The Dose-Response Relationship of Tranexamic Acid 

Patients aged 21–80 yr for elective cardiac operation performed by either of two surgeons (M.D.S. or K.E.G.) gave informed consent after institutional review board approval. Patients excluded from participation took warfarin or estrogens within 7 days of surgery, had active hematuria or a serum creatinine concentration of 2.0 mg *symbol* dl¹or more, had a personal or family history of abnormal bleeding, weighed more than 100 kg, or underwent intraaortic balloon counterpulsation before surgery. Patients receiving aspirin within 7 days of operation, heparin infusion within 8 h, or nonsteroidal antiinflammatory medication within 3 days participated. Inclusion of these patients permits application of the study results to a representative population of elective surgical patients. Enrolled patients underwent coronary revascularization, valve replacement, both procedures combined, or repair of atrial septal defect.

A table of random numbers determined patient allocation initially to one of five groups. The placebo group (P) received saline infusions. Four other groups received tranexamic acid beginning after induction of anesthesia but before skin incision as a loading dose over 30 min followed by a 12-h constant infusion. Table 1displays the different doses of tranexamic acid, which initially ranged from one-quarter (2.5 mg *symbol* kg sup -1 loading followed by 0.25 mg *symbol* kg sup -1 *symbol* h sup -1 infusion) to twice (20 mg *symbol* kg sup -1 loading followed by 2.0 mg *symbol* kg sup -1 *symbol* h sup -1) the recommended doses. [28] After enrollment of 60 patients without perioperative stroke or clinically evident deep vein thrombosis, a sixth group of patients receiving fourfold the recommended dose (40 mg *symbol* kg sup -1 loading followed by 4.0 mg *symbol* kg sup -1 *symbol* h sup -1) was added to the randomization scheme in such a way that each group would contain approximately equal numbers after a target enrollment of 155 patients. Power analysis predicted the need for approximately 25 patients per group to demonstrate a difference of at least 150 g chest tube drainage after operation between the placebo group and the group receiving the standard, recommended dose (10 mg *symbol* kg sup -1 followed by 1.0 mg *symbol* kg sup -1 *symbol* h sup -1). Coded infusion bags for both loading and infusion doses and sealed envelopes prepared by a pharmacist provided double-blinded conditions.

Beef lung heparin (Organon, West Orange, NJ), 400 units *symbol* kg sup -1, provided anticoagulation for ECC. Automated celite activated coagulation time (Hemochron, International Technidyne, Edison, NJ) greater than 480 s, determined in duplicate every 30 min, ensured continued anticoagulation. ECC employed nonocclusive roller pumps, membrane oxygenators (Maxima, Medtronic Cardiopulmonary, Anaheim, CA), cold sanguinous cardioplegic arrest, and systemic hypothermia to 25 degrees Celsius. The ECC circuit initially contained 5,000 U of beef lung heparin in 2 L of clear fluid prime (Plasmalyte A solution, Baxter Healthcare, Deerfield, IL). Termination of ECC required a distal esophageal temperature of 37 degrees Celsius or greater and urinary catheter temperature of 34 degrees Celsius or greater. Protamine (4 mg *symbol* kg sup -1) neutralized heparin to obtain an activated coagulation time within 15 s of baseline. [29] Infusion of residual blood from the extracorporeal circuit followed approximation of the sternal wound edges. A subsequent additional protamine infusion (30% of the initial dose) neutralized heparin in the pump blood and protected against heparin rebound, [30] as confirmed by repeat activated coagulation time measured 2 h after sternal closure.

The mass of blood collected via mediastinal and pleural drains for a period beginning with chest closure and lasting 12 h represented blood loss. No attempt was made to estimate blood loss before or during ECC or before insertion of mediastinal drainage tubes. The study ignored estimates of irrigation fluid, sponge and suction container losses, and soaking of linens; all of these measures of operative blood loss lack accuracy. All mediastinal blood lost after initial heparin administration until termination of ECC eventually was returned to the patient. After ECC, a citrated autotransfusion drainage system (Pleur-evac, Deknatel, Queens Village, NY) permitted return of shed blood to those patients experiencing initial rapid blood loss.

Transfusion of packed erythrocytes in the first 12 h after surgery required one or more of the following: hematocrit less than 21% measured at least 1 h after surgery, chest tube drainage of at least 250 ml *symbol* h sup -1, or a hematocrit less than 24% associated with hemodynamic signs of hypovolemia, defined as both heart rate greater than 110 beats/min and pulmonary arterial diastolic pressure less than 10 mmHg. Transfusions of fresh frozen plasma or platelets required a prolonged prothrombin time or thrombocytopenia respectively coupled with bleeding of 250 ml *symbol* h sup -1 or more. After 12 h, strict transfusion criteria no longer applied; clinician preference supervened. The number of units each of homologous blood, fresh frozen plasma, and platelets administered within 12 h and within 5 days of operation was recorded.

Before induction of anesthesia and 2 h after completion of initial protamine infusion, coagulation assessment included measurement of activated partial thromboplastin time, platelet count, plasma fibrinogen, serum fibrin-fibrinogen-related antigen (latex agglutination, Thrombo-Wellco Test, Burroughs Wellcome, Research Triangle Park, NC), and latex agglutination D-dimer (D-di, American Bioproducts, Parsippany, NJ). D-dimer is a specific breakdown product of factor XIIIa-stabilized, cross-linked fibrin.

Although blood loss did not distribute normally, its logarithmic transform did. Transformed blood loss data underwent analysis of variance followed by the Tukey HSD multiple comparison test. Back transformed mean and 95% confidence intervals provided summary statistics for blood loss.

Because most patients in the study did not receive a transfusion, traditional analytic techniques did not apply. A dichotomous variable, “occurrence of transfusion,” described whether or not a patient received any banked blood product. Multinomial logistic regression determined the effects of patient grouping and demographic factors on the occurence of transfusion. This technique uses a stepwise approach, Wald statistics, and the log-likelihood ratio test [31] to obtain the determinants of blood transfusion among selected variables.

Because analysis of the blood loss data must account for the multifactorial nature of bleeding after surgery, a stepwise multivariate linear regression technique sought those factors associated with bleeding. This method first uses univariate analysis of variance to identify potential contributors, then a stepwise multiple regression to include any of those potential contributors that provide further predictive information.

Analysis of variance compared group continuous demographic data, and contingency tables analyzed frequency data. Coagulation data underwent a two-way analysis of variance using as factors patient group, sample time (before operation or after operation), and their interaction. For any data not conforming to the normal distribution, nonparametric tests applied. All tests were two-tailed with P < 0.05 denoting significance.

Of 155 patients enrolled, 7 did not complete the study. One patient did not undergo surgery. Another patient (group W) sustained a torn atrium during surgery accompanied by rapid blood loss, hypotension, and intraoperative myocardial infarction. A third patient (group Q) bled so much during surgery that the surgeon insisted antifibrinolytic therapy (epsilon aminocaproic acid) be administered at that time. A fourth patient (group P) developed cardiogenic shock after bypass for repeat sternotomy for coronary grafting and expired soon after operation. For one patient in group Q and two patients in group F, human error resulting in violations of the study protocol terminated study participation.

Demographic variables for the remaining 121 men and 27 women did not differ among groups. Table 2displays these data by group, along with the duration of ECC, and the hematocrits measured before skin incision, after termination of ECC, and 2 h after arrival in the intensive care unit. None of these differed among groups with the exception of final hematocrit for the group that received one-fourth the usual dose of tranexamic acid. All aortocoronary bypass graft procedures except one each in groups P and F included an internal mammary artery graft.

Baseline coagulation studies did not differ among groups. Table 3displays the aggregate data. Most patients for aortocoronary graft procedures received heparin infusions up to 4 h before surgery, resulting in prolonged aPTT results just before skin incision. This population of cardiac surgical patients, like other patients with cardiac disease, [5,32,33] displayed fibrin degradation products and increased D-dimer concentrations before surgery. The aggregate median FDP concentration was 0 micro gram *symbol* ml sup -1 for preoperative specimens and 10 micro gram *symbol* ml sup -1 for preoperative specimens. The aggregate median D-dimer concentration was 0.0 fibrinogen equivalency units (FEU, micro gram *symbol* ml sup -1) before operation and 0.25 FEU after operation. As expected, postoperative specimens demonstrated a decrease compared to the preoperative value in both platelet count (pooled values 158 plus/minus 59 (SD) vs. 228 plus/minus 76 *symbol* 10⁹*symbol* 1 sup -1, P < 10 sup -15) and fibrinogen concentration (217 plus/minus 74 vs. 319 plus/minus 103 mg *symbol* dl sup -1, P < 10 sup -15). Groups differed with respect to postoperative D-dimer concentrations (P = 10 sup -5, Table 3). The placebo group demonstrated statistically significantly increased D-dimer concentrations compared to every group receiving TA (P less or equal to 0.022). Furthermore, groups differed (P = 0.043) with respect to the change, postoperative minus preoperative, in D-dimer concentration. The largest median changes occurred in groups D (-0.5 FEU) and F (-0.125 FEU), whereas groups P, Q, and H exhibited no change in median D-dimer concentration.

(Figure 1) displays the blood loss data for the six groups. Figure 2displays the associated back transformed mean and 95% confidence intervals for blood loss. Analysis of variance confirmed a dose-response relationship: patients in groups W, D, and F each bled significantly less than those in group P (P = 0.042, 0.005, and 0.030, respectively) but not less than those in group Q (P = 0.233, 0.062, and 0.210, respectively). As expected in a dose-response study, no group's blood loss differed from that of its immediate neighbor in dosing scale (i.e., group P with group Q, group Q with group H, group H with group W). Groups Q, H, W, D, and F form a logarithmic sequence. Because group P received no drug, its logarithm is undefined, precluding a strict dose-response mathematical model of this relationship.

Stepwise multivariate linear regression identified four factors accounting for chest tube drainage within 12 h of operation. Table 4lists the results of this analysis. Factors associated with decreased bleeding were drug treatment group, female gender, increased body weight, and repeat (rather than primary) sternotomy. Substitution of blood loss indexed by body weight or by body surface area did not eliminate the association of bleeding with drug treatment group.

In each of the six groups, similar numbers of patients received transfusion. Table 5summarizes the transfusion requirements of study patients. Although blood loss differed between those patients in groups P, Q, and H and those in groups W, D, and F, transfusions given did not achieve a statistically significant difference (P = 0.075). Stepwise multinomial logistic regression examined drug treatment group, initial hematocrit, operating surgeon, patient weight, kind of operation, and sternotomy status (primary vs. repeat) as possible factors associated with administration of packed erythrocytes within 12 h of operation and within 5 days of operation. Only initial hematocrit predicted (P < 0.00005) whether a patient would receive a transfusion within 12 h of operation. Each 3%(absolute) decrease in initial hematocrit increased the likelihood of transfusion by a factor of 1.74 (95% confidence interval 1.33–2.27). For transfusions within 5 days of operation, initial hematocrit again emerged as the sole significant predictor (P < 10 sup -5), with each 3%(absolute) decrease in initial hematocrit increasing the likelihood of transfusion by a factor of 2.08 (95% confidence interval 1.57–2.76).

Responsible medical practice includes consideration of the cost of pharmaceuticals prescribed. Even for drugs with high therapeutic indexes, clinicians should strive to administer the minimum dose known to achieve a desired goal, in this case, decreased bleeding after ECC. Recent reports of antifibrinolytic therapy recommend administration of very large doses. [6,11,12,19] The current data demonstrate no advantage to administration of very large doses of the antifibrinolytic drug TA.

An alternative interpretation of these data is that TA provides no substantial benefit to unselected patients undergoing routine cardiac surgery accompanied by heightened attention to surgical hemostasis and stringent transfusion criteria. Authors believe this alternative interpretation would not negatively impact the hemostatic efficacy of TA in selected patients, during certain higher risk procedures, or in centers less focused on bleeding and transfusion.

This study design attempted to minimize confounding effects on the primary outcome variable, measured blood loss, by excluding patients with known bleeding diatheses or recent warfarin ingestion, by using only two operating surgeons for all patients, by avoiding subjective measurement of blood loss, and by administering additional protamine after chest closure to obviate heparin rebound. Measurement of blood loss by weight instead of volume improves its precision. Specific criteria for blood transfusion minimized subjective influences on the decision to transfuse allogeneic blood products.

Patients in the six treatment groups proved similar in their demographic profiles. The unequal group sizes reflect the random nature of patient assignment to groups. The incidence of primary sternotomy, although not different statistically among groups, varied substantially (71–96%), suggesting the need to control for it as an independent factor. Indeed, patients undergoing primary sternotomy bled more (436 vs. 297 ml median). Could this have occurred merely from a disproportionate representation of repeat sternotomy patients in the adequately dosed groups W, D, and F (16% vs. 10% for groups P, Q, and H), i.e., a confounding by the hemostatic effect of TA? Multivariate analysis should separate these effects to prevent this confounding (see Blood Loss).

Although coagulation test data demonstrate the antifibrinolytic effect of TA compared to placebo, they do not suggest an association of this effect with hemostatic efficacy: after operation, all groups receiving TA show no differences with respect to FDP or the more specific D-dimer. Several forces might contribute to this negative result. First, the discontinuous semiquantitative nature of the D-dimer assay employed saps statistical power from the comparison test, rendering the sample sizes inadequate to demonstrate a modest effect. Second, the study design may have limited demonstration of such an effect: the fourfold dosed group's D-dimer summary statistics (0.0 median, 0.0–0.25 range) contrast with those of the smaller dosed groups. Perhaps an eightfold group would have revealed a progression of decreased D-dimer concentration with increasing TA dose. Third, ample evidence exists for a multifactorial mechanism of antifibrinolytic drugs on blood conservation. If these drugs also reduce bleeding by preserving platelet function [26,34,35] via plasmin and thrombin receptor blockade, [36,37] then the direct association of D-dimer and blood loss becomes diluted.

Multivariate analysis identified drug treatment group as the most significant factor affecting blood loss, with higher dose groups displaying decreased bleeding. Of the other three significant predictors of decreased bleeding—increased body weight, female gender, and repeat sternotomy—female gender does not agree with a previously published series, [38] in which women more often bled excessively after primary coronary artery surgery. That study, however, used a retrospective design. Furthermore, female gender did not survive logistic regression analysis by those authors and thus did not form part of a nomogram developed to predict excessive blood use. In agreement with the current results, a report of 20, 524 patients undergoing coronary artery surgery by Loop et al. documented the need to reoperate for bleeding in only 3.8% of women compared to 5.5% of men. [39] The lesser bleeding among female patients remains unexplained.

The current study comprised only 20 patients (13%) undergoing repeat sternotomy, a cohort too small to draw definitive conclusions regarding the decreased bleeding seen with repeat operation. Nevertheless, anecdotal reports of improved operating conditions owing to diminished capillary bleeding with antifibrinolytic administration for repeat sternotomy [19] agree with the current results, leading one to speculate some unknown interactive effect of antifibrinolytic with the thromboplastin-rich scar tissue encountered during repeat sternotomy. Patients also might have received additional attention to surgical hemostasis with repeat operation compared to primary operation. The effect of body weight on blood loss, while significant statistically, accounts for only 4.2% of the variance of the logarithm of blood loss, demonstrating minimal clinical significance. Indeed, the regression model predicts that a doubling of body weight reduces blood loss by only 9%.

Most patients bled little in this study (Figure 1). Minimal bleeding overall renders difficult the demonstration of hemostatic efficacy by pharmaceuticals. Does such success make any such demonstration moot? No, because adjunctive efforts may not always be available or provided. Despite surgical or design factors contriving to minimize overall blood loss, this study demonstrates that TA provides a small but measurable hemostatic efficacy in a heterogeneous population of patients undergoing ECC. Also, doses lower than those needed to affect antifibrinolytic activity in vitro (groups Q and H)[27,28] do not provide hemostatic benefit compared with the dose that does (group W). Furthermore, doses in excess of the effective dose (groups D and F) confer no additional benefit. Do these results disagree with those of aprotinin, in which higher doses demonstrated hemostatic efficacy [19] not seen earlier with modest doses?[25] Recent work with moderate doses of aprotinin suggest that these moderate doses demonstrate hemostatic efficacy. [10,12,40–45] Other factors, such asimproved ECC equipment, changes in patient demographics, or surgical technique modifications render suspect comparisons of data from 25 yr ago [25] with more recent investigations. Both TA and aprotinin appear to provide hemostatic benefit in moderate doses.

These data do not demonstrate an effect of TA dose on blood transfused, despite an effect on blood lost. How can this be? First, the criteria require significant hemodilution to trigger a transfusion, thus providing an overall low incidence of transfusion (98.5 total units given to 44 out of 148 patients). Because mean postoperative hematocrits exceeded 25% for each group and transfusion criteria specify a hematocrit less or equal to 22%, a group difference in transfusions would be unexpected. Blood loss, not transfusion requirement, was the primary outcome variable in this study. Studies that use more liberal transfusion criteria, such as a hematocrit less than 30% rather than 21%, and in which patients bleed more in general, i.e., a placebo group 12 h loss [nearly equal] 850 g instead of 552 g, can more easily demonstrate an effect of drug on transfusion requirement. As with blood loss, any potential effect of TA on blood transfused is not rendered moot by these factors; rather, a savings in blood transfused may attain clinical significance when other mitigating factors are absent or not effective. Likewise, routine administration of TA may be unwarranted when blood loss is minimal and transfusion criteria strict, i.e., hematocrit < 25%.

Although none of the patients in this study suffered a detectable thrombotic event, the clinician should exercise caution in interpreting these limited results when considering the safety of TA. For this cohort of 148 patients with zero strokes, the 95% confidence interval for stroke incidence is 0–2.0%. [46] Thus, these data are consistent with a true rate of 1.0% or less. Although a larger cohort could yield stricter predictions of safety, 300 patients would be required to achieve a 95% confidence interval incidence of 2.0% or less, and 3,000 patients for less than 0.1%. This study's design did not attempt to yield safety statistics of that magnitude. On the other hand, prophylactic use of TA for hemostasis after ECC lasted only 12 h in the current study. Previously reported anecdotes of thrombotic complications of antifibrinolytic therapy in patients without malignancy all feature long-term (greater than several days') infusion of drug.

Because excretion of TA depends solely upon glomerular filtration, patients with serum creatinine concentrations above 2.0 mg *symbol* dl sup -1 did not participate. Although this study provides data on neither the efficacy nor safety of TA administration to patients with significant impairment of renal function, prudent clinicians should consider modifying TA dosages to such patients.

Based on the postoperative blood loss response of this heterogeneous cardiac surgical population to a 16-fold range of TA doses, we recommend a 10 mg *symbol* kg sup -1 intravenous loading dose followed by an infusion of 1.0 mg *symbol* kg sup -1 *symbol* h sup -1 for prophylactic administration of TA. Larger doses provide no additional savings in blood loss, whereas smaller doses carry decreased hemostatic efficacy.

The authors thank Henry Rosenberg, M.D., for his suggestions, and Patricia Whooley, Nasreen Malik, Evelyn Ross, M.S., Ellen Rupp. R. Ph., and Lai Huynh, R. Ph., for their assistance.