- Email: [email protected]
Homologous blood transfusion as a risk factor for postoperative infection after coronary artery bypass graft operations
Homologous blood transfusion as a risk factor for postoperative infection after coronary artery bypass graft operations Homologous transfusions are immunosuppressive and associated with a higher risk of postoperative infection. In this retrospective analysis, we studied 238 consecutive patients who underwent first-time coronary operations by a single surgeon in 1988 to 1989 and collected clinical and laboratory data relevant to postoperative infections including pulmonary, urinary, and wound sites. Culture-proved postoperative infections occurred in 16 of the 238 patients (6.7 %), with only 3 (1.3%) being deep sternal wound infections. Seven of 16 (44%) of the infections were away from the wound sites, suggesting that nonsurgical variables contributed to at least some infections. Factors significantly associated with an increased risk of postoperative infection by univariate analysis included female sex, diabetes mellitus, and transfusion dose. Infections occurred in 3.9 % of patients receiving up to 2 units of red cells and whole blood, 6.9 % receiving 3 to 5 units, and 22 % of those receiving 6 units or more. Multiple linear and logistic regression analysis showed that transfusion dose was the most significant predictor of infection, days of fever, days of antibiotic therapy, and length of hospital stay. Homologous transfusion is associated (in a dose-dependent fashion) with a threefold to eightfold increased risk of postoperative infection in patients undergoing coronary artery operations. This increased risk of infection may be due to transfusion-induced immunosuppression of the patient. (J 'fHORAC CARDIOVASC SURG 1992;104:1092-9)
Paul J. Murphy, MD,a Cliff Connery, MD,b George L. Hicks, Jr., MD,b and Neil Blumberg, MD,a Rochester, N.Y.
Extensive data suggest that homologous blood transfusions have immunosuppressive effects on recipients. Transfusion-induced immunomodulation is believed responsible for increased renal allograft survival in patients who receive pretransplantation transfusion. 1 More recent studies suggest there may be other clinical consequences of blood transfusion-induced immunomodulation, including increased rates of solid tumor recurrence after resection.i decreased recurrence rates of From the Transfusion Medicine Unit; Department of Pathology and Laboratory Medicine, and Division of Cardiothoracic Surgery," Department of Surgery, University of Rochester Medical Center, Rochester, N.Y. Supported in part by a research grant from the American Association of Blood Banks Foundation, Arlington, Va. Received for publication Sept. 20, 1990. Accepted for publication Sept. 16, 1991. Address for reprints: Neil Blumberg, MD, University of Rochester Medical Center, Box 608, 601 Elmwood Ave., Rochester, NY 14642.
12/1/33818
1092
Crohn's disease.' and increased severity of viral infection." Other studies show an increased risk of postoperative bacterial infection in patients receiving homologous blood transfusions after abdominal or orthopedic operations."!' Few clinical studies have been designed specifically to examine the relationship between transfusion and postoperative infection in those undergoing cardiac operations. One large retrospective study'? examined patients undergoing a variety of cardiac procedures and found that blood transfusion was significantly associated with an increased risk of postoperative infection. Another, smaller study' ' found a similar association in those patients undergoing valve replacement but not in those who underwent coronary artery bypass grafting (CABG). The purpose of the present study is to assess further whether homologous blood transfusion is an independent risk factor for postoperative infection in those undergoing CABG.lnasmuch as the possibility exists that transfusion acts merely as a surrogate measure of other clinical variables known to predict infection, we designed the study to
Volume 104 Number 4 October 1992
address confounding variables such as length of the operation and preadmission anemia.
Methods We retrospectively reviewed the medical records of all patients who underwent first-time CABG in our institution from June 1988toJune 1989. These patients were treated by a single cardiothoracic surgeon (G.L.H.). The variables examined include those previously reported to be risk factors for wound infectionafter cardiac surgery-v!": age; sex;presence or absence of diabetes mellitus; smoking history; length of hospital stay; days of preoperative hospital stay including intensive care unit stay; number of coronary arteries bypassed; whether an internal mammary artery was used; duration of the operation; duration of extracorporeal circulation; whether reoperation was necessary; total white blood cell count; absolute numbers of neutrophils,lymphocytes,and monocytes; hematocrit values on admission and discharge; admission serum albumin; transfusionsduring the perioperative period; therapeutic use of antibiotics (all patients received postoperative antibiotic prophylaxis for 3 or 4 days); and days of temperatures above 38.30 C (\01 0 F).
Criteria for infection included strong clinical suspicion of infection by the attending surgeon, the institution of antibiotic therapy, and confirmation by a positive culture. When more liberal criteria for suspected infection were used," the results were identical to those reported when a positive culture was required for confirmation. Patient follow-upincluded reviewing the outpatient record for two office visits to the attending surgeon,both of which usually took place within I month after discharge from the hospital. A total of 248 consecutive patients were reviewed. Ten patients were excluded from the study because of preexisting bacterial infectionsat the time of hospital admission, for which they were treated with antibiotics (seven cases of urinary tract infection, two of pneumonia, and one of prostatitis). This left 238 patients evaluated for this study. Complete data were available for 234 of them, with one missing datum for each of the four patients with incomplete data. Because immunosuppression may be due to red cell, white cell,or plasma content of the transfused blood components, we used the following values to calculate the approximate red cell dose, white cell dose, and plasma dose received by each patient having a transfusion: (I) Red cells: Each unit of packed red bloodcellsor whole blood was considered to have 200 ml of red cells; a value of zero was used for platelets and fresh frozen plasma. (2) White cells: A white cell content of I X 109 cellsper unit of wholeblood or packed red cells and 0.4 X 109 white cells per unit of platelet concentrate was used. (3) Plasma: The plasma content was 300 ml for each unit of whole blood, 50 ml for a unit of packed red cells, 50 ml for a unit of platelet concentrate, and 250 ml for a unit of fresh frozen plasma. We also quantitated transfusion simply as number of components received; that is,each unit of red cells,platelet concentrate, fresh frozen plasma, and whole blood was counted as I unit of blood. Both parametric and nonparametric methods of data analysis were used, with JMP Software (SAS Institutes, Inc., Cary, N.C.) or Stat View SE + Graphics (Abacus Concepts, Inc., Berkeley, Calif.) on a Macintosh Ilcx microcomputer (Apple Computer, Cupertino, Calif.). Parametric results are generally reported (e.g., analysis of variance or t test), but these were alwaysconfirmedby a nonparametric equivalent when available
Infection after CABG operations
1093
(e.g., Kruskal-Wallis or Mann-Whitney). Risk ratios were calculated for factors significantly associated with postoperative infection by dividing the prevalence of infection in the high risk subgroup by the prevalence in the lower risk subgroup. N inetyfive-percent confidencelimits were calculated for risk ratios for single factors by the normal deviate and the x2 distribution. Because of the small number of infections in any single risk factor subgroup, attributable risk for each factor was not calculated. For regression analyses, no transformations ofvariables were used.
Results Of the 238 patients, 16 (6.7%) had a culture-proved postoperative infection during the inpatient or postdischarge period. We first analyzed these infections by a univariate approach to determine which clinical or laboratory variables were associated with infection (Table I). Length of stay, days of fever, and days of antibiotics were documented because of their potential utility as surrogate measures of infection and as an indication of morbidity and expense associated with infection. Because transfusion was one of the variables significantly different between the patients who had infections and those who did not (Table I), we also performed a similar univariate analysis to determine which clinical and laboratory variables were significantly different between those receiving or not receiving transfusion (Table II). The details of the patients with infection are shown in Table III. Only 3 of 238 (1.3%) patients had deep or severe sternal wound infections. There were a number of infections at sites other than the wounds, including pneumonia, empyema, and five urinary tract infections (7/16 infections, 44%). Two of these infections also led to generalized sepsis. Of these seven distant infections, six occurred in patients who had a transfusion. As can be seen from Table I, the patients having infections were significantly more likely to be female (50% versus 22% in the noninfected group), to be diabetic (44% versus 17%), and to have received significantly more blood transfusions (25 total components versus 4). In addition to being the quantitatively most striking difference between the patients with and without infections, transfusion appeared to be associated with an increased risk of infection in a dose-dependent manner. Infections were seen in 6 of 153 (3.9%) patients receiving up to 2 units of red cells or whole blood, 4 of 58 (6.9%) patients receiving 3 to 5 units, and 6 of 27 (22%) patients receiving 6 units or more (p = 0.0022 by x2 analysis) (Table IV). Other boundaries for grouping the red cell dose yielded the same results. Measures of infection-related expense and morbidity including length of stay, days of antibiotics, and fever, as expected, were significantly greater in the infected group than in the uninfected group. Interestingly, potentially
1094
The Journal of Thoracic and Cardiovascular Surgery
Murphy et al.
Table I. Characteristics ofpatients with culture-proved infection as compared with those without infection Variable No. of patients Male Female Age (yr) Length of stay (days) Diabetes Smoking Days of preop. rcu No. of grafts Duration of operation (min) Time on extracorporeal circulation (min) Reexploration required Total lymphocyte count (thousand / ILL) Preop. hematocrit value (%) Discharge hematocrit value (%) Preop. albumin (gm/dl) Total No. of blood components transfused WBC dose (x \09) Plasma dose (m!) Red cell dose (units) Days of antibiotics Days of fever
Infection
No infection
p Value
16 8/16 (50%) 8/16 (50%) 66 ± 7.5 25 ± 27 7/16 (44%) 3/16 (19%) 0.75 ± 1.3 3.6 ± 0.72 259 ± 84 \07 ± 28
222 174/222(78%) 48/222 (22%) 62 ± II II ± 3.7 37/222 (17%) 68/222 (31%) 0.46 ± 1.2 3.3 ± 0.97 223 ± 47 97 ± 28
NS* 0.0001 0.0182 NS* NS* NS* NS* (0.1076) NS*
3/16 (19%) 1.8 ± 0.78
12/222 (5%) 1.9 ± 0.74
NS* NS*
40 ± 4.9 29 ± 5.4 4.0 ± 0.55 25 ± 65
41 ± 4.3 29 ± 3.6 4.2 ± 0.42 4.0 ± 6.3
NS* NS* NS* 0.0001
14 ± 31 3500 ± 9200 12 ± 24 7.2 ± 5.2 3.3 ± 4.3
2.7 ± 4.1 530 ± 830 2.2 ± 3.4 4.1 ± 1.0 0.6 ± 0.8
0.0001 0.0001 0.0001 0.0001 0.0001
0.0226
The values shown are mean ± standard deviation unless otherwise noted. Differences in proportions were determined by x 2, with continuity correction if appropriate. The other p values were determined by a two-sided t test and confirmed with a nonparametric test. Duration of operation was significantly different only using the t test, but not the Mann- Whitney U nonparametric equivalent. *NS = Not significant (p> 0.10).
important variables such as smoking, age, number of preoperative days in intensive care, number of grafts, duration of the operation, and duration of extracorporeal circulation did not differ significantly between the patients with and without infections. There were no significant differences in preoperative lymphocytes, monocytes, neutrophils, or total white count (only data for lymphocytes are shown). Because we were specifically interested in the role transfusion might play in causing postoperative infection, we examined the clinical and laboratory variables associated with transfusion (Table II). Patients receiving transfusions were significantly more likely to be older (65 years versus 56 years in the nontransfused group), female (32% versus 4%), to have greater numbers of grafts (3.5 versus 3.1), longer operations (234 minutes versus 206), longer times on extracorporeal support (102 minutes versus 87), lower preoperative hematocrit values (40.3% versus 43.9%), and lower serum albumin values (4.1 gm/L versus 4.3). They were significantly less likely to be smokers (26% versus 40%) than patients not receiving transfusions. Although the patients receiving transfusions had higher rates of infection (8.3% versus 2.9%), more days
of antibiotics (4.3 versus 4.1), and more days of fever (0.9 versus 0.6), these differences were not significant when transfusion was treated as a dichotomous variable (i.e., yes/no). The length of stay of patients receiving transfusions was significantly longer than that of the other group (12.8 versus 10.2 days). Because the use of a continuous variable (i.e., dose) is a more sensitive measure of association than use of a dichotomous categoric variable (transfusion versus no transfusion), we proceeded to a multivariate modeling of transfusion dose as a predictor of infection and its related variables. In initial multivariate logistic regressions, we found that preoperative anemia was not a significant predictor of infection (p > 0.40). Thus this potential confounding variable did not explain the relationship between transfusion and postoperative infection. The following is the reasoning we used in selecting other variables in addition to transfusion dose for multivariate analysis. Only one factor, sex, was significantly variable between both the infection/no infection groups and the transfusion/no transfusion groups. In addition, need for reexploration is a variable that is confounded with transfusion, because reoperation is usually for bleeding, which would lead to
Volume 104 Number 4 October 1992
Infection after CABG operations
I 09 5
Table II. Characteristics ofpatients receiving transfusions as compared with patients not receiving transfusions Nontransfused
p Value
Variable
Transfused
No. of patients Male Female Age (yr) Length of stay (days) Diabetes Smoking Days of preop. leu No. of grafts Duration of operation (min) Time on extracorporeal circulation (min) Rexploration required Infection Total lymphocyte count (thousand / ILl) Preop. hematocrit value (%) Discharge hematocrit value (%) Preop. albumin (gm/dl) Days of antibiotics Days of fever
168 115/168 (68%) 53/168 (32%) 65 ± 9.7 12.8 ± 9.8 32/168 (19%) 43/168 (26%) 0.50 ± 1.26 3.5 ± 0.9 234 ± 52 102 ± 28
70 67/70 (96%) 3/70 (4%) 56 ± 9.8 10.2 ± 2.8 12/70 (17%) 28/70 (40%) 0.44 ± 1.25 3.1 ± 1.1 206 ± 41 87 ± 26
0.0001 0.0282 NS* 0.0269 NS* 0.0196 0.0001 0.0002
14/168 (8%) 14/168 (8.3%) 1.85 ± 0.74
1/70 (1%) 2/70 (2.9%) 2.02 ± 0.75
NS* NS* NS*
40.3 ± 4.3 29.2 ± 3.7 4.1 ± 0.46 4.3 ± 2.0 0.9 ± 1.7
43.9 ± 3.4 29.0 ± 3.8 4.3 ± 0.30 4.1 ± 1.1 0.6 ± 0.8
0.0001 NS* 0.0015 NS* NS*
0.0001
The values shown are mean ± standard deviation unless otherwise noted. Differences in the proportions were determined by X2 , with continuity correction if appropriate. The other p values were determined by a two-sided I test and confirmed by a nonparametric test. *NS = Not significant (p > 0.10).
increased numbers of transfusions. Reexploration was sufficiently uncommon that differences between the patients with and without infections did not achieve statistical significance at conventional levels (19% in the infected group versus 5% in the noninfected group; p = 0.11). We chose to include need for reexploration in our multivariate models, understanding that it might be acting as a covariate of transfusion dose. Duration of the operation, extracorporeal circulation time ("time on bypass"), and age are also potentially confounded variables with transfusion that were tested in multivariate models despite their lack of significant differences in patients with versus patients without infections. This was to confirm the lack of significance of these important covariates. Also, diabetes was significantly more common in patients having infections but not in those having transfusions. Because diabetes has been shown to be a prognostic factor in infection in many previous studies of both surgical and medical patients, we chose to include diabetes in all the models as well. Thus a total of eight variables (sex, age, diabetes, number of grafts, duration of the operation, time on bypass, and reoperation) were chosen in addition to blood transfusion (red cell dose) for multivariate analysis as predictors of infection, length of hospital stay, days of antibiotics, and days of fever. The results of these initial multivariate logistic and linear regressions are shown in the first column of Appendix
Tables I to 4. As expected, many of these variables, which did not differ between patients with and without infections, were not significant as predictors of infection or other end points. We created a final multivariate model for each regression by removing the least significant factors from the model, one by one, until only variables with p < 0.05 remained. Only red cell dose was a significant predictor of all four measures of morbidity: infection, days of hospital stay, days of antibiotics, and days offever. The only variable that was significant in at least two regressions was gender (female sex was associated with greater numbers of infections and longer hospital stays). Examination of the regression coefficients and p values makes it apparent that transfusion dose is the most important predictor of infection and morbidity in this cohort of patients. Finally, to rule out the possibility that a few patients with massive transfusion were skewing the results, regression models were created, which eliminated patients who had received more than 6 units of blood. Only red cell dose was a significant predictor of length of stay or days of antibiotics in these models. The models for predicting infection and days of fever did not achieve significance. Last, multivariate logistic and linear regressions were performed in which the only variables used were those that differed significantly among the patients with and without infections: red cell dose, sex, and diabetes (Table
1096
The Journal of Thoracic and Cardiovascular Surgery
Murphy et al.
Table III. Clinical characteristics ofpatients who had postoperative infections Days of
Reexplored?
Length of stay (days)*
4 38
No No
9 114
Superficial sternal wound UTI, septicemia, abscess of leg wound
F F
7 5
No No
9 12
61 80 67 64 66
M F M F M
2 10 3 37 4
No No No No Yes
10 31 17 21 22
Superficial Sternal wound with septicemia Superficial sternal wound Superficial sternal wound Deep sternal wound Superficial sternal wound Pneumonia with septicemia
57 63 63 62
F F M M
3 9 0 0
Yes No No No
16 19 12 9
Superficial sternal wound
63 76
M M F
2 263 18
No Yes No
9 41 53
UTI
Age (yr)
Sex
70 79
M F
58 56
72
No. of components
Type of infection
UTI UTI Deep sternal wound with septicemia Empyema
UTI
Culture results
antibiotics"
Staph. aureus Streptococcus and Candida albicans (UTI), Staph. aureus (septicemia), Pseudomonas (abscess) Staph. aureus Staph. aureus
5 19
Staph. aureus Staph. aureus Staph. aureus Enterobacter, Aerogenes Hemophilus influenza (sputum); E. coli (blood) E. coli Enterococcus Enterococcus, Acinetobacter Staph. aureus
7 14 4 4 15
Enterobacter cloacae Enterococcus, Pseudomonas Candida albicans
3 NA 12
3 4
4 6 4 4
UTI, Urinary tract infection; NA, not available. 'Length of stay does not include readmissions. Days of antibiotics includes only antibiotics given during the initial hospitalization. Antibiotics given as an outpatient or during readmission are not included.
Table IV. Relationship between number of units of red cells and whole blood transfused and prevalence of culture-proved postoperative infection Proportion infected Units transfused
o 1 2
3 4
5 ~6
No. 2 of 1 of 3 of 2 of I of 1 of 6 of
76* 32 45 26 19 13 27
%
2.6 3.1 6.7 7.7 5.3 7.7 22.2
'Six patients received at least I unit of fresh frozenplasma (5) or 6 units ofplatelets (I) but no red cellsor whole blood. None of these six patients had an infection.
I). This eliminated, before regression analysis, variables for which there were no significant differences between patients with and without infections (e.g., number of grafts, need for reoperation, duration of operation, time on bypass, and age). Sex, diabetes, and red cell dose were statistically significant as predictors of infection (regression results not shown) (p < 0.05). When only red cell dose and either sex or diabetes were included in the model, each ofthese factors remained significant as predictors of infection. Similar results were found when total blood
components, white cell dose, or plasma dose was substituted for red cell dose. By means of multivariate linear regressions, the predictive roles of sex, diabetes, and red cell dose for length of stay, days of antibiotics, and days of fever were also examined. Only red cell dose was uniformly statistically significant as a predictor of length of stay, days of antibiotics, and days of fever. When patients receiving more than 6 units of red cells were excluded from the regressions that used sex, diabetes, and red cell dose, only diabetes and transfusion dose remained significant as predictors of infection. The six infections occurring in the 27 patients receiving 6 units or more were all in female patients; thus female sex and transfusion dose are confounded and not separable in this subgroup. The rate of infection in female patients receiving less than 6 units was 2 of 50 (4%), which suggests that female sex may not itself be an independent risk factor for infection. Diabetes and transfusion dose were not confounded variables. Diabetic patients receiving 2 units of blood or more had an infection rate of 26% (n = 27) as compared with diabetic patients receiving 0 or I unit, who had no infections (0/17). Nondiabetic patients receiving 2 units or more had an infection rate of 6% (n = 103) versus 3% (n = 91) in those receiving 0 or I unit of blood. Nondiabetic patients receiving 6 units or more had an infection
Volume 104 Number 4 October 1992
rate of 16% (n = 25). Thus the effects oftransfusion dose and diabetes appear independent and perhaps cumulative. The relative risk ratios for infection were calculated as 3.5 for diabetes (95% confidence interval [1.2; 7.9]); 3.2 for female sex ( 1.2; 8.9), 3.6 for 2 units of blood transfused or more (1.0; 27), and 8.0for6 units of blood or more (2.2; 29). Discussion Postoperative infection is a significant problem in patients undergoing CABG and other cardiac surgical procedures, although serious infections are fortunately not common. Several studies have evaluated this group of patients to identify the risk factors for postoperative infection. 12-17 These potential risk factors include repeat thoracotomy, diabetes mellitus, length of hospital stay greater than 5 days before the operation, current cigarette smoking, obesity, duration of the operation, use of bilateral internal mammary arteries, age less than 2 years or greater than or equal to 65 years, and blood transfusion as a continuous variable. In this current retrospective study we examined the medical records of only those patients undergoing firsttime CABG operations. By limiting the study to a relatively short time span (I year) and by studying only patients of a single surgeon, we attempted to minimize differences in surgical technique, indications used for transfusion, anesthetic agents, and other perioperative factors. Culture-confirmed infections occurred in 6.7% of all patients. This figure includes infections away from the operative sites (44% of all infections). Our data demonstrate that homologous transfusion is associated with an increased risk of infection in patients undergoing CABG and appears to do so in a dose-dependent fashion. Although the number of infections studied is small, the highly significant relationship seen between transfusion dose and infection is unlikely to be coincidental and is consonant with findings in other postoperative settings. Interestingly, this dose-response relationship is seen whether looking at total number of components transfused, total red cell, plasma, or white cell dose. Thus, if this relationship is causal, our data do not permit hypothesizing which component of the transfused blood is most important in mediating posttransfusion immunomodulation in this setting. It is possible that all major components of transfused blood can impair host defenses against bacterial infection. When several factors that are significantly associated with postoperative infection by univariate analysis are subjected to multivariate analysis, transfusion dose is found to be the most significant predictor of infection,
Infection after CABG operations
1097
length of hospital stay, days on antibiotics, and days of fever. Other previously identified important variables, such as smoking, age, and sex, are significant predictors of infection in some models but are of much lesser predictive value than transfusion dose, at least in our cohort of patients. Diabetes was the one other clearly independent and significant predictor of postoperative infection. There were no significant relationships between infection and duration of operation, preoperative anemia, or extracorporeal circuit time, factors for which transfusion dose might be acting as a covariate. Transfusion dose was highly significant as a predictor of infection and of approximately equal importance to diabetes as a prognostic factor. This dose-response relationship between transfusion and infection is unexplained by previously reported prognostic factors for postoperative infection. The relationship between transfusion and increased rates of postoperative infection has also been noted in varying degrees in studies of patients with penetrating abdominal trauma' fractures.? and thermal injuries, 19 as well as in those undergoing operations for colorectal cancers,? Crohn's disease.t and total hip replacement.l? As previously reported," patients receiving only a single unit of red cells do not appear to have a significantly elevated risk of postoperative infection. Recipients of 2 units of blood or more have between a threefold and eightfold increased prevalence of infection in our study, and this is similar to that previously reported. The mechanism of this association is uncertain. However, it seems reasonable to hypothesize that the relationship between transfusion and postoperative infection is due in part to the decreases in immune function known to occur after homologous blood transfusions.P Such immunosuppressive changes include an increase in numbers of suppressor T-lymphocytes, decrease in numbers of helper T -lymphocytes, decrease in natural killer cell function, decrease in function of macrophages and monocytes, reduced responses in mixed lymphocyte culture, and decreases in cytokine production. Decreases in interleukin-2 production correlated with transfusion have been described in patients having cardiac operations." Posttransfusion immunosuppressive effects may also explain clinical consequences of homologous blood transfusion such as increased renal allograft survival in those receiving pretransplant transfusion, and, possibly, increased rates of solid tumor recurrence.v 11.20 In summary, we found the amount of homologous blood transfused to be a significant and apparently independent predictor of postoperative infection in patients undergoing CABG operations. We hypothesize that this is due in part to the immunosuppressive effects of homologous blood transfusions. Studies to further characterize this association immunologically and clinically would
The Journal of
1098
Murphy et al.
Thoracic and Cardiovascular Surgery
seem well worthwhile. Interventions to minimize this proposed effect, such as greater use of autologous transfusions, merit consideration. 10 REFERENCES 1. Opelz G, Sengar DPS, Mickey MR, et al. Effect of blood transfusions on subsequent kidney transplants. Transplant Proc 1973;5:253-9. 2. Blumberg N, Heal JM, Murphy P, et al. Association between transfusion of whole blood and recurrence of cancer. Br Med J 1986;293:530-3. 3. Williams JG, Hughes LE. Effect of perioperative blood transfusion on recurrence of Crohn's disease. Lancet 1989; 2:131-3. 4. Blumberg N, Heal JM. Evidence for plasma-mediated immunomodulation-transfusions of red cells are associated with a lower risk of AIDS than transfusions of plasma-rich blood components. Transplant Proc 1988;20:1138-42. 5. Dellinger EP, Oreskovich MR, Wertz MJ, et al. Risk of infection following laparotomy for penetrating abdominal injury. Arch Surg 1984;119:20-7. 6. Dawes LG, Aprahamian C, Condon RE, et al. The risk of infection after colon injury. Surgery 1986;100:796-801. 7. Tartter PI. Blood transfusion and infectious complications followingcolorectal cancer surgery. Br J Surg 1988;75:78992. 8. Tartter PI, Driefuss RM, Malon AM, et al. Relationship of postoperative septic complications and blood transfusions in patients with Crohn's disease. Am J Surg 1988;155:43-7. 9. Dellinger EP, Miller SD, Wertz MJ, et al. Risk of infection after open fracture of the arm or leg. Arch Surg 1988; 123:1320-7. 10. Murphy P, Heal JM, Blumberg N. Infection or suspected infection after hip replacement surgery with autologous or homologous blood transfusions. Transfusion 1991;31: 212-7.
II. Blumberg N, Triulzi DJ, Heal JM. Transfusion-induced immunomodulation and its clinical consequences.Transfusion Med Rev 1990;4(suppl 1):24-35. 12. Ottino G, DePaulis R, Pansini S, et al. Major sternal wound infection after open-heart surgery: a multivariate analysis of risk factors in 2,579 consecutive operative procedures. Ann Thorac Surg 1987;44:173-9. 13. Miholic J, Hudec M, Domanig E, et al. Risk factors for severe bacterial infections after valve replacement and aortocoronary bypass operations: analysis of 246 cases by logistic regression. Ann Thorac Surg 1985;40:224-8. 14. Nagachinta T, Stephens M, Reitz B, et al. Risk factors for surgical-wound infection followingcardiac surgery. J Infect Dis 1987;156:967-73. 15. Simchen E, Shapiro M, Marin G, et al. Risk factors for post-operative wound infection in cardiac surgery patients. Infect Control 1983;4:215-20. 16. Shuhaiber H, Chugh T, Portaian-Shuhaiber S, et a1. Wound infection in cardiac surgery. J Cardiovasc Surg 1987;28:139-42. 17. Loop F, Lytle B, Cosgrove DM, et a1. Sternal wound complications after isolated coronary artery bypass grafting: early and late mortality, morbidity, and cost of care. Ann Thorac Surg 1990;49:179-87. 18. Haley R, Schaberg D, McClish D, et a1. The accuracy of retrospective chart review in measuring nosocomial infection rates-results of validation studies in pilot hospitals. Am J EpidemioI1980;111:516-33. 19. Graves A, Cioffi W, Mason A Jr, et a1. Relationship of transfusion and infection in a burn population. J Trauma 1989;29:948-52. 20. Blumberg N, Heal J. Transfusion and recipient immune function. Arch Pathol Lab Med 1989;113:246-53. 21. Hisatomi K, Isomura T, Kawara T, et a1. Changes in lymphocyte subsets, mitogen responsiveness, and interleukin-2 production after cardiac operations. J THORAC CARDIOVASC SURG 1989;98:580-91.
Appendix Table 1. Results of logistic regressionsfor variables predicting infection
Variable
Initial model (coefficient ±SEM)
p Value
Final model (coefficient ±SEM)
p Value
±0.315 ± 0.03 ± 0.32 ± 0.55 ± 0.009 ± 0.03 ± 0.45 ± 0.061
0.025 0.84 0.13 0.06 0.30 0.11 0.20 0.12
-0.60 ± 0.28
0.032
Age Diabetes No. of grafts Duration of operation Time on bypass Need for reoperation Red cell dose
-0.71 -0.006 0.48 -1.03 -0.009 0.041 0.58 -0.10
-0.09 ± 0.038
0.017
Overall model
R2 (U)
= 0.217
0.0013
R2 (U)
Sex
= 0.145
0.0002
Model limited to :5,6 units (coefficient ±SEM)
Model not significant (p = 0.10)
p Value
R2 (U)
= 0.15
An initial model was created with the use of all eight variables. Then factors with p > 0.05 were eliminated from the model one by one until only factors with < 0.05 were left, yielding the final model. A third model restricted to patients receiving 2:6 units of red cells or whole blood was created but was not significant as a predictor of infection. SEM, Standard error of the mean.
p
Volume 104 Number 4 October 1992
Infection after CABG operations
1 09 9
Appendix Table 2. Results of linear regressions for variables predicting length of hospital stay
Variable Sex Age Diabetes No. of grafts Duration of operation Time on bypass Need for reoperation Red cell dose Overall model
Initial model (coefficient ±SEM) 2.3 0.059 -1.5 0.63 -0.024 -0.005 2.7 0.64
± ± ± ± ± ± ± ±
1.2 0.047 1.2 0.79 0.015 0.03 2.0 0.084
R2= 0.30
p Value
0.052 0.21 0.23 0.43 0.10 0.88 0.18 0.0001 0.0001
Final model (coefficient ±SEM)
p Value
2.8 ± 1.1
0.013
0.57 ± 0.066 R2= 0.27
Model limited to s;6 units (coefficient
±SEM)
p Value
0.0001
0.42 ± 0.13
0.0019
0.0001
R 2= 0.043
0.0019
An initial model was created with the use of all eight variables. Then factors with p > 0.05 were eliminated from the model one by one until only factors with p < 0.05 were left, yielding the final model. A third model restricted to patients receiving ::s6 units of red cells or whole blood was created, but only the final model after eliminating variables with p > 0.05 is shown. SEM, Standard error of the mean.
Appendix Table 3. Results of linear regressions for variables predicting days of antibiotics
Variable Sex Age Diabetes No. of grafts Duration of operation Time on bypass Need for reoperation Red cell dose Overall model
Initial model (coefficient ±SEM) 0.014 0.008 0.068 0.091 -0.006 -0.0002 -0.65 0.17
± ± ± ± ± ± ± ±
0.28 0.011 0.287 0.184 0.003 0.008 0.47 0.032
R 2 = 0.16
p Value
0.96 0.46 0.81 0.62 0.075 0.98 0.17 0.0001 0.0001
Final model (coefficient ±SEM)
-0.005 ± 0.003
0.18 ± 0.028 2
R = 0.14
Model limited to s;6 units (coefficient p Value
±SEM)
p Value
0.12 ± 0.051
0.019
0.031
0.0001
2
0.0001
R = 0.025
0.019
An initial model was created with the use of all eight variables. Then factors with p > 0.05 were eliminated from the model one by one until only factors with < 0.05 were left, yielding the final model. A third model restricted to patients receiving ::s6 units of red cells or whole blood was created, but only the final model after eliminating variables with p> 0.05 is shown. SEM, Standard error of the mean.
p
Appendix Table 4. Results of linear regressions for variables predicting days offever
Variable Sex Age Diabetes No. of grafts Duration of operation Time on bypass Need for reoperation Red cell dose Overall model
Initial model (coefficient ±SEM) 0.054 -0.01 -0.15 0.4 -0.003 -0.011 -0.03 0.14
± 0.20 ± 0.008 ± 0.21 ± 0.14 ± 0.003 ± 0.006 ± 0.35 ± 0.Ql5
R2= 0.34
p Value
0.79 0.21 0.47 0.004 0.17 0.055 0.93 0.0001 0.0001
Final model (coefficient ±SEM)
0.42 ± 0.13 -0.016 ± 0.005
Model limited to s;6 units (coefficient p Value
±SEM)
p Value
0.002 0.002
0.13 ± 0.013
0.0001
R2 = 0.33
0.0001
Overall model not significant (p = 0.09)
An initial model was created with the use of all eight variables. Then factors with p > 0.05 were eliminated from the model one by one until only factors with p < 0.05 were left, yielding the final model. A third model restricted to patients receiving ::s6 units of red cells or whole blood was created but was not significant as a predictor of days of fever. SEM, Standard error of the mean.
Paul J. Murphy, MD,a Cliff Connery, MD,b George L. Hicks, Jr., MD,b and Neil Blumberg, MD,a Rochester, N.Y.
Extensive data suggest that homologous blood transfusions have immunosuppressive effects on recipients. Transfusion-induced immunomodulation is believed responsible for increased renal allograft survival in patients who receive pretransplantation transfusion. 1 More recent studies suggest there may be other clinical consequences of blood transfusion-induced immunomodulation, including increased rates of solid tumor recurrence after resection.i decreased recurrence rates of From the Transfusion Medicine Unit; Department of Pathology and Laboratory Medicine, and Division of Cardiothoracic Surgery," Department of Surgery, University of Rochester Medical Center, Rochester, N.Y. Supported in part by a research grant from the American Association of Blood Banks Foundation, Arlington, Va. Received for publication Sept. 20, 1990. Accepted for publication Sept. 16, 1991. Address for reprints: Neil Blumberg, MD, University of Rochester Medical Center, Box 608, 601 Elmwood Ave., Rochester, NY 14642.
12/1/33818
1092
Crohn's disease.' and increased severity of viral infection." Other studies show an increased risk of postoperative bacterial infection in patients receiving homologous blood transfusions after abdominal or orthopedic operations."!' Few clinical studies have been designed specifically to examine the relationship between transfusion and postoperative infection in those undergoing cardiac operations. One large retrospective study'? examined patients undergoing a variety of cardiac procedures and found that blood transfusion was significantly associated with an increased risk of postoperative infection. Another, smaller study' ' found a similar association in those patients undergoing valve replacement but not in those who underwent coronary artery bypass grafting (CABG). The purpose of the present study is to assess further whether homologous blood transfusion is an independent risk factor for postoperative infection in those undergoing CABG.lnasmuch as the possibility exists that transfusion acts merely as a surrogate measure of other clinical variables known to predict infection, we designed the study to
Volume 104 Number 4 October 1992
address confounding variables such as length of the operation and preadmission anemia.
Methods We retrospectively reviewed the medical records of all patients who underwent first-time CABG in our institution from June 1988toJune 1989. These patients were treated by a single cardiothoracic surgeon (G.L.H.). The variables examined include those previously reported to be risk factors for wound infectionafter cardiac surgery-v!": age; sex;presence or absence of diabetes mellitus; smoking history; length of hospital stay; days of preoperative hospital stay including intensive care unit stay; number of coronary arteries bypassed; whether an internal mammary artery was used; duration of the operation; duration of extracorporeal circulation; whether reoperation was necessary; total white blood cell count; absolute numbers of neutrophils,lymphocytes,and monocytes; hematocrit values on admission and discharge; admission serum albumin; transfusionsduring the perioperative period; therapeutic use of antibiotics (all patients received postoperative antibiotic prophylaxis for 3 or 4 days); and days of temperatures above 38.30 C (\01 0 F).
Criteria for infection included strong clinical suspicion of infection by the attending surgeon, the institution of antibiotic therapy, and confirmation by a positive culture. When more liberal criteria for suspected infection were used," the results were identical to those reported when a positive culture was required for confirmation. Patient follow-upincluded reviewing the outpatient record for two office visits to the attending surgeon,both of which usually took place within I month after discharge from the hospital. A total of 248 consecutive patients were reviewed. Ten patients were excluded from the study because of preexisting bacterial infectionsat the time of hospital admission, for which they were treated with antibiotics (seven cases of urinary tract infection, two of pneumonia, and one of prostatitis). This left 238 patients evaluated for this study. Complete data were available for 234 of them, with one missing datum for each of the four patients with incomplete data. Because immunosuppression may be due to red cell, white cell,or plasma content of the transfused blood components, we used the following values to calculate the approximate red cell dose, white cell dose, and plasma dose received by each patient having a transfusion: (I) Red cells: Each unit of packed red bloodcellsor whole blood was considered to have 200 ml of red cells; a value of zero was used for platelets and fresh frozen plasma. (2) White cells: A white cell content of I X 109 cellsper unit of wholeblood or packed red cells and 0.4 X 109 white cells per unit of platelet concentrate was used. (3) Plasma: The plasma content was 300 ml for each unit of whole blood, 50 ml for a unit of packed red cells, 50 ml for a unit of platelet concentrate, and 250 ml for a unit of fresh frozen plasma. We also quantitated transfusion simply as number of components received; that is,each unit of red cells,platelet concentrate, fresh frozen plasma, and whole blood was counted as I unit of blood. Both parametric and nonparametric methods of data analysis were used, with JMP Software (SAS Institutes, Inc., Cary, N.C.) or Stat View SE + Graphics (Abacus Concepts, Inc., Berkeley, Calif.) on a Macintosh Ilcx microcomputer (Apple Computer, Cupertino, Calif.). Parametric results are generally reported (e.g., analysis of variance or t test), but these were alwaysconfirmedby a nonparametric equivalent when available
Infection after CABG operations
1093
(e.g., Kruskal-Wallis or Mann-Whitney). Risk ratios were calculated for factors significantly associated with postoperative infection by dividing the prevalence of infection in the high risk subgroup by the prevalence in the lower risk subgroup. N inetyfive-percent confidencelimits were calculated for risk ratios for single factors by the normal deviate and the x2 distribution. Because of the small number of infections in any single risk factor subgroup, attributable risk for each factor was not calculated. For regression analyses, no transformations ofvariables were used.
Results Of the 238 patients, 16 (6.7%) had a culture-proved postoperative infection during the inpatient or postdischarge period. We first analyzed these infections by a univariate approach to determine which clinical or laboratory variables were associated with infection (Table I). Length of stay, days of fever, and days of antibiotics were documented because of their potential utility as surrogate measures of infection and as an indication of morbidity and expense associated with infection. Because transfusion was one of the variables significantly different between the patients who had infections and those who did not (Table I), we also performed a similar univariate analysis to determine which clinical and laboratory variables were significantly different between those receiving or not receiving transfusion (Table II). The details of the patients with infection are shown in Table III. Only 3 of 238 (1.3%) patients had deep or severe sternal wound infections. There were a number of infections at sites other than the wounds, including pneumonia, empyema, and five urinary tract infections (7/16 infections, 44%). Two of these infections also led to generalized sepsis. Of these seven distant infections, six occurred in patients who had a transfusion. As can be seen from Table I, the patients having infections were significantly more likely to be female (50% versus 22% in the noninfected group), to be diabetic (44% versus 17%), and to have received significantly more blood transfusions (25 total components versus 4). In addition to being the quantitatively most striking difference between the patients with and without infections, transfusion appeared to be associated with an increased risk of infection in a dose-dependent manner. Infections were seen in 6 of 153 (3.9%) patients receiving up to 2 units of red cells or whole blood, 4 of 58 (6.9%) patients receiving 3 to 5 units, and 6 of 27 (22%) patients receiving 6 units or more (p = 0.0022 by x2 analysis) (Table IV). Other boundaries for grouping the red cell dose yielded the same results. Measures of infection-related expense and morbidity including length of stay, days of antibiotics, and fever, as expected, were significantly greater in the infected group than in the uninfected group. Interestingly, potentially
1094
The Journal of Thoracic and Cardiovascular Surgery
Murphy et al.
Table I. Characteristics ofpatients with culture-proved infection as compared with those without infection Variable No. of patients Male Female Age (yr) Length of stay (days) Diabetes Smoking Days of preop. rcu No. of grafts Duration of operation (min) Time on extracorporeal circulation (min) Reexploration required Total lymphocyte count (thousand / ILL) Preop. hematocrit value (%) Discharge hematocrit value (%) Preop. albumin (gm/dl) Total No. of blood components transfused WBC dose (x \09) Plasma dose (m!) Red cell dose (units) Days of antibiotics Days of fever
Infection
No infection
p Value
16 8/16 (50%) 8/16 (50%) 66 ± 7.5 25 ± 27 7/16 (44%) 3/16 (19%) 0.75 ± 1.3 3.6 ± 0.72 259 ± 84 \07 ± 28
222 174/222(78%) 48/222 (22%) 62 ± II II ± 3.7 37/222 (17%) 68/222 (31%) 0.46 ± 1.2 3.3 ± 0.97 223 ± 47 97 ± 28
NS* 0.0001 0.0182 NS* NS* NS* NS* (0.1076) NS*
3/16 (19%) 1.8 ± 0.78
12/222 (5%) 1.9 ± 0.74
NS* NS*
40 ± 4.9 29 ± 5.4 4.0 ± 0.55 25 ± 65
41 ± 4.3 29 ± 3.6 4.2 ± 0.42 4.0 ± 6.3
NS* NS* NS* 0.0001
14 ± 31 3500 ± 9200 12 ± 24 7.2 ± 5.2 3.3 ± 4.3
2.7 ± 4.1 530 ± 830 2.2 ± 3.4 4.1 ± 1.0 0.6 ± 0.8
0.0001 0.0001 0.0001 0.0001 0.0001
0.0226
The values shown are mean ± standard deviation unless otherwise noted. Differences in proportions were determined by x 2, with continuity correction if appropriate. The other p values were determined by a two-sided t test and confirmed with a nonparametric test. Duration of operation was significantly different only using the t test, but not the Mann- Whitney U nonparametric equivalent. *NS = Not significant (p> 0.10).
important variables such as smoking, age, number of preoperative days in intensive care, number of grafts, duration of the operation, and duration of extracorporeal circulation did not differ significantly between the patients with and without infections. There were no significant differences in preoperative lymphocytes, monocytes, neutrophils, or total white count (only data for lymphocytes are shown). Because we were specifically interested in the role transfusion might play in causing postoperative infection, we examined the clinical and laboratory variables associated with transfusion (Table II). Patients receiving transfusions were significantly more likely to be older (65 years versus 56 years in the nontransfused group), female (32% versus 4%), to have greater numbers of grafts (3.5 versus 3.1), longer operations (234 minutes versus 206), longer times on extracorporeal support (102 minutes versus 87), lower preoperative hematocrit values (40.3% versus 43.9%), and lower serum albumin values (4.1 gm/L versus 4.3). They were significantly less likely to be smokers (26% versus 40%) than patients not receiving transfusions. Although the patients receiving transfusions had higher rates of infection (8.3% versus 2.9%), more days
of antibiotics (4.3 versus 4.1), and more days of fever (0.9 versus 0.6), these differences were not significant when transfusion was treated as a dichotomous variable (i.e., yes/no). The length of stay of patients receiving transfusions was significantly longer than that of the other group (12.8 versus 10.2 days). Because the use of a continuous variable (i.e., dose) is a more sensitive measure of association than use of a dichotomous categoric variable (transfusion versus no transfusion), we proceeded to a multivariate modeling of transfusion dose as a predictor of infection and its related variables. In initial multivariate logistic regressions, we found that preoperative anemia was not a significant predictor of infection (p > 0.40). Thus this potential confounding variable did not explain the relationship between transfusion and postoperative infection. The following is the reasoning we used in selecting other variables in addition to transfusion dose for multivariate analysis. Only one factor, sex, was significantly variable between both the infection/no infection groups and the transfusion/no transfusion groups. In addition, need for reexploration is a variable that is confounded with transfusion, because reoperation is usually for bleeding, which would lead to
Volume 104 Number 4 October 1992
Infection after CABG operations
I 09 5
Table II. Characteristics ofpatients receiving transfusions as compared with patients not receiving transfusions Nontransfused
p Value
Variable
Transfused
No. of patients Male Female Age (yr) Length of stay (days) Diabetes Smoking Days of preop. leu No. of grafts Duration of operation (min) Time on extracorporeal circulation (min) Rexploration required Infection Total lymphocyte count (thousand / ILl) Preop. hematocrit value (%) Discharge hematocrit value (%) Preop. albumin (gm/dl) Days of antibiotics Days of fever
168 115/168 (68%) 53/168 (32%) 65 ± 9.7 12.8 ± 9.8 32/168 (19%) 43/168 (26%) 0.50 ± 1.26 3.5 ± 0.9 234 ± 52 102 ± 28
70 67/70 (96%) 3/70 (4%) 56 ± 9.8 10.2 ± 2.8 12/70 (17%) 28/70 (40%) 0.44 ± 1.25 3.1 ± 1.1 206 ± 41 87 ± 26
0.0001 0.0282 NS* 0.0269 NS* 0.0196 0.0001 0.0002
14/168 (8%) 14/168 (8.3%) 1.85 ± 0.74
1/70 (1%) 2/70 (2.9%) 2.02 ± 0.75
NS* NS* NS*
40.3 ± 4.3 29.2 ± 3.7 4.1 ± 0.46 4.3 ± 2.0 0.9 ± 1.7
43.9 ± 3.4 29.0 ± 3.8 4.3 ± 0.30 4.1 ± 1.1 0.6 ± 0.8
0.0001 NS* 0.0015 NS* NS*
0.0001
The values shown are mean ± standard deviation unless otherwise noted. Differences in the proportions were determined by X2 , with continuity correction if appropriate. The other p values were determined by a two-sided I test and confirmed by a nonparametric test. *NS = Not significant (p > 0.10).
increased numbers of transfusions. Reexploration was sufficiently uncommon that differences between the patients with and without infections did not achieve statistical significance at conventional levels (19% in the infected group versus 5% in the noninfected group; p = 0.11). We chose to include need for reexploration in our multivariate models, understanding that it might be acting as a covariate of transfusion dose. Duration of the operation, extracorporeal circulation time ("time on bypass"), and age are also potentially confounded variables with transfusion that were tested in multivariate models despite their lack of significant differences in patients with versus patients without infections. This was to confirm the lack of significance of these important covariates. Also, diabetes was significantly more common in patients having infections but not in those having transfusions. Because diabetes has been shown to be a prognostic factor in infection in many previous studies of both surgical and medical patients, we chose to include diabetes in all the models as well. Thus a total of eight variables (sex, age, diabetes, number of grafts, duration of the operation, time on bypass, and reoperation) were chosen in addition to blood transfusion (red cell dose) for multivariate analysis as predictors of infection, length of hospital stay, days of antibiotics, and days of fever. The results of these initial multivariate logistic and linear regressions are shown in the first column of Appendix
Tables I to 4. As expected, many of these variables, which did not differ between patients with and without infections, were not significant as predictors of infection or other end points. We created a final multivariate model for each regression by removing the least significant factors from the model, one by one, until only variables with p < 0.05 remained. Only red cell dose was a significant predictor of all four measures of morbidity: infection, days of hospital stay, days of antibiotics, and days offever. The only variable that was significant in at least two regressions was gender (female sex was associated with greater numbers of infections and longer hospital stays). Examination of the regression coefficients and p values makes it apparent that transfusion dose is the most important predictor of infection and morbidity in this cohort of patients. Finally, to rule out the possibility that a few patients with massive transfusion were skewing the results, regression models were created, which eliminated patients who had received more than 6 units of blood. Only red cell dose was a significant predictor of length of stay or days of antibiotics in these models. The models for predicting infection and days of fever did not achieve significance. Last, multivariate logistic and linear regressions were performed in which the only variables used were those that differed significantly among the patients with and without infections: red cell dose, sex, and diabetes (Table
1096
The Journal of Thoracic and Cardiovascular Surgery
Murphy et al.
Table III. Clinical characteristics ofpatients who had postoperative infections Days of
Reexplored?
Length of stay (days)*
4 38
No No
9 114
Superficial sternal wound UTI, septicemia, abscess of leg wound
F F
7 5
No No
9 12
61 80 67 64 66
M F M F M
2 10 3 37 4
No No No No Yes
10 31 17 21 22
Superficial Sternal wound with septicemia Superficial sternal wound Superficial sternal wound Deep sternal wound Superficial sternal wound Pneumonia with septicemia
57 63 63 62
F F M M
3 9 0 0
Yes No No No
16 19 12 9
Superficial sternal wound
63 76
M M F
2 263 18
No Yes No
9 41 53
UTI
Age (yr)
Sex
70 79
M F
58 56
72
No. of components
Type of infection
UTI UTI Deep sternal wound with septicemia Empyema
UTI
Culture results
antibiotics"
Staph. aureus Streptococcus and Candida albicans (UTI), Staph. aureus (septicemia), Pseudomonas (abscess) Staph. aureus Staph. aureus
5 19
Staph. aureus Staph. aureus Staph. aureus Enterobacter, Aerogenes Hemophilus influenza (sputum); E. coli (blood) E. coli Enterococcus Enterococcus, Acinetobacter Staph. aureus
7 14 4 4 15
Enterobacter cloacae Enterococcus, Pseudomonas Candida albicans
3 NA 12
3 4
4 6 4 4
UTI, Urinary tract infection; NA, not available. 'Length of stay does not include readmissions. Days of antibiotics includes only antibiotics given during the initial hospitalization. Antibiotics given as an outpatient or during readmission are not included.
Table IV. Relationship between number of units of red cells and whole blood transfused and prevalence of culture-proved postoperative infection Proportion infected Units transfused
o 1 2
3 4
5 ~6
No. 2 of 1 of 3 of 2 of I of 1 of 6 of
76* 32 45 26 19 13 27
%
2.6 3.1 6.7 7.7 5.3 7.7 22.2
'Six patients received at least I unit of fresh frozenplasma (5) or 6 units ofplatelets (I) but no red cellsor whole blood. None of these six patients had an infection.
I). This eliminated, before regression analysis, variables for which there were no significant differences between patients with and without infections (e.g., number of grafts, need for reoperation, duration of operation, time on bypass, and age). Sex, diabetes, and red cell dose were statistically significant as predictors of infection (regression results not shown) (p < 0.05). When only red cell dose and either sex or diabetes were included in the model, each ofthese factors remained significant as predictors of infection. Similar results were found when total blood
components, white cell dose, or plasma dose was substituted for red cell dose. By means of multivariate linear regressions, the predictive roles of sex, diabetes, and red cell dose for length of stay, days of antibiotics, and days of fever were also examined. Only red cell dose was uniformly statistically significant as a predictor of length of stay, days of antibiotics, and days of fever. When patients receiving more than 6 units of red cells were excluded from the regressions that used sex, diabetes, and red cell dose, only diabetes and transfusion dose remained significant as predictors of infection. The six infections occurring in the 27 patients receiving 6 units or more were all in female patients; thus female sex and transfusion dose are confounded and not separable in this subgroup. The rate of infection in female patients receiving less than 6 units was 2 of 50 (4%), which suggests that female sex may not itself be an independent risk factor for infection. Diabetes and transfusion dose were not confounded variables. Diabetic patients receiving 2 units of blood or more had an infection rate of 26% (n = 27) as compared with diabetic patients receiving 0 or I unit, who had no infections (0/17). Nondiabetic patients receiving 2 units or more had an infection rate of 6% (n = 103) versus 3% (n = 91) in those receiving 0 or I unit of blood. Nondiabetic patients receiving 6 units or more had an infection
Volume 104 Number 4 October 1992
rate of 16% (n = 25). Thus the effects oftransfusion dose and diabetes appear independent and perhaps cumulative. The relative risk ratios for infection were calculated as 3.5 for diabetes (95% confidence interval [1.2; 7.9]); 3.2 for female sex ( 1.2; 8.9), 3.6 for 2 units of blood transfused or more (1.0; 27), and 8.0for6 units of blood or more (2.2; 29). Discussion Postoperative infection is a significant problem in patients undergoing CABG and other cardiac surgical procedures, although serious infections are fortunately not common. Several studies have evaluated this group of patients to identify the risk factors for postoperative infection. 12-17 These potential risk factors include repeat thoracotomy, diabetes mellitus, length of hospital stay greater than 5 days before the operation, current cigarette smoking, obesity, duration of the operation, use of bilateral internal mammary arteries, age less than 2 years or greater than or equal to 65 years, and blood transfusion as a continuous variable. In this current retrospective study we examined the medical records of only those patients undergoing firsttime CABG operations. By limiting the study to a relatively short time span (I year) and by studying only patients of a single surgeon, we attempted to minimize differences in surgical technique, indications used for transfusion, anesthetic agents, and other perioperative factors. Culture-confirmed infections occurred in 6.7% of all patients. This figure includes infections away from the operative sites (44% of all infections). Our data demonstrate that homologous transfusion is associated with an increased risk of infection in patients undergoing CABG and appears to do so in a dose-dependent fashion. Although the number of infections studied is small, the highly significant relationship seen between transfusion dose and infection is unlikely to be coincidental and is consonant with findings in other postoperative settings. Interestingly, this dose-response relationship is seen whether looking at total number of components transfused, total red cell, plasma, or white cell dose. Thus, if this relationship is causal, our data do not permit hypothesizing which component of the transfused blood is most important in mediating posttransfusion immunomodulation in this setting. It is possible that all major components of transfused blood can impair host defenses against bacterial infection. When several factors that are significantly associated with postoperative infection by univariate analysis are subjected to multivariate analysis, transfusion dose is found to be the most significant predictor of infection,
Infection after CABG operations
1097
length of hospital stay, days on antibiotics, and days of fever. Other previously identified important variables, such as smoking, age, and sex, are significant predictors of infection in some models but are of much lesser predictive value than transfusion dose, at least in our cohort of patients. Diabetes was the one other clearly independent and significant predictor of postoperative infection. There were no significant relationships between infection and duration of operation, preoperative anemia, or extracorporeal circuit time, factors for which transfusion dose might be acting as a covariate. Transfusion dose was highly significant as a predictor of infection and of approximately equal importance to diabetes as a prognostic factor. This dose-response relationship between transfusion and infection is unexplained by previously reported prognostic factors for postoperative infection. The relationship between transfusion and increased rates of postoperative infection has also been noted in varying degrees in studies of patients with penetrating abdominal trauma' fractures.? and thermal injuries, 19 as well as in those undergoing operations for colorectal cancers,? Crohn's disease.t and total hip replacement.l? As previously reported," patients receiving only a single unit of red cells do not appear to have a significantly elevated risk of postoperative infection. Recipients of 2 units of blood or more have between a threefold and eightfold increased prevalence of infection in our study, and this is similar to that previously reported. The mechanism of this association is uncertain. However, it seems reasonable to hypothesize that the relationship between transfusion and postoperative infection is due in part to the decreases in immune function known to occur after homologous blood transfusions.P Such immunosuppressive changes include an increase in numbers of suppressor T-lymphocytes, decrease in numbers of helper T -lymphocytes, decrease in natural killer cell function, decrease in function of macrophages and monocytes, reduced responses in mixed lymphocyte culture, and decreases in cytokine production. Decreases in interleukin-2 production correlated with transfusion have been described in patients having cardiac operations." Posttransfusion immunosuppressive effects may also explain clinical consequences of homologous blood transfusion such as increased renal allograft survival in those receiving pretransplant transfusion, and, possibly, increased rates of solid tumor recurrence.v 11.20 In summary, we found the amount of homologous blood transfused to be a significant and apparently independent predictor of postoperative infection in patients undergoing CABG operations. We hypothesize that this is due in part to the immunosuppressive effects of homologous blood transfusions. Studies to further characterize this association immunologically and clinically would
The Journal of
1098
Murphy et al.
Thoracic and Cardiovascular Surgery
seem well worthwhile. Interventions to minimize this proposed effect, such as greater use of autologous transfusions, merit consideration. 10 REFERENCES 1. Opelz G, Sengar DPS, Mickey MR, et al. Effect of blood transfusions on subsequent kidney transplants. Transplant Proc 1973;5:253-9. 2. Blumberg N, Heal JM, Murphy P, et al. Association between transfusion of whole blood and recurrence of cancer. Br Med J 1986;293:530-3. 3. Williams JG, Hughes LE. Effect of perioperative blood transfusion on recurrence of Crohn's disease. Lancet 1989; 2:131-3. 4. Blumberg N, Heal JM. Evidence for plasma-mediated immunomodulation-transfusions of red cells are associated with a lower risk of AIDS than transfusions of plasma-rich blood components. Transplant Proc 1988;20:1138-42. 5. Dellinger EP, Oreskovich MR, Wertz MJ, et al. Risk of infection following laparotomy for penetrating abdominal injury. Arch Surg 1984;119:20-7. 6. Dawes LG, Aprahamian C, Condon RE, et al. The risk of infection after colon injury. Surgery 1986;100:796-801. 7. Tartter PI. Blood transfusion and infectious complications followingcolorectal cancer surgery. Br J Surg 1988;75:78992. 8. Tartter PI, Driefuss RM, Malon AM, et al. Relationship of postoperative septic complications and blood transfusions in patients with Crohn's disease. Am J Surg 1988;155:43-7. 9. Dellinger EP, Miller SD, Wertz MJ, et al. Risk of infection after open fracture of the arm or leg. Arch Surg 1988; 123:1320-7. 10. Murphy P, Heal JM, Blumberg N. Infection or suspected infection after hip replacement surgery with autologous or homologous blood transfusions. Transfusion 1991;31: 212-7.
II. Blumberg N, Triulzi DJ, Heal JM. Transfusion-induced immunomodulation and its clinical consequences.Transfusion Med Rev 1990;4(suppl 1):24-35. 12. Ottino G, DePaulis R, Pansini S, et al. Major sternal wound infection after open-heart surgery: a multivariate analysis of risk factors in 2,579 consecutive operative procedures. Ann Thorac Surg 1987;44:173-9. 13. Miholic J, Hudec M, Domanig E, et al. Risk factors for severe bacterial infections after valve replacement and aortocoronary bypass operations: analysis of 246 cases by logistic regression. Ann Thorac Surg 1985;40:224-8. 14. Nagachinta T, Stephens M, Reitz B, et al. Risk factors for surgical-wound infection followingcardiac surgery. J Infect Dis 1987;156:967-73. 15. Simchen E, Shapiro M, Marin G, et al. Risk factors for post-operative wound infection in cardiac surgery patients. Infect Control 1983;4:215-20. 16. Shuhaiber H, Chugh T, Portaian-Shuhaiber S, et a1. Wound infection in cardiac surgery. J Cardiovasc Surg 1987;28:139-42. 17. Loop F, Lytle B, Cosgrove DM, et a1. Sternal wound complications after isolated coronary artery bypass grafting: early and late mortality, morbidity, and cost of care. Ann Thorac Surg 1990;49:179-87. 18. Haley R, Schaberg D, McClish D, et a1. The accuracy of retrospective chart review in measuring nosocomial infection rates-results of validation studies in pilot hospitals. Am J EpidemioI1980;111:516-33. 19. Graves A, Cioffi W, Mason A Jr, et a1. Relationship of transfusion and infection in a burn population. J Trauma 1989;29:948-52. 20. Blumberg N, Heal J. Transfusion and recipient immune function. Arch Pathol Lab Med 1989;113:246-53. 21. Hisatomi K, Isomura T, Kawara T, et a1. Changes in lymphocyte subsets, mitogen responsiveness, and interleukin-2 production after cardiac operations. J THORAC CARDIOVASC SURG 1989;98:580-91.
Appendix Table 1. Results of logistic regressionsfor variables predicting infection
Variable
Initial model (coefficient ±SEM)
p Value
Final model (coefficient ±SEM)
p Value
±0.315 ± 0.03 ± 0.32 ± 0.55 ± 0.009 ± 0.03 ± 0.45 ± 0.061
0.025 0.84 0.13 0.06 0.30 0.11 0.20 0.12
-0.60 ± 0.28
0.032
Age Diabetes No. of grafts Duration of operation Time on bypass Need for reoperation Red cell dose
-0.71 -0.006 0.48 -1.03 -0.009 0.041 0.58 -0.10
-0.09 ± 0.038
0.017
Overall model
R2 (U)
= 0.217
0.0013
R2 (U)
Sex
= 0.145
0.0002
Model limited to :5,6 units (coefficient ±SEM)
Model not significant (p = 0.10)
p Value
R2 (U)
= 0.15
An initial model was created with the use of all eight variables. Then factors with p > 0.05 were eliminated from the model one by one until only factors with < 0.05 were left, yielding the final model. A third model restricted to patients receiving 2:6 units of red cells or whole blood was created but was not significant as a predictor of infection. SEM, Standard error of the mean.
p
Volume 104 Number 4 October 1992
Infection after CABG operations
1 09 9
Appendix Table 2. Results of linear regressions for variables predicting length of hospital stay
Variable Sex Age Diabetes No. of grafts Duration of operation Time on bypass Need for reoperation Red cell dose Overall model
Initial model (coefficient ±SEM) 2.3 0.059 -1.5 0.63 -0.024 -0.005 2.7 0.64
± ± ± ± ± ± ± ±
1.2 0.047 1.2 0.79 0.015 0.03 2.0 0.084
R2= 0.30
p Value
0.052 0.21 0.23 0.43 0.10 0.88 0.18 0.0001 0.0001
Final model (coefficient ±SEM)
p Value
2.8 ± 1.1
0.013
0.57 ± 0.066 R2= 0.27
Model limited to s;6 units (coefficient
±SEM)
p Value
0.0001
0.42 ± 0.13
0.0019
0.0001
R 2= 0.043
0.0019
An initial model was created with the use of all eight variables. Then factors with p > 0.05 were eliminated from the model one by one until only factors with p < 0.05 were left, yielding the final model. A third model restricted to patients receiving ::s6 units of red cells or whole blood was created, but only the final model after eliminating variables with p > 0.05 is shown. SEM, Standard error of the mean.
Appendix Table 3. Results of linear regressions for variables predicting days of antibiotics
Variable Sex Age Diabetes No. of grafts Duration of operation Time on bypass Need for reoperation Red cell dose Overall model
Initial model (coefficient ±SEM) 0.014 0.008 0.068 0.091 -0.006 -0.0002 -0.65 0.17
± ± ± ± ± ± ± ±
0.28 0.011 0.287 0.184 0.003 0.008 0.47 0.032
R 2 = 0.16
p Value
0.96 0.46 0.81 0.62 0.075 0.98 0.17 0.0001 0.0001
Final model (coefficient ±SEM)
-0.005 ± 0.003
0.18 ± 0.028 2
R = 0.14
Model limited to s;6 units (coefficient p Value
±SEM)
p Value
0.12 ± 0.051
0.019
0.031
0.0001
2
0.0001
R = 0.025
0.019
An initial model was created with the use of all eight variables. Then factors with p > 0.05 were eliminated from the model one by one until only factors with < 0.05 were left, yielding the final model. A third model restricted to patients receiving ::s6 units of red cells or whole blood was created, but only the final model after eliminating variables with p> 0.05 is shown. SEM, Standard error of the mean.
p
Appendix Table 4. Results of linear regressions for variables predicting days offever
Variable Sex Age Diabetes No. of grafts Duration of operation Time on bypass Need for reoperation Red cell dose Overall model
Initial model (coefficient ±SEM) 0.054 -0.01 -0.15 0.4 -0.003 -0.011 -0.03 0.14
± 0.20 ± 0.008 ± 0.21 ± 0.14 ± 0.003 ± 0.006 ± 0.35 ± 0.Ql5
R2= 0.34
p Value
0.79 0.21 0.47 0.004 0.17 0.055 0.93 0.0001 0.0001
Final model (coefficient ±SEM)
0.42 ± 0.13 -0.016 ± 0.005
Model limited to s;6 units (coefficient p Value
±SEM)
p Value
0.002 0.002
0.13 ± 0.013
0.0001
R2 = 0.33
0.0001
Overall model not significant (p = 0.09)
An initial model was created with the use of all eight variables. Then factors with p > 0.05 were eliminated from the model one by one until only factors with p < 0.05 were left, yielding the final model. A third model restricted to patients receiving ::s6 units of red cells or whole blood was created but was not significant as a predictor of days of fever. SEM, Standard error of the mean.