Original Articles

Factors Affecting Transfusion of Fresh Frozen Plasma, Platelets, and Red Blood Cells During Elective Coronary Artery Bypass Graft Surgery Randal Covin, MD; Maureen O’Brien, PhD; Gary Grunwald, PhD; Bradley Brimhall, MD; Gulshan Sethi, MD; Steven Walczak, PhD; William Reiquam, MD; Chitra Rajagopalan, MD; A. Laurie Shroyer, PhD

● Context.—The ability to predict the use of blood components during surgery will improve the blood bank’s ability to provide efficient service. Objective.—Develop prediction models using preoperative risk factors to assess blood component usage during elective coronary artery bypass graft surgery (CABG). Design.—Eighty-three preoperative, multidimensional risk variables were evaluated for patients undergoing elective CABG-only surgery. Main Outcome Measures.—The study endpoints included transfusion of fresh frozen plasma (FFP), platelets, and red blood cells (RBC). Multivariate logistic regression models were built to assess the predictors related to each of these endpoints. Setting.—Department of Veterans Affairs (VA) health care system. Patients.—Records for 3034 patients undergoing elective

CABG-only procedures; 1033 patients received a blood component transfusion during CABG. Results.—Previous heart surgery and decreased ejection fraction were significant predictors of transfusion for all blood components. Platelet count was predictive of platelet transfusion and FFP utilization. Baseline hemoglobin was a predictive factor for more than 2 units of RBC. Some significant hospital variation was noted beyond that predicted by patient risk factors alone. Conclusions.—Prediction models based on preoperative variables may facilitate blood component management for patients undergoing elective CABG. Algorithms are available to predict transfusion resources to assist blood banks in improving responsiveness to clinical needs. Predictors for use of each blood component may be identified prior to elective CABG for VA patients. (Arch Pathol Lab Med. 2003;127:415–423)

B

well as the mean number of units per patient, varies markedly from hospital to hospital. For example, the mean number of red blood cells (RBC) transfused in CABG ranges from 0 to 6.3 units per patient, and the frequency of transfusion ranges from 16% to 100%.1–5 Significant transfusion variation has also been noted with fresh frozen plasma (FFP), platelets, and cryoprecipitate.1,3–5 This variability may be related to differences in case severity, surgical skill and experience, or institutional transfusion practices.6 Obviously, given the variation in blood transfusion during CABG, it would be advantageous to identify patients requiring greater than the standard number of blood components during surgery. Approximately 7000 to 8000 cases of cardiac surgery are performed annually within the Department of Veterans Affairs (VA) health care system at 43 medical centers nationwide. Of these cardiac surgery procedures, approximately 80% are CABG-only procedures. The predominantly older male VA patients undergoing CABG often have multiple comorbid conditions, increasing their operative risk. As the world’s largest managed care organization, the VA health care system also operates with constrained financial resources and must provide quality medical care in the most cost-effective manner possible. Transfusion practices in many institutions require that 2 units of RBC be crossmatched for CABG. However, other VA hospitals

lood component transfusion has been an important part of coronary artery bypass graft surgery (CABG) since its inception. In the early days of CABG, almost all patients received blood components or whole blood. Despite current reductions in transfusion requirements for patients undergoing CABG, many patients continue to require transfusion because of (1) the increased number of acutely ill patients undergoing CABG, (2) surgical complications for patients with repeat cardiac surgery procedures, and (3) excessive bleeding among patients with coronary artery disease on anticoagulant therapy. In addition, bypass systems and hypothermia further compromise hemostasis. The proportion of patients undergoing transfusion, as

Accepted for publication October 3, 2002. From the Departments of Pathology (Drs Covin, Brimhall, Reiquam, and Shroyer), Medicine (Dr Shroyer), and Preventive Medicine and Biometrics (Dr Grunwald), University of Colorado Health Sciences Center School of Medicine, and the Departments of Cardiac Research (Drs O’Brien and Shroyer) and Pathology (Dr Rajagopalan), Denver Veterans Affairs Medical Center, Denver, Colo; the Department of Cardiothoracic Surgery, University of Arizona Health Sciences Center, and the Tucson Veterans Affairs Medical Center, Tucson, Ariz (Dr Sethi); and the University of Colorado College of Business and Administration, Denver (Dr Walczak). Reprints: A. Laurie Shroyer, PhD, Cardiac Research (112R), Denver Veterans Affairs Medical Center, 1055 Clermont St, Denver, CO 80220 (e-mail: [email protected]). Arch Pathol Lab Med—Vol 127, April 2003

Transfusion During Coronary Artery Bypass—Covin et al 415

have policies requiring crossmatching 4 units of RBC, reserving at least 1 apheresis platelet unit (or 6–10 randomdonor platelet units) and 2 units of FFP for each CABG procedure. In our experience at the Denver VA Medical Center, certain patients require transfusion of large numbers of blood components. Financial and space constraints limit the number of blood components that may be stored in inventory at any one time. A model to predict patients at risk for requiring more than the standard number of blood components would improve the blood bank’s ability to provide efficient service to the clinical and surgical staff as well as improve inventory management. The present study was carried out to determine which combinations of preoperative demographic, clinical, and laboratory factors are the best predictors of blood component usage during elective CABG. Although several previous studies have evaluated RBC transfusion during CABG, very few, if any, have extensively evaluated FFP and platelet usage during CABG. The overall goal for the use of these predictive models, therefore, was to facilitate the most efficient approach for the use of blood bank services in elective CABG procedures performed within the VA health care system. MATERIALS AND METHODS Patients Data were extracted from the VA ‘‘Processes, Structures, and Outcomes of Care in Cardiac Surgery’’ (PSOCS) data set derived from the VA Cooperative Studies Program No. 5. From September 1992 to December 1996, the PSOCS study team gathered data from 14 VA medical centers for 4969 cardiac surgery patient records. Relevant to this PSOCS study, more than 150 patient risk factors and more than 450 process of care variables were recorded, in addition to the traditional patient outcomes of care, which included 30-day operative mortality and morbidity (PSOCS data forms available upon request). In November 1999, this team’s substudy proposal received approval by the PSOCS Executive Committee to extract a subset of the database records and fields to evaluate the predictors of blood component use for elective CABG-only procedures. For this substudy, 83 preoperative variables were extracted from the PSOCS data set for evaluation (see Appendix). Intraoperative transfusion of all blood components was also extracted. These transfusion data represent the primary study outcomes for this PSOCS substudy analysis. Patients were included for analysis if they underwent isolated, elective CABG-only procedures. Exclusion criteria included simultaneous CABG and other concurrent procedures, such as cardiac valve surgery, aneurysmectomy, great vessel repair, and electrophysiology, as well as urgent (clinical condition–mandated surgery within 12 to 72 hours) or emergent (clinical condition– mandated immediate surgery) procedures. In addition, patients with preoperative intra-aortic balloon pump support, documented myocardial infarction within 7 days of surgery, percutaneous coronary angioplasty within 72 hours of surgery, or an American Society of Anesthesiologists classification of V (moribund patient unlikely to survive 24 hours with or without operation) were excluded. All procedures were performed ‘‘on pump.’’

Dependent Variables The primary endpoints of interest were (1) transfusion of more than 2 units of FFP, (2) transfusion of platelets, and (3) transfusion of more than 2 units of RBC. Only blood components transfused intraoperatively were counted; pre- and postoperative blood usage was not evaluated.

Independent Variables Using available literature related to predictors of blood component use as well as multidisciplinary clinical review, a concep416 Arch Pathol Lab Med—Vol 127, April 2003

tual model for this study was developed to identify the patient preoperative risk characteristics that may be associated with transfusion of blood components. Demographic factors include parameters such as age and gender. Surrogate variables for coronary artery disease severity include prior heart surgery, cardiomegaly, use of diuretics, preoperative use of heparin, and ejection fraction. Comorbidity was assessed using variables such as history of cerebrovascular disease, preoperative serum creatinine, serum albumin, platelet count, hemoglobin, and history of gastrointestinal bleeding. Finally, patients’ health behavior and lifestyle choices were assessed by current smoking status and body mass index. More than 80 preoperative variables (see Appendix) were extracted from the larger data set to permit construction of a comprehensive model.

Statistical Analysis Statistical analyses were done in S-plus,7 including the DESIGN and HMISC software libraries of Azola and Harrell.8 For each of the dichotomous study endpoints, logistic regression was used to estimate factors affecting the use of each of the 3 blood components separately. A full model, including all candidate variables, was estimated, and then backward selection with a cutoff of P # .05 was used to attempt to obtain smaller sets of variables yielding acceptable predictions for each outcome. Model performance was quantified using the c-index for discrimination (how well the model discriminates between those who received and those who did not receive a specific blood component), as well as the Hosmer-Lemeshow test for calibration (how well the estimated model probabilities match empirical probabilities of blood component use within deciles of predicted use estimates).9 The c-index, representing the area under the receiver-operating characteristic curve, may theoretically range from 0.5 (correct prediction due to chance alone) to 1.0 (perfect prediction). All patients were included in developing models. Bootstrap and cross-validation methods were used to adjust for overfitting and for evaluating the performance of the model when applied to other samples from the same patient population. The effect of hospital on blood component transfusion was evaluated by including a hospital indicator variable. The bootstrap methods described by Harrell et al10 are designed to avoid the overoptimism associated with evaluating models on the same sample of patients to which they were fit while avoiding the problems associated with splitting the data into separate ‘‘learning’’ and ‘‘test’’ sets (unusual data splits, reduced sample size, etc). Bootstrapping also includes the variability due to the modeling process and backward selection. Model performance, as quantified by the c-index, was assessed with this adjustment for overfitting.

RESULTS Patients Of the 4969 cardiac surgery records contained in the PSOCS database, 3998 (80.5%) represented CABG-only procedures. Of these CABG-only procedures, 3126 (78.2%) were recorded as elective surgery at the time of operation. An additional 84 cases were excluded for clinical reasons because of the use of preoperative intra-aortic balloon pump, myocardial infarction within 7 days of surgery, percutaneous coronary angioplasty performed within 72 hours of surgery, or an American Society of Anesthesiologists classification of V. Eight more patients were excluded from all analyses because they were missing the outcome of blood component usage. Demographic data for these 3034 patients are shown in Table 1. Predictors of Blood Component Use Overall, 37.3% (1133 patients of 3034) of the patients received a blood component transfusion during CABG. A total of 261 patients received FFP (8.6%), and 101 of these Transfusion During Coronary Artery Bypass—Covin et al

Table 1. Patient Demographic Data* Risk Factor

% of Patients

Vital statistics Male Female Age, y 6 SD Body mass index, kg/m2 6 SD

Frequency

99.4 0.6 63.3 6 9.2 28.1 6 4.9

Medical history Prior heart surgery Prior myocardial infarction Cerebrovascular disease Peripheral vascular disease Chronic obstructive pulmonary disease Current smoker Liver disease Gastrointestinal bleeding Diabetes mellitus

9.4 57.8 17.6 27.7 14.9 25.9 1.8 2.0 27.2

Medications Aspirin Heparin Diuretics

58.8 29.3 22.6

Laboratory and radiologic evaluation Platelet count, 3109/L .150 100–150 ,100

91.7 6.9 1.4

Hemoglobin, g/dL .12 10–12 ,10

89.9 8.6 1.5

Creatinine, mg/dL ,1.5 1.5–3.5 .3.00 Albumin, ,3.5 mg/dL

82.5 15.9 1.6 9.0

No. of coronary arteries with .50% occlusion 0 or 1 2 3

13.3 38.4 48.3

Ejection fraction $0.55 0.45–0.54 0.35–0.44 0.25–0.34 ,0.25

47.7 23.8 17.1 9.4 2.0

Cardiomegaly

15.5

Other Rales ST segment depression on ECG Functional status Independent Dependent Angina class† I II III IV

5.8 13.7 97.1 2.9 19.7 15.3 27.0 38.0

NYHA Congestive heart failure class I 42.3 II 23.2 III 24.2 IV 10.4 * ECG indicates electrocardiography; NYHA, New York Heart Association. † Canadian Cardiovascular Society functional class. Arch Pathol Lab Med—Vol 127, April 2003

Table 2. Intraoperative Use of FFP, Platelets, and RBC During Elective Coronary Artery Bypass Graft Surgery* %

Units of platelets tranfused 0 2721 1 27 2 8 3 2 4 3 5 5 6 52 7 10 8 37 9 1 10 131 11 1 12 9 14 1 15 2 16 5 20 15 30 2 40 2 Mean per patient 0.91 units

89.7 0.9 0.2 0.1 0.1 0.2 1.7 0.3 1.2 0.0 4.3 0.0 0.3 0.0 0.1 0.2 0.5 0.1 0.1 ...

Units of FFP transfused 0 1 2 3 4 5 6 7 8 9 10 12 16 28 Mean per patient

91.4 0.8 4.5 0.3 2.4 0.1 0.3 0.0 0.1 0.0 0.1 0.0 0.0 0.0 ...

2773 23 137 8 72 2 9 1 3 1 2 1 1 1 0.26 units

Units of RBC transfused 0 1994 65.7 1 243 8.0 2 435 14.3 3 171 5.6 4 105 3.5 5 33 1.1 6 30 1.0 7 8 0.2 8 5 0.2 9 2 0.1 10 5 0.2 12 1 0.0 16 2 0.1 Mean per patient 0.85 units ... * FFP indicates fresh frozen plasma; RBC, red blood cells.

patients received more than 2 units of FFP (3.3% of the total and 38.7% of those who received FFP). Overall, 313 patients received platelet transfusions (10.3%). Additionally, 1040 patients received RBC transfusions (34.3%), of which 362 received more than 2 units of RBC during surgery (11.9% of the total and 34.8% of the patients who received RBC). The distribution of the number of units of FFP, platelets, and RBC transfused during CABG is shown in Table 2. In backward selection with multiple logistic regression analysis, patient demographic, clinical, and laboratory factors independently associated with transfusion of more than 2 units of FFP are shown in Table 3a. Factors Transfusion During Coronary Artery Bypass—Covin et al 417

Table 3a. Factors Associated With Transfusion of More Than 2 Units of FFP* Odds Ratio (95% CI)

P Value

8.28 (5.39, 12.72)

,.001

Platelet count, 310 /L 100–150 ,100

2.50 (1.40, 4.47) 4.46 (1.59, 12.52)

.002 .005

History of gastrointestinal bleeding

3.16 (1.15, 8.68)

.03

Ejection fraction 0.45–0.54 0.35–0.44 0.25–0.34 ,0.25

0.74 1.71 1.94 2.36

.33 .046 .04 .19

Risk Factor

Prior heart surgery 9

Heparin c-Index c-Index adjusted for overfitting Hosmer-Lemeshow goodness of fit * FFP indicates fresh frozen plasma.

(0.41, (1.01, (1.04, (0.65,

1.35) 2.91) 3.61) 8.53)

1.59 (1.04, 2.43) 0.768 0.724 P 5 .89

.03

Table 3b. Factors Associated With Transfusion of Platelets Risk Factor

Odds Ratio (95% CI)

P Value

Platelet count, 310 /L ,100 100–150

7.55 (3.85, 14.80) 1.83 (1.23, 2.74)

,.001 .003

Prior heart surgery

4.83 (3.58, 6.51)

,.001

Creatinine, mg/dL 1.5–3.0 .3.0

1.25 (0.91, 1.72) 3.05 (1.53, 6.09)

.16 .002

Ejection 0.45–0.54 0.35–0.44 0.25–0.34 ,0.25

1.02 1.45 1.35 2.60

1.41) 2.02) 2.05) 5.27)

.88 .03 .16 .008

Diuretic use Albumin, ,3.5 mg/dL Age (per 10 y) c-Index c-Index adjusted for overfitting Hosmer-Lemeshow goodness of fit

1.54 (1.16, 2.04) 1.53 (1.06, 2.22) 1.21 (1.05, 1.50) 0.724 0.694 P 5 .53

.002 .02 .01

9

associated with platelet transfusion and transfusion of more than 2 units of RBC are shown in Table 3, b and c, respectively. Odds ratios for predictive models are also shown in Table 3, a through c, to permit calculation of estimated probabilities of blood component use. Model Performance Receiver-operating characteristic curves for each of the 3 outcomes are shown in Figure 1. The models best predict RBC use (c-index 5 0.752 adjusted for overfitting and backward selection) with somewhat less ability to predict FFP and platelet use (c-indices 5 0.724 and 0.694, respectively, adjusted for overfitting and backward selection). There is a tendency for patients with a high probability of using one blood component to have a high probability of using other components also. Correlation coefficients of estimated probabilities of blood component use were as follows: (RBC, FFP) 5 0.80; (RBC, platelets) 5 0.54; (FFP, platelets) 5 0.41; and P , .001 for all. Hospital Effects There is substantial variability between hospitals in the use of all blood components, as revealed in Figure 2. The 418 Arch Pathol Lab Med—Vol 127, April 2003

(0.74, (1.04, (0.89, (1.29,

incidence of blood component usage in hospitals varied from 1.6% to 28.4% for RBC, from 0% to 9.7% for FFP, and from 4.8% to 18.4% for platelets. Table 4 documents cindices describing predictive ability of the backward selection models shown in Table 3. c-Indices for models using only hospitals to predict blood usage were much lower than those for models using only preoperative predictors. In all cases, including a hospital indicator in addition to preoperative predictors significantly improved predictive power above that achieved using only preoperative predictor variables. The improvements in predictive ability of the models as quantified by c-indices are relatively slight, although the improvements are highly significant statistically since the sample size is relatively large. COMMENT Predictive models have been developed using preoperative demographic, clinical, and laboratory factors to predict transfusion of more than 2 units of FFP and RBC and to predict platelet transfusion during elective CABG. Although previous studies have defined excessive bleeding based on the volume of chest tube drainage11 or transfusion of more than 5 units of RBC,12 we defined our threshTransfusion During Coronary Artery Bypass—Covin et al

Table 3c. Factors Associated With Transfusion of More Than 2 Units of RBC* Odds Ratio (95% CI)

P Value

14.61 (5.22, 40.93)

,.001

Hemoglobin, g/dL 10–12 ,10.0

3.21 (2.31, 4.44) 8.06 (4.12, 15.77)

,.001 ,.001

Prior heart surgery

3.36 (2.44, 4.63)

,.001

Ejection fraction 0.45–0.54 0.35–0.44 0.25–0.34 ,0.25

0.84 1.17 2.02 3.32

.27 .36 ,.001 ,.001

History of gastrointestinal bleeding

2.32 (1.21, 4.45)

.01

Creatinine, mg/dL 1.5–3.0 .3.0

1.44 (1.08, 1.93) 2.06 (1.01, 4.22)

.01 .048

Age (per 10 y)

1.59 (1.36, 1.86)

,.001

NYHA class II III IV

1.50 (1.10, 2.05) 1.55 (1.14, 2.10) 1.17 (0.77, 1.76)

.01 .005 .46

History of peripheral vascular disease 1.33 (1.03, 1.71) Body mass index, kg/m2 0.94 (0.92, 0.97) Current smoker 0.70 (0.51, 0.95) c-Index 0.775 c-Index adjusted for overfitting 0.752 Hosmer-Lemeshow goodness of fit P 5 .24 * NYHA indicates New York Heart Association; RBC, red blood cells.

.03 ,.001 .02

Risk Factor

Female gender

old as the transfusion of more than 2 units of FFP or RBC. We considered this a reasonable threshold because many transfusion facilities, as a matter of policy, type and crossmatch 2 units of RBC for an elective CABG-only procedure. The need for more than 2 units of RBC would require the blood bank to perform additional crossmatches during surgery and would, therefore, require increased resources. A number of pharmacologic and nonpharmacologic methods for reducing transfusion requirements13–16 are currently used, and many CABG surgeries are performed without the need for transfusion.17–22 Antiplatelet drugs now used commonly (eg, ticlopidine, clopidogrel, and abciximab) were very rarely used in the VA patient population, and data on their use were not collected during the study period. In our study, most surgical procedures (62.6%) did not require transfusion. The number of patients not requiring blood products may possibly be explained by the following factors: 1. This substudy evaluated only intraoperative blood product use and did not include the postoperative period when blood products may be commonly administered; 2. Patients with emergent procedures were excluded; and 3. General practice patterns for use of blood products may be somewhat unique to this subset of Veterans Administration hospitals with cardiac surgery programs. A large, diverse set of multidimensional risk factors was used to develop the most comprehensive prediction models possible. The most consistent predictor across all 3 outcomes was prior heart surgery, a fairly common condition (9.4% of patients) with a large odds ratio (.3); this risk factor was also the most highly significant (P , .001) predictor for all 3 outcomes. Prior heart surgery was found Arch Pathol Lab Med—Vol 127, April 2003

(0.61, (0.84, (1.39, (1.72,

1.15) 1.63) 2.93) 6.38)

to be a predictor of excessive bleeding and blood component transfusion in several previous studies.3,11,20 In addition, decreased ejection fraction was found to be a significant risk. Magovern and colleagues20 also identified decreased ejection fraction (,0.30) as a predictor for RBC transfusion during CABG in a multivariate logistic regression model. Additional factors associated with transfusion of more than 2 units of FFP in our study included platelet counts ,150 3 109/L, a history of gastrointestinal bleeding, and the preoperative use of heparin. Other factors associated with platelet transfusion were platelet counts ,150 3 109/ L, creatinine levels .3.0 mg/dL (270 mmol/L), diuretic usage, albumin levels ,3.5 mg/dL, and age. Surgenor and colleagues3 identified reoperation, graft type, and RBC transfusion as factors associated with transfusion of FFP and platelets. In their study, however, platelet count did not correlate with platelet transfusion. In our study, female gender, increasing age, a history of gastrointestinal bleeding or peripheral vascular disease, current nonsmoker status, low hemoglobin level, renal insufficiency, low body mass index, and a New York Heart Association classification of II or greater were associated with an increased risk of excessive intraoperative transfusion of RBC. It should be noted, however, that few women were included in our study (0.6% of total); thus, firm conclusions regarding the importance of the female gender in predicting RBC transfusion cannot be drawn from this study. In spite of this limitation, these findings are consistent with those found in previous studies.3,11,20,23 Increasing age,3,11,17,20,23,24 decreased preoperative hematocrit or red cell mass/volume,3,12,17–20,23,24 and renal insufficiency or failure have all been associated with excessive bleeding.3,20,25 Transfusion During Coronary Artery Bypass—Covin et al 419

Figure 1. Model performance for predicting transfusion of more than 2 units of red blood cells (RBC) and fresh frozen plasma (FFP) and platelet transfusion. FFP (——), platelets (········), and RBC (– ·– ·– ).

Figure 2. Incidence (%) of blood component use by hospital for red blood cells (RBC), fresh frozen plasma (FFP), and platelets.

A decreased body mass index has been shown to be associated with transfusion,20 although in one study, a low body mass index did not improve the predictive ability of the model using multivariate logistic regression analysis.17 Lower weight has also been shown to be a predictor of blood component transfusion.19 Active tobacco use or dependence has been identified as a risk factor for RBC transfusion.3 The contrary finding in our study may result from more sickly current smokers 420 Arch Pathol Lab Med—Vol 127, April 2003

having recently quit, resulting in their classification as nonsmokers. Previous reports have also identified abnormal coagulation studies,3,11,12 type 1 diabetes mellitus,3,20 decreased platelet count,11 decreased albumin (,4.0 g/dL),20 and the number or type of bypass grafts3 with excessive bleeding or increased risk of RBC transfusion during CABG, whereas our study did not. Other factors previously associated with increased risk of RBC transfusion or excesTransfusion During Coronary Artery Bypass—Covin et al

Table 4. Adjusted and Unadjusted C-indices for Prediction Models With and Without Hospital Effect* Unadjusted

Adjusted

Hospital P Value

RBC Model only Model and hospital Hospital only

0.775 0.827 0.68

0.752 0.804 0.664

... ,.001 ,.001

FFP Model only Model and hospital Hospital only

0.768 0.797 0.686

0.724 0.741 0.646

... ,.001 ,.001

Platelets Model only 0.724 0.694 ... Model and hospital 0.745 0.707 ,.001 Hospital only 0.619 0.595 ,.001 * FFP indicates fresh frozen plasma; RBC, red blood cells.

sive bleeding include: cardiogenic shock,3 first treatment or episode of transmural myocardial infarction,3 lower core temperature,11 prolonged cardiopulmonary bypass time,11,18 duration of surgery,23 large-volume intraoperative transfusion of salvaged RBC,11 urgent or emergent surgery,20 and catheterization-induced coronary occlusion.20 Intraoperative events are often not predictable preoperatively, but their inclusion would increase the predictive ability of our models. However, since it is not possible to have intraoperative information available to the blood bank at the time the units are prepared for upcoming surgery, such information was not included in predictive models for the purposes of these analyses. Whereas earlier studies have shown an association between preoperative use of aspirin and transfusion requirements in CABG,26–29 more recent studies have not shown an association.30–32 Our study did not indicate that aspirin usage was a risk factor for excessive transfusion of RBC, FFP, or platelets intraoperatively. Aspirin usage in the PSOCS data set was defined as ingestion of any medication(s) containing aspirin within 1 week prior to surgery. The model for predicting the transfusion of more than 2 units of RBC performed best (c-index 5 0.752), while the model for predicting platelet transfusion performed least well (c-index 5 0.694). The performance of our model to predict transfusion of more than 2 units of FFP was intermediate between that of RBC and platelets (c-index 5 0.724). The addition of a hospital indicator to the model significantly improved model performance in all 3 outcomes, although improvements in predictive power were relatively slight. These results support those found in other studies.4,5 For all 3 outcomes, the addition of preoperative risk variables, however, explained large variations in blood component use beyond that explained by hospital variation alone. Thus, most of the variation in blood component use is not simply due to hospital-specific practices, but rather, is related to individual patient preoperative risk factors. This result, therefore, may provide important insights into future research in this field. Previously cited studies differ from this study. For example, the study by Cosgrove and colleagues17 evaluated red cell transfusion over the patients’ entire hospitalization. Ferraris and Gildengorin12 developed a regression model to predict which patients would likely require more than 5 units of packed RBC postoperatively. Despotis and Arch Pathol Lab Med—Vol 127, April 2003

colleagues11 evaluated pre-, intra-, and postoperative variables involved in postoperative blood loss and transfusion of hemostatic blood components after cardiopulmonary bypass. Another study examined preoperative variables related primarily to the likelihood of red cell transfusion but also included data for other blood components.3 Magovern and colleagues20 evaluated preoperative predictors of red cell transfusion after CABG and included patients requiring emergency surgery. Our study reviewed preoperative variables and intraoperative blood usage, including RBC, FFP, and platelets, for elective CABG patients. In addition, our study was performed from the perspective of transfusion service management—using algorithms to predict patients requiring more blood components than are typically set up for patients undergoing elective CABG. Some of the most notable limitations to this substudy are outlined in the following paragraphs. 1. This substudy focused on the intraoperative use of blood components. 2. Most of the patients for this substudy (99.4%) were male. 3. The period for the PSOCS data captured is prior to common use of an ‘‘off-pump’’ procedure or recent use of antiplatelet drugs. 4. The prothrombin time and the international normalized ratio could not be included in the substudy analyses, as these PSOCS study values were recorded inconsistently as either a prothrombin time or an international normalized ratio. 5. Information was unavailable about the various thromboplastin reagents (international sensitivity index) used for the prothrombin time assay at the 14 different hospitals and whether or not the reagents were changed during this time. In addition, we could not express the prothrombin time or the international normalized ratio as a ratio using the midpoint of the reference range because the use of different thromboplastin reagents would result in different reference ranges. Also, the reference ranges used at the time of transfusion for each hospital were not available. 6. The number of therapeutic platelet transfusion events could not be easily quantitated. Data for the number of platelets transfused revealed peaks at 0, 1, 6, 8, and 10 units transfused, with smaller peaks at 2, 7, 12, 16, and 20 units transfused (Table 2). At the time of the study, both random-donor platelets and apheresis platelets were being transfused. Only the number of units transfused, not the number of transfusion events, was recorded. For example, in a patient who received 10 units of platelets, we could not be sure if the patient received 10 units of randomdonor platelets as a 10-pack (one transfusion event) or 8 units of random-donor platelets (as an 8-pack) and 2 units of apheresis platelets (3 transfusion events). This study includes several additional components that, to our knowledge, have not been reported in previous investigations. We used a large sample size of more than 3000 CABG operations with more than 80 wide-ranging variables available for evaluation. This extensive database allowed us to include a number of variables that ordinarily would not be available for evaluation in a single-center study. The bootstrap method of model development and cross-validation allowed us to use the entire data set for developing our models; the addition of a hospital indicator Transfusion During Coronary Artery Bypass—Covin et al 421

allowed us to evaluate variations in transfusion practice among different hospitals. Finally, we evaluated transfusion of FFP, platelets, and RBC and created separate models for each of these 3 blood components. This allowed us to use only those factors that were significant for transfusion of the specific blood component. A set of algorithms was developed to predict transfusion of more than 2 units of FFP and RBC, as well as to predict platelet transfusion in elective CABG-only surgery. Although not all patients requiring greater numbers of components can be predicted with 100% accuracy, these algorithms can be used by the blood bank to improve its efficiency and effectiveness to meet the needs of clinical and surgical teams. Prototype software based on these VA predictive models (provided in a Microsoft Excel spreadsheet format) for the calculation of estimated blood component use is available upon request. A nurse from the Denver VA Cardiac Research Unit was able to find all of the required patient data for cardiac surgery patients in less than 20 minutes using the VA’s computer system. In the future, risk algorithms may be useful for quality improvement initiatives related to blood component management. In comparison to other centers, a hospital’s riskadjusted blood component use may be employed as a screen to monitor and assess hospital performance. This project was funded in part by the Department of Veterans Affairs Health Services Research and Development grant IHY 99214-1 (Dr Shroyer, Principal Investigator at the Denver Veterans Affairs Medical Center). Funding was also provided by the Department of Pathology, University of Colorado Health Sciences Center, Denver, Colo. The initial data capture for this study was funded by the Department of Veterans Affairs Cooperative Studies Program Cooperative Studies Program Grant 5 (Drs Hammermeister, Grover, and Sethi, Co-Principal Investigators). References 1. Goodnough LT, Johnston MF, Toy PT. The variability of transfusion practice in coronary artery bypass surgery. JAMA. 1991;265:86–90. 2. Hasley PB, Lave JR, Hanusa BH, et al. Variation in the use of red blood cell transfusions: a study of four common medical and surgical procedures. Med Care. 1995;11:1145–1160. 3. Surgenor DM, Churchill WH, Wallace EL, et al. Determinants of red cell, platelet, plasma, and cryoprecipitate transfusions during coronary artery bypass graft surgery: the Collaborative Hospital Transfusion Study. Transfusion. 1996;36: 521–532. 4. Stover EP, Siegel LC, Parks R, et al. Variability in transfusion practice for coronary artery bypass surgery persists despite national consensus guidelines: a 24-institution study. Anesthesiology. 1998;88:327–333. 5. Surgenor DM, Churchill WH, Wallace EL, et al. The specific hospital significantly affects red cell and component transfusion practice in coronary artery bypass graft surgery: a study of five hospitals. Transfusion. 1998;38:122–134. 6. Bracey AW, Radovacevic R. The hematologic effects of cardiopulmonary bypass and the use of hemotherapy in coronary artery bypass grafting. Arch Pathol Lab Med. 1993;118:411–416. 7. S-plus 2000. Guide to statistics. Available at: http://www.insightful.com/ support/documentation.asp?did51. Accessed 2000. 8. Azola C, Harrell FE. An introduction to S-plus and the HMISC and DESIGN libraries. Available at: http://hesweb1.med.virginia.edu/biostat/s/doc/splus.pdf. Accessed 2001.

9. Hosmer DW, Lemeshow S. Applied Logistic Regression. New York, NY: John Wiley and Sons; 1989. 10. Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387. 11. Despotis GJ, Filos KS, Zoys TN, Hogue CW, Spitznagel E, Lappas DG. Factors associated with excessive postoperative blood loss and hemostatic transfusion requirements: a multivariate analysis in cardiac surgical patients. Anesth Analg. 1996;82:13–21. 12. Ferraris VA, Gildengorin V. Predictors of excessive blood use after coronary artery bypass grafting: a multivariate analysis. J Thorac Cardiovasc Surg. 1989;98: 492–497. 13. Ferraris VA, Ferraris SP. Limiting excessive postoperative blood transfusion after cardiac procedures: a review. Texas Heart Inst J. 1995;22:216–230. 14. Cooley DA. Conservation of blood during cardiovascular surgery. Am J Surg. 1995;170:53S–59S. 15. Janssens M, Hartstein G, David J. Reduction in requirements for allogeneic blood products: pharmacologic methods. Ann Thorac Surg. 1996;62:1944–1950. 16. Hardy J-F, Belisle S, Janvier G, Samama M. Reduction in requirements for allogeneic blood products: nonpharmacologic methods. Ann Thorac Surg. 1996; 62:1935–1943. 17. Cosgrove DM, Loop FD, Lytle BW, et al. Determinants of blood utilization during myocardial revascularization. Ann Thorac Surg. 1985;40:380–384. 18. Scott WJ, Rode R, Castlemain B, et al. Efficacy, complications, and cost of a comprehensive blood conservation program for cardiac operations. J Thorac Cardiovasc Surg. 1992;5:1001–1007. 19. Paone G, Spencer T, Silverman NA. Blood conservation in coronary artery surgery. Surgery. 1994;116:672–678. 20. Magovern JA, Sakert T, Benckart DH, et al. A model for predicting transfusion after coronary artery bypass grafting. Ann Thorac Surg. 1996;61:27–32. 21. Rosengart TK, Helm RE, DeBois WJ, Garcia N, Krieger KH, Isom OW. Open heart operations without transfusion using a multimodality blood conservation strategy in 50 Jehovah’s Witness patients: implications for a ‘‘bloodless’’ surgical technique. J Am Coll Surg. 1997;184:618–629. 22. Helm RE, Rosengart TK, Gomez M, et al. Comprehensive multimodality blood conservation: 100 consecutive CABG operations without transfusion. Ann Thorac Surg. 1998;65:125–136. 23. Shevde K, Pagala M, Kashikar A, et al. Gender is an essential determinant of blood transfusion in patients undergoing coronary artery bypass graft procedure. J Clin Anesth. 2000;12:109–116. 24. Bilfinger TV, Conti VR. Blood conservation in coronary artery bypass surgery: prediction with assistance of a computer model. Thorac Cardiovasc Surg. 1989;37:365–368. 25. Anderson RJ, O’Brien M, Whinney SM, et al. Renal failure predisposes patients to adverse outcome after coronary artery bypass surgery. Kidney Int. 1999;55:1057–1062. 26. Ferraris VA, Ferraris SP, Lough FC, Berry WR. Preoperative aspirin ingestion increases operative blood loss after coronary artery bypass grafting. Ann Thorac Surg. 1988;45:71–74. 27. Sethi GK, Copeland JG, Goldman S, Moritz T, Zadina K, Henderson WG. Implications of preoperative administration of aspirin in patients undergoing coronary artery bypass grafting. Department of Veterans Affairs Cooperative Study on Antiplatelet Therapy. J Am Coll Cardiol. 1990;15:15–20. 28. Taggart DP, Siddiqui A, Wheatley DJ. Low-dose preoperative aspirin therapy, postoperative blood loss, and transfusion requirements. Ann Thorac Surg. 1990;50:424–428. 29. Bashein G, Nessly ML, Rice AL, Counts RB, Misbach GA. Preoperative aspirin therapy and reoperation for bleeding after coronary artery bypass surgery. Ann Intern Med. 1991;151:89–93. 30. Rawitscher RE, Jones JW, McCoy TA, Lindsley DA. A prospective study of aspirin’s effect on red blood cell loss in cardiac surgery. J Cardiovasc Surg. 1991; 32:1–7. 31. Reich DL, Patel GC, Vela-Cantos F, Bodian C, Lansman S. Aspirin does not increase homologous blood requirements in elective coronary bypass surgery. Anesth Analg. 1994;79:4–8. 32. Tuman KJ, McCarthy RJ, O’Connor CJ, Ivankovich AD. Aspirin does not increase allogeneic blood transfusion in reoperative coronary artery surgery. Anesth Analg. 1996;83:1178–1184.

APPENDIX Preoperative Independent Variables Extracted Use of aspirin or aspirin containing drugs within 1 week New York Heart Association functional class Prior heart surgery Current smoker History of cigarette smoking History of chronic obstructive pulmonary disease History of hypertension requiring medical prescription 422 Arch Pathol Lab Med—Vol 127, April 2003

Family history of premature heart disease History of liver disease within the last 5 years History of alcohol abuse within the last year History of illicit drug use within the last 2 years History of peripheral vascular disease History of cerebrovascular disease Height

Transfusion During Coronary Artery Bypass—Covin et al

Weight Distance from home to Veterans Affairs Medical Center Birth date Admission date Current smoker History of cigarette smoking Diabetes Type of prior heart surgery General medical condition New York Heart Association functional classification Use of thrombolytic therapy within the last 48 hours in chart Record of illicit drug use noted in chart Gender Total no. of distal anastomoses performed With saphenous vein With left internal mammary artery graft With cephalic or upper limb vein With right internal mammary anastomosis With gastroepiploic artery With synthetic/cryopreserved graft Proximal anastomoses performed Caseload by cardiothoracic (CT) surgery in last quarter of cases requiring cardiopulmonary bypass (CPB) CT attending—average hours per week in direct patient care activities Year of most senior general surgery resident Average turnaround time for stat complete blood count during quarter Average duration of rotation of most senior resident during quarter Chemistry laboratory proximity to main Veterans Affairs hospital No. of operating rooms capable of handling CPB Size of primary cardiac surgery operating room Age of primary CPB machine CT surgery attendings CT residents/fellows

Arch Pathol Lab Med—Vol 127, April 2003

Prior myocardial infarction Preoperative use of intra-aortic balloon pump History of gastrointestinal bleeding in the last 30 days Hepatomegaly Serum creatinine Hemoglobin Platelet count Glucose Albumin Prothrombin time/international normalized ratio Partial thromboplastin time Cardiac catheterization within 6 months Where was most recent cardiac catheterization performed Date most recent cardiac catheterization was performed Complications—cardiac catheterization Type of complication—cardiac catheterization CT attending supervising other simultaneous cases Primary surgeon First assistant surgeon CT attending scrubbed during cannulation CT attending scrubbed during CPB CT attending scrubbed during weaning from bypass Most senior surgeon present during cannulation Most senior surgeon present during CPB Most senior surgeon present during weaning from CPB Primary anesthesiologist Anesthesiologist attending supervising other cases Most senior anesthesiologist present during cannulation Most senior anesthesiologist present during CPB Most senior anesthesiologist present during weaning from CPB Age Body surface area/body mass index Hospital indicator

Transfusion During Coronary Artery Bypass—Covin et al 423