CCBYNC Open access

Tranexamic acid use and postoperative outcomes in patients undergoing total hip or knee arthroplasty in the United States: retrospective analysis of effectiveness and safety

BMJ 2014; 349 doi: (Published 12 August 2014) Cite this as: BMJ 2014;349:g4829
  1. Jashvant Poeran, assistant professor1,
  2. Rehana Rasul, biostatistician1,
  3. Suzuko Suzuki, fellow anesthesiology2,
  4. Thomas Danninger, research fellow anesthesiology2,
  5. Madhu Mazumdar, professor1,
  6. Mathias Opperer, research fellow anesthesiology2,
  7. Friedrich Boettner, attending orthopedic surgeon3,
  8. Stavros G Memtsoudis, attending anesthesiologist and senior scientist2, clinical professor of anesthesiology and public health4
  1. 1Institute of Healthcare Delivery Science, Mount Sinai Hospital System / Department of Health Evidence and Policy, Icahn School of Medicine at Mount Sinai, New York, NY, USA
  2. 2Department of Anesthesiology, Hospital for Special Surgery, 535 East 70th Street, New York, NY, USA
  3. 3Department of Orthopedic Surgery, Hospital for Special Surgery, New York, NY, USA
  4. 4Department of Healthcare Policy and Research, Weill Cornell Medical College, New York, NY, USA
  1. Correspondence to: S G Memtsoudis memtsoudiss{at}
  • Accepted 17 July 2014


Objective To determine the effectiveness and safety of perioperative tranexamic acid use in patients undergoing total hip or knee arthroplasty in the United States.

Design Retrospective cohort study; multilevel multivariable logistic regression models measured the association between tranexamic acid use in the perioperative period and outcomes.

Setting 510 US hospitals from the claims based Premier Perspective database for 2006-12.

Participants 872 416 patients who had total hip or knee arthroplasty.

Intervention Perioperative intravenous tranexamic acid use by dose categories (none, ≤1000 mg, 2000 mg, and ≥3000 mg).

Main outcome measures Allogeneic or autologous transfusion, thromboembolic complications (pulmonary embolism, deep venous thrombosis), acute renal failure, and combined complications (thromboembolic complications, acute renal failure, cerebrovascular events, myocardial infarction, in-hospital mortality).

Results While comparable regarding average age and comorbidity index, patients receiving tranexamic acid (versus those who did not) showed lower rates of allogeneic or autologous transfusion (7.7% v 20.1%), thromboembolic complications (0.6% v 0.8%), acute renal failure (1.2% v 1.6%), and combined complications (1.9% v 2.6%); all P<0.01. In the multilevel models, tranexamic acid dose categories (versus no tranexamic acid use) were associated with significantly (P<0.001) decreased odds for allogeneic or autologous blood transfusions (odds ratio 0.31 to 0.38 by dose category) and no significantly increased risk for complications: thromboembolic complications (odds ratio 0.85 to 1.02), acute renal failure (0.70 to 1.11), and combined complications (0.75 to 0.98).

Conclusions Tranexamic acid was effective in reducing the need for blood transfusions while not increasing the risk of complications, including thromboembolic events and renal failure. Thus our data provide incremental evidence of the potential effectiveness and safety of tranexamic acid in patients requiring orthopedic surgery.


Reducing blood loss and the need for blood transfusions surrounding orthopedic surgery remains a major concern among clinicians during the perioperative period. Many interventions have been developed over the past decades to achieve this goal, including controlled hypotensive anesthesia1 and various blood salvage techniques. In addition, pharmacologic approaches have become more popular in recent years. Especially, tranexamic acid has seen a renaissance among patients requiring orthopedic surgery, with numerous publications showing clinical efficacy and cost effectiveness.2 3 4 5 6 Indeed, a recent study found that the use of tranexamic acid may even make the use of blood salvage equipment unnecessary.7 Despite these promising results, valid data on safety are lacking, as large sample sizes are needed to determine this outcome. Thus concerns about the routine use of tranexamic acid remain.5 8 9 Data on perioperative outcomes, especially those related to thromboembolic events and renal complications, which have traditionally been of concern in the setting of antifibrinolytic use, are rare. Further, no population based data are available detailing outcomes in a large cohort outside of randomized controlled trials, which often only include selected patients based on stringent inclusion criteria and are thus not reflective of real world practice and are burdened by low external validity.3

Utilizing a large national database, we compared the characteristics and outcomes between patients receiving tranexamic acid and those that did not and analyzed if the use of tranexamic acid is independently associated with altered odds for blood transfusions and perioperative complications, particularly thromboembolic events and acute renal failure. We hypothesized that the characteristics associated with treated and untreated patients differed, and that tranexamic acid decreases the odds for blood transfusions while not increasing the risk of perioperative complications.


Data source and study design

For this retrospective cohort study we used the Premier Perspective database10 (Premier, Charlotte, NC) containing information on surgical hospital discharges from January 2006 to October 2012. This database provides complete billing information on a patient’s hospital stay as well as information on international classification of diseases-ninth revision clinical modification codes (ICD-9 CM) and current procedural terminology codes. Billed items are standardized by the database vendor after the hospital both reviews and consents to the items.

Study sample

We included cases if they had an indication of elective total hip or knee arthroplasty by the presence of ICD-9 CM codes 81.51 and 81.54, respectively. Cases were excluded if information on sex was unavailable (n=10), discharge status was unknown, patients were still listed as in-patients at the end of the data collection period (n=291), or patients had both a total hip and a total knee arthroplasty during the same hospital stay (n=193).

Study variables

The main intervention variable was the use of intravenous tranexamic acid on the day of surgery (further referred to as perioperative tranexamic acid use), which we categorized into four groups based on billing information retrieved dosing: none, ≤1000 mg, 2000 mg, and ≥3000 mg. Patient characteristics included age, sex, and race (white, black, Hispanic, other). Healthcare related variables included type of insurance (commercial, Medicaid, Medicare, uninsured, other), hospital location (rural, urban), hospital bed size (<300, 300-499, ≥500), hospital teaching status, and the mean annual number of total hip and knee arthroplasties per hospital. Procedure related variables included type of procedure (total hip or knee arthroplasty, unilateral or bilateral for both), type of anesthesia (general, neuraxial, general and neuraxial combined, other, unknown), use of peripheral nerve block, use of anticoagulants (antiplatelets (aspirin, other), warfarin, heparin, other), and year of procedure. Analogous to a previous report by our study group,11 we used billed items to define type of anesthesia. The same applied to the definition of the use of anticoagulants (see supplementary appendix 1) for which we also took into account simultaneous use of multiple medications.

We used the Deyo adaptation of the Charlson comorbidity index to measure overall comorbidity burden.12 Individual Elixhauser comorbidities,13 and the presence of sleep apnea (not included in either index) were evaluated.

Primary outcome variables included transfusion (allogeneic or both allogeneic and autologous), thromboembolic complications (pulmonary embolism, deep venous thrombosis), and acute renal failure. In addition, we considered a combined complication variable, which included thromboembolic complications and acute renal failure as well as in-hospital mortality, cerebrovascular events, and acute myocardial infarction. Secondary outcome variables included mechanical ventilation, admission to an intensive care unit, length of hospital stay in days, and cost of hospital stay in US dollars. Supplementary appendix 2 gives an overview of the ICD-9 codes used.

Univariable analysis

The association between tranexamic acid use and study variables was assessed using χ2 tests for categorical variables and t tests for continuous variables. Median and interquartile range were reported for length of hospitalization and cost of hospital stay due to their skewed distribution; significance between groups was measured using the Mann-Whitney rank sum test. To facilitate overall readability of the main manuscript text we chose to illustrate the univariable analyses by categorical “yes or no” tranexamic acid use instead of the more detailed categorization by dosage; the latter tables are provided in supplementary appendices 3 to 5.

Multilevel logistic regression analysis

We built separate multilevel multivariable logistic regression models to measure the association between use of tranexamic acid and the binary outcome variables. To account for correlation of patients within hospitals we included a random intercept term that varies at the level of each hospital. All hospitals (clusters) had sufficient patients (n>30) according to previously recommended sample sizes for this type of model to reduce bias.14 We adjusted models using all available demographic, healthcare related, procedure related, and comorbidity variables that were significant (P<0.1515) in the univariable analysis. Odds ratios (95% confidence intervals) and P values are reported and used together as a measure of overall significance.

Propensity score matching

To test the robustness of results and their sensitivity to the methodology chosen, we conducted a propensity score analysis. We calculated propensity scores from a multilevel logistic regression model with the outcome of tranexamic acid use (categorized into “yes or no” as opposed to the more detailed categorization used in the multilevel model) and the same covariates used in the primary analysis. A patient who received tranexamic acid (case) was matched with three patients who did not receive this drug (controls) by comparing their propensity scores.16 We measured the balance between the groups by comparing standardized differences on the original study sample and the matched sample.17 Although there is no consensus on a standard threshold to consider as an acceptable balance, a standardized difference of less than 10% or 0.1 to indicate negligible differences between groups has been suggested.18 The Cochran-Mantel-Haenszel estimate of the common odds ratio to control for the three pairs of matches and 95% confidence intervals were estimated on the matched sample to evaluate the effect of use of tranexamic acid on outcomes.

All analyses were performed in SAS v9.3 statistical software (SAS Institute, Cary, NC). The SAS procedure GLIMMIX was used for multilevel regression analyses. To match samples for propensity score analysis we used the SAS macro OneToManyMTCH with 8-digit to 1-digit match without replacement.


The study sample consisted of 872 416 cases of elective total hip or knee arthroplasty from 510 hospitals.

Univariable results

Table 1 lists the patient characteristics, healthcare related variables, and procedure related variables by tranexamic acid use. Except for average age (65.9 years for the tranexamic acid groups v 65.8 years for the no tranexamic acid group), all differences between the group of patients receiving tranexamic acid versus the group not receiving tranexamic acid were significant. Most notably, tranexamic acid was given more often to white patients (82.7% v 75.7%), in medium sized (300-499 beds) hospitals (57.7% v 37.4%), and in hospitals with a higher mean annual number of total hip or knee arthroplasties (776.6 v 731.2). Moreover, perioperative tranexamic acid use increased dramatically, from almost 0% in 2006 to 11.2% in 2012.

Table 1

 Patient characteristics and healthcare and procedure related variables by tranexamic acid use. Values are numbers (percentages) unless stated otherwise

View this table:

Table 2 shows the comorbidity burden by tranexamic acid use. The mean Deyo-Charlson comorbidity index differed only slightly between the groups (0.72 for patients in the tranexamic acid group v 0.74 for patients in the no tranexamic acid group, P=0.0267). The incidence of individual comorbidities was similar in both groups for most comorbidities.

Table 2

 Comorbidities by tranexamic acid use. Values are numbers (percentages) unless stated otherwise

View this table:

Primary and secondary outcome variables

Table 3 shows the primary and secondary outcome variables by tranexamic acid use. Compared with patients who did not receive tranexamic acid, patients receiving tranexamic acid had lower rates of all binary outcomes: allogeneic or autologous transfusion (7.7% v 20.1%, P<0.001), thromboembolic complications (0.6% v 0.8%, P=0.0057), combined complications (1.9% v 2.6%, P<0.001), need for mechanical ventilation (0.1% v 0.2%, P=0.0003), and admission to an intensive care unit (3.1% v 7.5%, P<0.001). Median length of hospital stay was three days for both groups; median cost of hospital stay was $14 890 for the tranexamic acid group compared with $15 110 for the group not receiving tranexamic acid, P<0.001.

Table 3

 Outcome variables by tranexamic acid use. Values are numbers (percentages) unless stated otherwise

View this table:

Multilevel logistic regression analysis

When controlling for covariates, the use of tranexamic acid was significantly associated with a decreased need for allogeneic or autologous blood transfusions (odds ratio varying from 0.31 to 0.38 by dose category), and allogeneic blood transfusions (odds ratio 0.29 to 0.37), with no significantly increased risk for complications: thromboembolic complications (0.85 to 1.02), acute renal failure (0.70 to 1.11), combined complications (0.75 to 0.98), and admission to an intensive care unit (0.73 to 1.01) (table 4). For the dosage categories, 2000 mg tranexamic acid seemed to have the best effectiveness and safety profile. For all models, the C statistics were high (range 0.83 to 0.90).

Table 4

 Results from multilevel logistic regression model and propensity score analysis for primary outcomes. Values are odds ratios† (95% confidence intervals) unless stated otherwise

View this table:

Propensity score matching

Out of the 20 051 patients who received tranexamic acid (cases), 5486 were successfully matched to patients who did not receive tranexamic acid (controls). The matched sample was well balanced (standardized differences <10%) for almost all variables (see supplementary appendix 3). Similar to the multilevel models, the propensity score analyses showed decreased odds for allogeneic or autologous transfusion (odds ratio 0.50, 95% confidence interval 0.45 to 0.55) and allogeneic transfusion (0.47, 0.42 to 0.53) in patients given tranexamic acid. In this model too, we found no increased risk for complications: thromboembolic complications (0.86, 0.59 to 1.25), acute renal failure (0.74, 0.57 to 0.96), combined complications (0.75, 0.61 to 0.92), and admission to an intensive care unit (0.85, 0.74 to 0.99).


In this population based study of 872 416 total hip and knee arthroplasty procedures, the use of tranexamic acid was significantly associated with an up to 69% reduction in the need for allogeneic or autologous blood transfusions. Further, irrespective of the use of anticoagulants, tranexamic acid use was not associated with an increased risk for perioperative complications, including thromboembolic events and acute renal failure. In a univariable context, we also found the use of tranexamic acid to be associated with reduced healthcare utilization: lower rates of advanced care need, lower length of hospital stay, and lower costs of hospital stay.

Strengths and limitations of this study

The main strengths of our study are the large sample size, use of data from actual, everyday practice (establishing generalizability), and the multivariable multilevel analysis controlling not only for individual level factors but also for hospital clusters. In particular, the ability to control for the use of anticoagulants and type of anesthesia—both important determinants of transfusion risk—is unique for population based databases. Further, the large sample size allows for the study of safety concerns regarding the incidence of rare complications such as pulmonary embolism or deep venous thrombosis. Meta-analyses are limited in their utility to assess generalizable safety concerns about perioperative tranexamic acid use, as patients with, for example, a history of cardiovascular disease or those taking warfarin or low molecular weight heparin, are often excluded from these trials.3

Our study has several limitations. First, our analysis utilized data from an administrative database, and detailed clinical information was missing, including hemoglobin levels or other transfusion triggers. We expect the multilevel model to account in part for this limitation as it adjusts for practice variations among hospitals (of which transfusion practices are a part). In addition, our outcome is the actual transfusion being administered regardless of the existence of triggers. As in all other guidelines, transfusion triggers are not static and, especially in the perioperative period, may be dynamic and only partially dependent on hemoglobin levels, as rapidly changing variables such as patients’ symptoms and expected trajectory may also play important roles. Thus, lack of detailed clinical data will remain a problem for studies that use retrospective data. Furthermore, having detailed clinical information still would not guarantee perfect validity of data as there will always be unmeasured factors influencing decisions and outcomes. This problem is dealt with in a randomized clinical trial setting, but then again at the cost of loss of generalizability to more general populations. Another important facet of the lack of detailed clinical data refers to the selective use of tranexamic acid in patients with arterial stents or a history of thromboembolic events, both considered relative contraindications for tranexamic acid use by some practitioners. However, the pseudorandomized approach of the propensity score analysis showed the same results as the multilevel analysis. Moreover, although the Elixhauser comorbidities (but also other patient characteristics) do not specifically capture these contraindications, they might act as a partial proxy, thus reducing the effect of this limitation. Another limitation refers to residual confounding. Although we included many important covariates in our analytic models while also accounting for correlation of patients within hospitals, residual confounding might remain. However, the multilevel models showed high C statistics (up to 0.90) indicating good model discrimination between subjects for each level of the outcome. The use of ICD-9 codes and billing data may also be associated with (registration) bias. However, this bias should be equally distributed between our treatment groups, thus reducing its impact. Although we did have information on the dose of tranexamic acid using billing data, we do not know with complete certainty how much of the billed medication was actually administered to the patients as this is a topic of major controversy. We therefore have limited our statements to the categorical (“yes or no”) use of tranexamic acid. Finally, in respect to the safety of tranexamic acid we were only able to study complications that occurred during the patients’ hospital stay, which is an inherent limitation of our data source. This may cause an underestimation of the actual incidence of complications. However, one study showed that more than 90% of complications in unilateral arthroplasties occur within four days after surgery, suggesting that most complications should be encompassed within our dataset.19

Comparison with existing literature

The use of tranexamic acid has been shown to be effective in reducing blood transfusions both in small, randomized controlled trials and in meta-analytic publications.2 3 4 5 6 7 9 Our study validates these findings, by providing data on effectiveness from information gathered in a wide range of settings representing actual, “real world” practice. This is important, as the information gathered from randomized controlled trials conducted in single institutional—often academic—settings frequently lacks external validity, as participants tend to be highly selected. The effect size, as measured by the reduction in the odds for the need of allogeneic or autologous blood transfusions by up to 69% was large even by conservative standards. This finding has significant implications for two reasons: firstly, total hip and knee arthroplasty are common procedures, with over one million interventions annually in the United States alone,20 and the utilization is expected to increase dramatically in the future.21 Secondly, joint arthroplasties are associated with significant blood loss with relatively high transfusion rates compared with other elective surgeries.22 In this context, the use of tranexamic acid in patients undergoing total joint arthroplasty may have a profound clinical and economic impact if used in appropriate candidates—that is, those at high risk for requiring blood transfusions.23 24

When analyzing the impact of tranexamic acid use on complications we found no increased risk for adverse outcomes in general, and for thromboembolic events and acute renal failure in particular. In fact, the multilevel model showed significantly decreased risks for some complications. These complications have been put forward by several clinicians as major reasons for a conservative use of tranexamic acid given previously published concerns with agents of this category.25 As tranexamic acid inhibits fibrinolysis, safety concerns are based on the fact that interference with the coagulation cascade may promote a procoagulable state and thus increase the risk for complications such as pulmonary embolism, deep venous thrombosis, myocardial infarctions, and cerebrovascular events.25 This is of particular concern as patients who require joint arthroplasty have been identified as an especially vulnerable group for clotting based complications as the source of major morbidity and mortality.26 Further, previous publications have identified the use of certain antifibrinolytics to be associated with increased mortality in surgical patients, leading to the withdrawal of aprotinin from the US market.27 Renal failure was identified as a major contributor to this outcome and has since been the focus of many outcome studies related to antifibrinolytics.28 Although comparative evaluations between aprotinin and other agents such as tranexamic acid have been published showing improved safety profiles with tranexamic acid, such studies are scarce for the population of patients requiring orthopedic procedures.28 Even though a meta-analysis showed efficacy of either agent in patients requiring orthopedic surgery, the authors concluded that safety data are needed before recommending the use of these agents in this patient population.29 Despite encouraging results derived from our analysis regarding the safety of tranexamic acid, we cannot provide support for the ubiquitous use of tranexamic acid in all patients requiring joint arthroplasty as the differential impact on complications among patient subpopulations remains to be studied. In this context, studies with aprotinin have suggested that the use of this antifibrinolytic agent among low to intermediate risk patients requiring cardiac surgery may increase mortality risk, while not having the same effect in high risk patients.30 Thus, a conservative approach taking into account appropriate stratification strategies for bleeding risk seems prudent when deciding to use any antifibrinolytic perioperatively. Future studies might focus on this subgroup specific effectiveness and safety of tranexamic acid.

Conclusions and implications

Utilizing population based data we found that tranexamic acid was effective in reducing the need for blood transfusions while not increasing the risk of complications, including thromboembolic events and renal failure. Although our data provide incremental evidence of the potential effectiveness and safety of tranexamic acid in patients requiring orthopedic surgery, this study has limitations inherent to observational analyses. Moreover, outcome data in subpopulations of patients remain to be studied. Therefore, the prudent identification of patients most likely to benefit from tranexamic acid—that is, those at increased risk of bleeding—is warranted. Additional studies focusing not only on subgroup specific effectiveness and safety but also on optimal dosing schemes are needed.

What is already known on this topic

  • Tranexamic acid has been shown to reduce perioperative blood loss and blood transfusions in orthopedic surgery

  • Safety concerns remain, however, as small and highly selective populations were studied

  • Large scale effectiveness studies are lacking

What this study adds

  • Tranexamic acid is associated with a decreased risk for blood transfusions, while not increasing the risk of complications, including thromboembolic events and renal failure

  • Our data provide incremental evidence of the potential effectiveness and safety of tranexamic acid in patients requiring orthopedic surgery


Cite this as: BMJ 2014;349:g4829


  • Contributors: JP, RR, MM, and SGM designed the study and attained the Premier Perspective database. RR and JP analyzed data under guidance of SGM and MM. All authors contributed to the interpretation of the results, particularly TD, MO, SS, and FB with their clinical viewpoints. All authors reviewed and approved the final manuscript. All authors, particularly JP and RR, had full access to all of the data in the study. All authors gave their final approval of the submitted manuscript and agreed to be accountable for all aspects of the work. JP and RR take responsibility for the integrity of the data and the accuracy of the data analysis. SGM is the guarantor.

  • Funding: Contributions of JP, RR, and MM on this project were partly funded by the Clinical Translational Science Center, New York, NY. SGM is funded by the Anna Maria and Stephen Kellen Career development award, New York, NY. The study sponsors had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript. The views expressed in this article are those of the authors and do not necessarily represent the views of the sponsors or authors’ affiliated institutions.

  • Competing interests: All authors have completed the ICMJE uniform disclosure form at and declare: no support from any organization for the submitted work; no financial relationships with any organizations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work.

  • Ethical approval: These data meet the requirements of deidentification as defined by the Health Insurance Portability and Accountability Act and were exempt from consent requirements of the Hospital for Special Surgery institutional review board (No 2012-050-CR2).

  • Data sharing: Data were purchased from Premier and as such are restricted for this project and cannot be shared because of these restrictions on use of data. Syntax is available from the corresponding author (

  • Transparency: The senior author, SGMemtsoudis (the manuscript’s guarantor), affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.

This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 3.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See:


View Abstract