (academic status, region, and vol-

Figure 1. Observed Variation in Hospital-Specific Transfusion Rates for Primary IsolatedCABG Surgery With Cardiopulmonary Bypass During 2008 (N=798 Sites)

100

80

40

60

20

0

0 600500400300200100No. of Isolated Primary CABG Operations

Red blood cells

Tran

sfus

ion

Rat

e, %

100

80

40

60

20

0

0 600500400300200100No. of Isolated Primary CABG Operations

Fresh-frozen plasma

Tran

sfus

ion

Rat

e, %

100

80

40

60

20

0

0 600500400300200100No. of Isolated Primary CABG Operations

Platelets

Tran

sfus

ion

Rat

e, %

CABG indicates coronary artery bypass graft. Each solid circle represents a unique hospital, with the observedtransfusion rate percentages for that hospital (red blood cells, fresh-frozen plasma, and platelets) plotted againstthe hospital’s 2008 volume of isolated primary CABG operations. The solid line indicates the overall mean trans-fusion rate across all hospitals. The dashed lines indicate the upper and lower 99.9% prediction limits based onthe binomial distribution.

USE OF BLOOD TRANSFUSION IN CABG SURGERY

1570 JAMA, October 13, 2010—Vol 304, No. 14 (Reprinted) ©2010 American Medical Association. All rights reserved.

at Duke University on November 9, 2010 www.jama.comDownloaded from

ume). This model was used to explorethe effect of each hospital-level factorwhile adjusting for differences in pa-tient case mix. The effect of each hos-pital-level covariate was summarized byreporting odds ratios (ORs) with 95%confidence intervals (CIs). The abilityof hospital and patient factors to ex-plain between-hospital variation intransfusion rates was examined. Toquantify between-hospital variation, thepredicted log odds from model 3 wasaveraged within each hospital and de-composed as the sum of 3 compo-nents (namely, the contributions of pa-tient factors, hospital covariates, andhospital random effects). The percent-age of between-hospital variation ex-plained by hospital covariates was cal-culated as the squared Pearsoncorrelation between the hospital fac-tor component of the average log-odds and the sum of all 3 compo-nents. An analogous calculation wasused to quantify the percentage of varia-tion explained by patient factors (ie,case mix).

Hierarchical logistic regression withrandom intercepts was used to assessthe association between the percent-age of patients receiving RBCs at a hos-pital and the patient-level end point ofall-cause mortality. All-cause mortal-ity was defined as death during the samehospitalization as surgery or after dis-charge but within 30 days of surgery.To minimize misclassification error inthe hospital-specific transfusion rates,only hospitals with at least 100 on-pump isolated CABG operations wereincluded in these analyses. Hospitalswere assigned to 4 groups according tothe percentage of patients receiving RBCtransfusion intraoperatively or postop-eratively (7.8%-43.2%, 43.3%-55.9%,56.0%-65.6%, and 65.7%-92.8%). Ratesof mortality were compared across cat-egories of hospital transfusion rates withand without adjustment for patient fac-tors (model 2 covariates).

Parameters of the various models wereestimated using a penalized quasi-likelihood approximation as imple-mented in SAS version 9.2 PROCGLIMMIX (SAS Institute, Cary, North

Carolina). R statistical package version2.9.0 (R Foundation for Statistical Com-puting, Vienna, Austria) was also used.P�.05 was considered significant. Alltests were 2-sided and were not ad-justed for multiple comparisons.

RESULTSAmong the 102 592 cases of primaryisolated CABG surgery with CPB sub-mitted from 798 hospitals in 2008,102 470 cases (99.9%) had complete in-formation about perioperative RBC,platelet, and fresh-frozen plasma trans-fusions. The rates of perioperative trans-fusion were 56.1% (95% CI, 55.8%-56.4%) for packed RBCs, 19.3% (95%CI, 19.1%-19.6%) for fresh-frozenplasma, and 24.7% (95% CI, 24.5%-25.0%) for platelets. Patients receiv-ing RBC transfusion (n=57 445) weremore likely to be women, were older,had received adenosine diphosphate in-hibitors, had lower preoperative he-matocrit, and exhibited other tradi-tional risk factors for morbidity andmortality compared with those pa-tients who did not receive RBC trans-fusions (TABLE 1).

Between-Hospital Variationin Blood Usage

There was dramatic variability in the ob-served hospital-specific transfusionrates for all 3 blood products in 102 470patients undergoing isolated primaryCABG surgery at 798 hospitals(Figure 1). To ensure that between-center differences would not be domi-nated by random statistical variation,we also analyzed the subset of hospi-tals performing at least 100 eligible on-pump CABG operations during theyear. At these 408 sites (n = 82 446cases), the frequency of blood transfu-sion rates ranged from 7.8% to 92.8%for RBCs, 0% to 97.5% for fresh-frozen plasma, and 0.4% to 90.4% forplatelets.

The estimated distribution of trans-fusion rates based on hierarchical mod-eling is shown in TABLE 2. Accordingto this model, hospitals at the 99th per-centile of the distribution were 4.6 timesmore likely to use RBCs (90.6%/

19.7%=4.6), 31.2 times more likely touse fresh-frozen plasma (71.7%/2.3%=31.2), and 22.5 times more likelyto use platelets (76.4%/3.4%=22.5)compared with hospitals at the 1st per-centile of the distribution. This widevariation was not explained by patientrisk factors. Comparing the 1st and 99thpercentiles, hospitals at the 99th per-centile of the distribution were 7.7 timesmore likely to use RBCs (94.5%/12.2%), 34.8 times more likely to usefresh-frozen plasma (73.1%/2.1%), and

Figure 2. Estimated Distribution ofHospital-Specific Transfusion Rates forPrimary Isolated CABG Surgery With CPBDuring 2008 (N=102 470)

0 20 80 10040 60

0.020

0.025

0.030

0.035

0.015

0.010

0.005

0

Transfusion Rate, %

Red blood cells

Den

sity

0 20 80 10040 60

0.020

0.025

0.030

0.035

0.015

0.010

0.005

0

Transfusion Rate, %

Fresh-frozen plasma

Den

sity

0 20 80 10040 60

0.020

0.025

0.030

0.035

0.015

0.010

0.005

0

Transfusion Rate, %

Platelets

Den

sity

CABG indicates coronary artery bypass graft; CPB, car-diopulmonary bypass. The “true” hospital-specificblood usage rate and its variability among hospitalsare estimated by hierarchical models (random effectmodels). X-axes are blood product transfusion rates(percentages) divided into intervals of 5%. Y-axes rep-resent distribution density. The area of each bar rep-resents the estimated proportion of hospitals fallinginto the indicated range of transfusion rates.

USE OF BLOOD TRANSFUSION IN CABG SURGERY

©2010 American Medical Association. All rights reserved. (Reprinted) JAMA, October 13, 2010—Vol 304, No. 14 1571

at Duke University on November 9, 2010 www.jama.comDownloaded from

24.3 times more likely to use platelets(77.6%/3.2%), even after adjusting forpatient risk factors (Table 2).

Hospital Characteristicsand Blood Usage

TABLE 3 shows unadjusted and ad-justed associations between hospitalcharacteristics and blood usage. The fre-quency of perioperative RBC usage de-creased across categories of increasingCABG surgery volume from 61.4%(95% CI, 59.4%-63.4%) in quartile 1 to

51.6% (95% CI, 48.0%-55.2%) in quar-tile 4 (P� .001). The adjusted ORs forRBC usage were inversely related to vol-ume and were statistically significant forquartile 2 (OR, 0.71; 95% CI, 0.59-0.86), quartile 3 (OR, 0.61; 95% CI,0.49-0.76), and quartile 4 (OR, 0.51;95% CI, 0.38-0.66) compared withquartile 1.

There was also substantial geo-graphic variation in RBC usage. In theunadjusted analysis, blood usage wassignificantly lower in all 8 regions com-

pared with the West South Central re-gion. After adjusting for patient risk fac-tors, blood usage was found to besignificantly lower in 7 of the 8 geo-graphic regions (OR of these 7 regionsranged between 0.45 and 0.67). Bloodusage was more than 2-fold lower in theMountain (OR, 0.45; 95% CI, 0.31-0.64) and New England (OR, 0.46; 95%CI, 0.29-0.72) regions compared withthe West South Central region.

A significant association (OR, 1.32;95% CI, 1.04-1.69; P= .03) was ob-served between academic hospital sta-tus and perioperative RBC usage, afteradjusting for patient-level risk factors.However, these 3 hospital characteris-tics combined only explained 11.1% ofthe variation in hospital risk-adjustedRBC usage. Case mix explained 20.1%of the variation between hospitals inRBC usage.

Hospital-Specific RBC TransfusionRates and All-Cause Mortality

In both unadjusted and adjusted analy-ses, there was no significant associa-tion between hospital-specific RBCtransfusion rates and all-cause mortal-ity (eTable, available at http://www.jama.com).

COMMENTOur large observational study showsthat there is enormous variability in therates of transfusion of RBCs, fresh-frozen plasma, and platelets in pa-tients undergoing isolated primaryCABG surgery across a large number ofUS hospitals, even after adjusting for pa-tient- and hospital-level risk factors.Our analysis of blood transfusion prac-tices represents patients who have un-dergone surgery at 798 US hospitals. Be-cause the STS database includes themajority of US patients who under-went cardiac surgery,21 our findings pre-sent a comprehensive picture of trans-fusion practices in patients undergoingCABG surgery.

Almost 20 years ago, the study byGoodnough et al10 showed significantvariability in transfusion practice in 540patients who underwent cardiac sur-gery across 18 institutions and drew at-

Table 1. Demographic Characteristics and Outcomes of Patients Who Underwent PrimaryIsolated CABG Surgery With or Without Perioperative RBC Transfusion During 2008

Characteristics

Transfusion, No. (%) of Patients

Without RBC(n = 45 025)

With RBC(n = 57 445)

Age, median (IQR), y 62 (55-69) 67 (60-75)

Men 39 294 (87.3) 36 013 (62.7)

Academic hospital 3825 (8.5) 6432 (11.2)

Demographic regionWest South Central 3226 (7.2) 5049 (8.8)

South Atlantic 10 808 (24.0) 12 964 (22.6)

Plains 3718 (8.3) 4365 (7.6)

Pacific 5300 (11.8) 6262 (10.9)

New England 1636 (3.6) 2266 (3.9)

Mountain 2306 (5.1) 2257 (3.9)

Mid-Atlantic 4806 (10.7) 6559 (11.4)

Great Lakes 8993 (20.0) 12 362 (21.5)

East South Central 4232 (9.4) 5361 (9.3)

Weight, median (IQR), kg 92 (81-104) 82 (71-95)

BMI, median (IQR) 29.8 (26.7-33.6) 28.1 (25.0-32.1)

Diabetes 16 171 (35.9) 24 939 (43.4)

Hypertension 36 665 (81.4) 49 634 (86.4)

Current or recent smoker 14 384 (32.0) 14 775 (25.7)

Peripheral vascular disease 4677 (10.4) 9709 (16.9)

ADP inhibitor (thienopyridine) 4235 (9.4) 8668 (15.1)

Glycoprotein IIb/IIIa inhibitor 1822 (4.1) 2690 (4.7)

Preoperative hematocrit, median (IQR), % 41 (39-44) 37 (34-41)

Preoperative creatinine, median (IQR), mg/dL 1.0 (0.9-1.1) 1.0 (0.9-1.3)

Dialysis for renal failure 376 (0.84) 1966 (3.42)

History of myocardial infarction 18 382 (40.8) 26 835 (46.7)

Ejection fraction, median (IQR), % 55 (45-60) 55 (43-60)

Duration, median (IQR), minCardiopulmonary bypass 85 (66-108) 93 (72-117)

Cross clamp 60 (45-79) 64 (48-85)

No. of distal bypass grafts, median (IQR) 3.0 (3.0-4.0) 3.0 (3.0-4.0)

Reoperation for bleeding 66 (0.2) 2318 (4.0)

Postoperative length of stay, median (IQR), d 5.0 (4.0-6.0) 6.0 (5.0-8.0)

Readmission within 30 d 3150 (7.0) 6342 (11.0)

All-cause mortalitya 196 (0.4) 1330 (2.3)Abbreviations: ADP, adenosine diphosphate; BMI, body mass index, calculated as weight in kilograms divided by height

in meters squared; CABG, coronary artery bypass graft; IQR, interquartile range; RBC, red blood cell.SI conversion: To convert creatinine to µmol/L, multiply by 88.4.aDefined as death during the same hospitalization as surgery or after discharge but within 30 days of surgery.

USE OF BLOOD TRANSFUSION IN CABG SURGERY

1572 JAMA, October 13, 2010—Vol 304, No. 14 (Reprinted) ©2010 American Medical Association. All rights reserved.

at Duke University on November 9, 2010 www.jama.comDownloaded from

tention to this problem. Several subse-quent studies provided additional dataon this topic, but these are no longercontemporary, had no or limited risk-adjustment, and were limited insize.11,13,14 Despite nearly 2 decades ofawareness of inconsistent transfusionpractices and the publication of clini-cal practice guidelines, there has beenno improvement in disparate transfu-sion practices. For example, Good-nough et al10 found that the transfu-sion rates for RBCs, fresh-frozen plasma,

and platelets ranged from 17% to 100%,0% to 90%, and 0% to 80%, respec-tively. In our analysis, transfusion rateswere similar. This variability cannot beattributed to inclusion of hospitals withsmall denominators. Indeed, in hospi-tals reporting at least 100 eligible on-pump CABG operations (82 446 casesat 408 sites), transfusion rates amongpatients undergoing primary isolatedon-pump CABG surgery still rangedfrom 7.8% to 92.8% for RBCs, 0% to97.5% for fresh-frozen plasma, and

0.4% to 90.4% for platelets. Moreover,the variation persisted after adjust-ment for a large number of patient andhospital factors.

We found that patients at academichospitals and those in the lowest quar-tile of volume were more likely to re-ceive RBC transfusion compared withother hospitals. We also observed varia-tion in RBC usage based on geo-graphic region. These differences areunexplained and warrant further study.Of note, these 3 hospital characteris-

Table 2. Estimated Percentiles of the Distribution of Hospital-Specific True Transfusion Rates in Patients Who Underwent Primary IsolatedCABG Surgery During 2008a

Model

Probability of Transfusion in Percentile, %

1st 10th 25th 50th 75th 90th 99th

Red blood cellsUnadjusted 19.7 36.0 47.5 60.7 72.4 81.0 90.6

Adjusted 12.2 29.1 43.5 60.7 75.6 85.3 94.5

Fresh-frozen plasmaUnadjusted 2.3 6.3 11.0 19.6 32.5 47.0 71.7

Adjusted 2.1 6.1 10.8 19.6 33.0 48.0 73.1

PlateletsUnadjusted 3.4 8.9 15.0 25.3 39.5 54.0 76.4

Adjusted 3.2 8.6 14.7 25.3 39.9 55.0 77.6Abbreviation: CABG, coronary artery bypass graft.aPercentile was derived from estimated random effect distribution and adjusted models included hospital-specific random intercepts and patient-level covariates, as described in

the “Methods” section.

Table 3. Association Between Perioperative RBC Transfusion and Hospital Characteristics in Patients Who Underwent Primary Isolated CABGSurgery During 2008

OutcomeNo. of

PatientsTransfused, %

(95% CI)

Unadjusted Adjusteda

OR (95% CI) P Value OR (95% CI) P Value

Region 102 470 .04 .007

West South Central 8275 61.0 (55.6-66.5) 1 [Reference] 1 [Reference]

East South Central 9593 55.9 (51.1-60.7) 0.65 (0.48-0.87) .003 0.79 (0.54-1.15) .21

Great Lakes 21 355 57.9 (55.4-60.3) 0.69 (0.55-0.87) .001 0.67 (0.51-0.90) .007

Mid-Atlantic 11 365 57.7 (53.7-61.8) 0.66 (0.51-0.85) .001 0.53 (0.38-0.74) �.001

Mountain 4563 49.5 (44.8-54.1) 0.48 (0.36-0.64) �.001 0.45 (0.31-0.64) �.001

New England 3902 58.1 (50.4-65.8) 0.62 (0.43-0.89) .01 0.46 (0.29-0.72) �.001

Pacific 11 562 54.2 (49.5-58.8) 0.74 (0.58-0.94) .02 0.55 (0.40-0.75) �.001

Plains 8083 54.0 (49.3-58.7) 0.56 (0.43-0.73) �.001 0.54 (0.39-0.76) �.001

South Atlantic 23 772 54.5 (51.3-57.8) 0.63 (0.50-0.80) �.001 0.60 (0.45-0.81) �.001

Hospital volume, quartileb 102 470 �.001 �.001

1 (n = 441) 25 530 61.4 (59.4-63.4) 1 [Reference] 1 [Reference]

2 (n = 177) 25 509 56.9 (54.4-59.3) 0.74 (0.64-0.85) �.001 0.71 (0.59-0.86) .001

3 (n = 116) 25 783 54.4 (51.7-57.1) 0.65 (0.56-0.77) �.001 0.61 (0.49-0.76) �.001

4 (n = 64) 25 648 51.6 (48.0-55.2) 0.58 (0.47-0.71) �.001 0.51 (0.38-0.66) �.001

Academic status 102 470 .008 .03

Nonacademic 92 213 55.3 (53.8-56.8) 1 [Reference] 1 [Reference]

Academic hospital 10 257 62.7 (59.0-66.4) 1.29 (1.07-1.56) .008 1.32 (1.04-1.69) .03Abbreviations: CABG, coronary artery bypass graft; CI, confidence interval; OR, odds ratio; RBC, red blood cell.aAdjusted models include all 3 classes of hospital characteristics variables (region, hospital volume, and academic status) and patient-level predictors as described in the “Methods” section.bThe annual hospital volume for primary isolated CABG surgery during 2008 was categorized into 4 groups (quartiles) with an approximately equal number of patients in each group as

described in the “Methods” section (quartile 1=lowest volume, quartile 4=highest volume).

USE OF BLOOD TRANSFUSION IN CABG SURGERY

©2010 American Medical Association. All rights reserved. (Reprinted) JAMA, October 13, 2010—Vol 304, No. 14 1573

at Duke University on November 9, 2010 www.jama.comDownloaded from

tics combined only explained 11.1% ofthe variation in hospital risk-adjustedRBC usage.

Our study has several limitations.First, data on RBC, platelet, and fresh-frozen plasma transfusions have not un-dergone audit; therefore, we cannot beabsolutely sure of the accuracy of datareported by sites. Some of the variabil-ity in observed transfusion rates mightbe due to differences in the accuracywith which programs document us-age. For example, sites may rely on onlyone or a combination of paper or elec-tronic medical record, blood bank rec-ords, or both. However, our analysisonly included patients who had dataavailable on blood product usage at bothtime points (intraoperative and post-operative).

Our study’s primary goal was to as-sess the variability between hospitalswith respect to transfusion; therefore,as such it was not prospectively de-signed to focus on the association be-tween hospital transfusion rate and ad-verse outcome. Nevertheless, ourlimited analysis (eTable) appears to sug-gest that there is no strong associationbetween hospital transfusion rate andmortality. This does not necessarilycontradict the large body of literatureshowing an association between trans-fusion and adverse outcome becausethose studies focused on patient-levelrisk.2-9 Our analysis of mortality fo-cused on comparing groups of hospi-tals according to their hospital-leveltransfusion rates. We specifically didnot compare mortality of individual pa-tients who did vs did not receive trans-fusion. We can state, however, that evenif higher transfusion rates at some hos-pitals are not deleterious they may stillrepresent potentially unnecessary carethat is costly. The acquisition costs ofa unit of RBCs, fresh-frozen plasma, andplatelets (apharesis) were $214, $60,and $539, respectively, in 2006.1 Thesecosts underestimate the true direct andindirect costs of transfusion.25 For ex-ample, a recent analysis estimated thetotal cost of each RBC transfusion torange from $522 to $1183 (mean cost,$761 per RBC unit).26 Therefore, even

if unnecessary transfusion is not del-eterious, a reduction in the observedvariability might result in significantcost savings.

As is the case in other areas of medi-cine, the degree of variability in clini-cal practice we observed represents apotential quality improvement oppor-tunity. This is particularly complex inrelation to transfusion practice in CABGsurgery. The decision to transfuse hasmultiple triggers, resulting from a widearray of clinical scenarios and the con-sequent inability to apply standard-ized algorithms. The multiplicity ofhealth care practitioners in CABG sur-gery care generates differences of opin-ion about safety and efficacy. Transfu-sion thresholds will change during thecourse of care; the threshold for a rap-idly bleeding patient is different thanfor a stable patient postoperatively. Im-provement in quality related to trans-fusion practice in CABG surgery is amultifactorial, complex but criticallyimportant, challenge. Studies have dem-onstrated that use of a blood conser-vation program significantly im-proves transfusion rates over time.27-31

This may be a more effective way of im-proving transfusion rates, as opposedto publishing guidelines, which may notbe that helpful as our study suggests.In addition, the role of lack of data fromrandomized trials cannot be over-stated. To our knowledge, there hasnever been a large randomized trial ofthe safety and efficacy of blood trans-fusion in cardiac surgery15; therefore,some of the variability we observed maybe due to honest differences betweenclinicians in the perceived benefits andrisks of transfusion.

Author Contributions: Dr Bennett-Guerrero had fullaccess to all of the data in the study and takes re-sponsibility for the integrity of the data and the ac-curacy of the data analysis.Study concept and design: Bennett-Guerrero, O’Brien,Ferguson, Gammie, Song.Acquisition of data: Peterson, Gammie, Song.Analysis and interpretation of data: Bennett-Guerrero,Zhao, O’Brien, Gammie, Song.Drafting of the manuscript: Bennett-Guerrero, O’Brien,Gammie, Song.Critical revision of the manuscript for important in-tellectual content: Zhao, O’Brien, Ferguson, Peterson,Gammie, Song.Statistical analysis: Zhao, O’Brien, Song.Obtained funding: Peterson, Song.

Administrative, technical, or material support:Ferguson, Gammie, Song.Study supervision: Bennett-Guerrero, O’Brien,Ferguson, Gammie, Song.Financial Disclosures: Dr Bennett-Guerrero is princi-pal investigator (grant R01 HL101382-01 from the Na-tional Institutes of Health) for a multicenter study as-sessing the impact of blood transfusion on peripheraland cerebral oxygenation and the microcirculation. Heis also a named inventor on a patent application re-lated to methods of washing red blood cells. No otherauthors have any disclosures.Funding/Support: This study was supported by the So-ciety of Thoracic Surgeons (STS) through the Na-tional Adult Cardiac Surgery Database and the DukeClinical Research Institute (DCRI).Role of the Sponsors: This study was sponsored bythe STS. Specifically, the DCRI has a contract with theSTS to be their National Cardiac Data Warehouse andAnalysis Center. In this role, the DCRI independentlyharvests data from each participating STS center, cre-ates a national analysis database, and performs sta-tistical analyses. The proposal for this study was sub-mitted to and approved of by the STS NationalDatabase Publications Committee. After approval, themanuscript was reviewed by the coauthors and a fi-nal version was approved by the publications com-mittee. The STS was involved in the design and con-duct of the study; in the collection, management,analysis, and interpretation of the data; and in the re-view and approval of the manuscript.Disclaimer: Dr Peterson, a contributing editor forJAMA, was not involved in the editorial review of orthe decision to publish this article.Online-Only Material: An eTable is available at http://www.jama.com.

REFERENCES

1. US Department of Health and Human Services. The2007 Nationwide Blood Collection and Utilization Sur-vey Report. Washington, DC: Dept of Health and Hu-man Services; 2007.2. Engoren MC, Habib RH, Zacharias A, Schwann TA,Riordan CJ, Durham SJ. Effect of blood transfusion onlong-term survival after cardiac operation. Ann Tho-rac Surg. 2002;74(4):1180-1186.3. Koch CG, Li L, Duncan AI, et al. Morbidity and mor-tality risk associated with red blood cell and blood-component transfusion in isolated coronary artery bypassgrafting. Crit Care Med. 2006;34(6):1608-1616.4. Koch CG, Li L, Duncan AI, et al. Transfusion in coro-nary artery bypass grafting is associated with re-duced long-term survival. Ann Thorac Surg. 2006;81(5):1650-1657.5. Koch CG, Li L, Sessler DI, et al. Duration of red-cell storage and complications after cardiac surgery.N Engl J Med. 2008;358(12):1229-1239.6. Kuduvalli M, Oo AY, Newall N, et al. Effect of peri-operative red blood cell transfusion on 30-day and1-year mortality following coronary artery bypasssurgery. Eur J Cardiothorac Surg. 2005;27(4):592-598.7. Murphy GJ, Reeves BC, Rogers CA, Rizvi SI, CullifordL, Angelini GD. Increased mortality, postoperative mor-bidity, and cost after red blood cell transfusion in pa-tients having cardiac surgery. Circulation. 2007;116(22):2544-2552.8. Scott BH, Seifert FC, Grimson R. Blood transfu-sion is associated with increased resource utilisation,morbidity and mortality in cardiac surgery. Ann CardAnaesth. 2008;11(1):15-19.9. Surgenor SD, DeFoe GR, Fillinger MP, et al. Intra-operative red blood cell transfusion during coronaryartery bypass graft surgery increases the risk of post-operative low-output heart failure. Circulation. 2006;114(1)(suppl):I43-I48.10. Goodnough LT, Johnston MF, Toy PT; Transfu-

USE OF BLOOD TRANSFUSION IN CABG SURGERY

1574 JAMA, October 13, 2010—Vol 304, No. 14 (Reprinted) ©2010 American Medical Association. All rights reserved.

at Duke University on November 9, 2010 www.jama.comDownloaded from

sion Medicine Academic Award Group. The variabil-ity of transfusion practice in coronary artery bypasssurgery. JAMA. 1991;265(1):86-90.11. Covin R, O’Brien M, Grunwald G, et al. Factorsaffecting transfusion of fresh frozen plasma, plate-lets, and red blood cells during elective coronary ar-tery bypass graft surgery. Arch Pathol Lab Med. 2003;127(4):415-423.12. Maddux FW, Dickinson TA, Rilla D, et al. Insti-tutional variability of intraoperative red blood cell uti-lization in coronary artery bypass graft surgery. Am JMed Qual. 2009;24(5):403-411.13. Snyder-Ramos SA, Möhnle P, Weng YS, et al; In-vestigators of the Multicenter Study of PerioperativeIschemia; MCSPI Research Group. The ongoing vari-ability in blood transfusion practices in cardiac surgery.Transfusion. 2008;48(7):1284-1299.14. Stover EP, Siegel LC, Parks R, et al; Institutionsof the Multicenter Study of Perioperative Ischemia Re-search Group. Variability in transfusion practice forcoronary artery bypass surgery persists despite na-tional consensus guidelines: a 24-institution study.Anesthesiology. 1998;88(2):327-333.15. Ferraris VA, Ferraris SP, Saha SP, et al; Society ofThoracic Surgeons Blood Conservation Guideline TaskForce; Society of Cardiovascular Anesthesiologists Spe-cial Task Force on Blood Transfusion. Perioperativeblood transfusion and blood conservation in cardiacsurgery: the Society of Thoracic Surgeons and the So-ciety of Cardiovascular Anesthesiologists Clinical Prac-tice Guideline. Ann Thorac Surg. 2007;83(5)(suppl):S27-S86.16. Peterson ED, Coombs LP, DeLong ER, Haan CK,Ferguson TB. Procedural volume as a marker of qual-ity for CABG surgery. JAMA. 2004;291(2):195-201.

17. Ferguson TB Jr, Peterson ED, Coombs LP, et al;Society of Thoracic Surgeons and the National Car-diac Database. Use of continuous quality improve-ment to increase use of process measures in patientsundergoing coronary artery bypass graft surgery: a ran-domized controlled trial. JAMA. 2003;290(1):49-56.18. Ferguson TB Jr, Coombs LP, Peterson ED; Soci-ety of Thoracic Surgeons National Adult CardiacSurgery Database. Preoperative beta-blocker useand mortality and morbidity following CABG surgeryin North America. JAMA. 2002;287(17):2221-2227.19. Edwards FH. Evolution of the Society of ThoracicSurgeons National Cardiac Surgery Database. J Inva-sive Cardiol. 1998;10(8):485-488.20. Ferguson TB Jr, Dziuban SW Jr, Edwards FH, et al.The STS National Database: current changes and chal-lenges for the new millennium. Committee to Estab-lish a National Database in Cardiothoracic Surgery, TheSociety of Thoracic Surgeons. Ann Thorac Surg. 2000;69(3):680-691.21. Jacobs JP, Edwards FH, Shahian DM, et al. Suc-cessful linking of the STS Adult Cardiac Surgery Da-tabase to CMS Medicare data to examine the pen-etration, completeness, and representativeness of theSTS database. Paper presented at: 46th Annual Meet-ing of the Society of Thoracic Surgeons (STS); Janu-ary 25, 2008; Fort Lauderdale, FL.22. Welke KF, Ferguson TB Jr, Coombs LP, et al. Va-lidity of the Society of Thoracic Surgeons National AdultCardiac Surgery Database. Ann Thorac Surg. 2004;77(4):1137-1139.23. Spiegelhalter DJ. Funnel plots for comparing in-stitutional performance. Stat Med. 2005;24(8):1185-1202.

24. Shahian DM, O’Brien SM, Filardo G, et al; Soci-ety of Thoracic Surgeons Quality Measurement TaskForce. The Society of Thoracic Surgeons 2008 car-diac surgery risk models, part 1: coronary artery by-pass grafting surgery. Ann Thorac Surg. 2009;88(1)(suppl):S2-S22.25. Shander A, Hofmann A, Gombotz H, TheusingerOM, Spahn DR. Estimating the cost of blood: past,present, and future directions. Best Pract Res ClinAnaesthesiol. 2007;21(2):271-289.26. Shander A, Hofmann A, Ozawa S, TheusingerOM, Gombotz H, Spahn DR. Activity-based costs ofblood transfusions in surgical patients at four hospitals.Transfusion. 2010;50(4):753-765.27. DeAnda A Jr, Baker KM, Roseff SD, et al. Devel-oping a blood conservation program in cardiac surgery.Am J Med Qual. 2006;21(4):230-237.28. Helm RE, Rosengart TK, Gomez M, et al. Com-prehensive multimodality blood conservation: 100 con-secutive CABG operations without transfusion. AnnThorac Surg. 1998;65(1):125-136.29. Rosengart TK, Helm RE, DeBois WJ, Garcia N,Krieger KH, Isom OW. Open heart operations with-out transfusion using a multimodality blood conser-vation strategy in 50 Jehovah’s Witness patients: im-plications for a “bloodless” surgical technique. J AmColl Surg. 1997;184(6):618-629.30. Shibata K, Takamoto S, Kotsuka Y, Sato H. Ef-fectiveness of combined blood conservation mea-sures in thoracic aortic operations with deep hypo-thermic circulatory arrest. Ann Thorac Surg. 2002;73(3):739-743.31. Van der Linden P, De Hert S, Daper A, et al. A stan-dardized multidisciplinary approach reduces the use ofallogeneicbloodproducts inpatientsundergoingcardiacsurgery. Can J Anaesth. 2001;48(9):894-901.

Reality, no matter how widened and heightened ourperceptions, never ceases to be anything but the effecton us of an infinite mystery.

—Laurens van der Post (1906-1996)

USE OF BLOOD TRANSFUSION IN CABG SURGERY

©2010 American Medical Association. All rights reserved. (Reprinted) JAMA, October 13, 2010—Vol 304, No. 14 1575

at Duke University on November 9, 2010 www.jama.comDownloaded from