Safety of Tranexamic Acid in Hip and Knee Arthroplasty in High-risk Patients

The Mount Sinai Hospital Institutional Review Board approved this retrospective cohort study (Project No. 14-0067). Included data were extracted from the Premier Healthcare database,^14,15  a national all-payer claims database representing 20 to 25% of US hospital discharges. Hospitals participating in Premier Healthcare are mainly concentrated in the South of the United States (approximately 40%), with equal distributions of hospitals located in the Northeast, West, and Midwest (approximately 20% each). This data set contains detailed billing information on hospitalizations. International Classification of Diseases, Ninth Revision (ICD-9) procedure codes for total hip arthroplasty (81.51) and total knee arthroplasty (81.54) were used to define the study cohort including data from 2013 to 2016. Of note, the data set purchased from Premier Healthcare included only ICD-9 codes; International Classification of Diseases, Tenth Revision codes—introduced in the United States in October 2015—were converted to ICD-9 codes using a proprietary algorithm. Before any exclusion criteria, the Premier Healthcare data set included 820,816 total hip and knee arthroplasties performed between 2013 and 2016. Exclusion criteria pertained to patients who met one or more of the following criteria: unknown sex (n = 280), unknown discharge status (n = 457), classification as a nonelective procedure (n = 49,172), patient under 18 yr of age (n = 119), and classification as an outpatient procedure (n = 5,677).

A data analysis and statistical plan was written, date-stamped, and recorded in the investigators’ files before the data were accessed. The exposure of interest was the use of tranexamic acid in patients with preexisting comorbidities for which concerns existed of associated increased risks with tranexamic acid use. Tranexamic acid use was defined using pharmacy billing information, which also provided information on the dose (1,000 mg, 2,000 mg, 3,000 mg or more, or not specified) and the mode of administration, classified into “parenteral” or “other/not specified”; the latter likely represents intraarticular use. Tranexamic acid use was applied as a binary variable in analyses; however, information on the distribution of dose and mode of administration is presented as well.

The main focus of this study is the use of tranexamic acid in high-risk patients. Thus, three separate definitions of high-risk patients were applied, based on preexisting comorbidities that were defined by ICD-9 codes in line with previous studies.^16–18  To distinguish between preexisting (comorbidities) and “new-onset” (complications during hospitalization), we applied the “present-on-admission” indicator provided in the Premier Healthcare data set.

The three separate high-risk definitions were based on the following:

1. Patients’ history of VTE (including deep venous thrombosis and pulmonary embolism), MI, seizures, or ischemic stroke/transient ischemic attacks.

2. Patients’ history of renal disease: Concerns have been noted on tranexamic acid administration in these patients in the context of cardiac surgery and other clinical contexts¹⁹ ; it has also been used as exclusion criterion in trials focusing on tranexamic acid in lower-extremity joint arthroplasty.^19–21 

3. Patients’ history of atrial fibrillation: This was added after discussions flowing from presenting our preliminary results at the annual meeting of the American Association of Hip and Knee Surgeons on November 3, 2018, in Dallas, Texas. Here, atrial fibrillation was suggested to be included because it represents a thromboembolic risk factor. Likewise, it has also been evaluated in another recent study on safety of tranexamic acid using institutional data.²² 

The two coprimary outcomes were (1) blood transfusion use and (2) composite complication of new-onset VTE, MI, seizures, or ischemic stroke/transient ischemic attacks. Complications were evaluated separately as well as in a “composite complication” variable. Blood transfusion was defined as a patient having a billing charge description or ICD-9 code indicating the transfusion of any blood product during the hospital stay. The aforementioned clinical complications were only counted if they occurred during the hospital stay, given the short course of tranexamic acid treatment in these patients and the half-life of 2 h. Secondary outcomes included acute renal failure, 90-day readmission caused by the aforementioned clinical complications (based on the ICD diagnosis codes associated with the readmission), in-hospital mortality, length of hospital stay, and cost of hospitalization. In total, these are 12 outcome variables.

Patient demographics included age, sex, and race (White, Black, other). Healthcare-related variables included insurance type (commercial, Medicaid, Medicare, uninsured, other/unknown), hospital location (urban, rural), hospital size (less than 300, 300 to 499, 500 beds or more), teaching status, and annual number of total hip and knee arthroplasties performed per hospital. Procedure-related variables were year and type of procedure, type of anesthesia, use of patient-controlled analgesia, nonopioid analgesics (nonsteroidal anti-inflammatory drugs, cyclooxygenase-2 inhibitors, ketamine, pregabalin/gabapentin and acetaminophen), and use of peripheral nerve blocks. Comorbidity-related variables (in addition to the aforementioned comorbidity definitions of interest) included comorbidities from the Charlson Comorbidity Index,²³  history of substance use/abuse (including smoking), chronic pain conditions, psychiatric comorbidities, opioid use disorder, and obstructive sleep apnea. The latter five and postsurgical analgesic variables were included given their association with opioid use and perioperative complications. We additionally assessed anticoagulant/antiplatelet use, categorized into aspirin or other antiplatelets, heparin, warfarin, other anticoagulants, more than one of the preceding, and none. Any medication- related variable was defined as the patient having a billing charge (using descriptions of generic names) at any point during the hospital stay. ICD-9 codes used for definitions of outcomes and comorbidities are presented in the Supplemental Digital Content (supplemental table 1, https://links.lww.com/ALN/C584). Our study was not burdened by missing data as the main variables of interest were defined based on whether or not a patient record had the associated ICD code or billing description. If this was not the case, then they were considered not to have that outcome/exposure/confounder.

First, univariable associations between tranexamic acid use and the aforementioned study variables were assessed; percentages and medians (with interquartile ranges) were compared. Univariable group differences easily reach statistical significance in large sample sizes; we therefore opted to report both standardized differences and P values (from chi-square and Mann–Whitney U test for categorical and continuous variables, respectively). A standardized difference of 0.1 (or 10%) generally indicates a meaningful difference in covariate distribution between groups.²⁴ 

Mixed-effects models²⁵  measured associations between tranexamic acid use and outcomes. This was done separately for patients in high-risk and non–high-risk groups by applying interaction terms between the aforementioned high-risk definitions and tranexamic acid use. Given the focus on tranexamic acid use in high-risk patients, we only present effect estimates from high-risk groups in the article text; effect estimates from non–high-risk groups can be found in the Supplemental Digital Content (supplemental tables 2 through 4, https://links.lww.com/ALN/C584). Mixed-effects models account for correlation of patients within hospitals; this is important in this context as patients treated at the same hospital are more likely to receive similar treatment and care than a patient treated at another hospital. Given the large study sample size and decreased need for parsimony, models were adjusted for all covariates available. Additionally, as a sensitivity analysis, these same analyses were performed using a cohort only including hospitals with a high use of tranexamic acid (greater than or equal to 75% hospital-specific tranexamic acid use), to address the issue of potential selective use of tranexamic acid.

We report odds ratios and Bonferroni-adjusted 99.9% CIs taking into account the number of hypotheses tested for in the main analyses (72 hypotheses reflecting 12 outcomes and 6 (non–)high-risk groups of interest: three definitions of high-risk and their subsequent three complementary non–high-risk counterparts). The reported 99.9% CI values are computed from 1 - (0.05/72) = 99.9%. This was applied to reduce the risk of type I errors (false-positive associations); however, this step may increase the likelihood of type II errors (false-negative associations).²⁶  Given this conservative approach and our main interest in safety outcomes, we consider any association in multivariable analyses that demonstrates statistical significance (after applying the Bonferroni correction) as meaningful. For the continuous outcomes (length of stay and cost of hospitalization), instead of odds ratios, we report the percentage of change; as for these models, a γ distribution with a log link function was applied given that these variables are highly skewed.^27,28 

We additionally assessed trends in tranexamic acid use by the three aforementioned high-risk groups because it may provide information on differential use of tranexamic acid between those with and without comorbidities. Hypothesis testing was two-sided. All analyses were performed using SAS version 9.4 statistical software (SAS Institute, USA).

During the peer-review process and presentation of preliminary results at national professional meetings, the following adjustments were made to our initial analyses:

• ▪

  Addition of atrial fibrillation as an additional definition of high risk.

• ▪

  Removal of analyses representing Charlson Comorbidity Index categories as a definition of high risk.

• ▪

  Addition of a history of seizure(s) and new-onset seizure(s) as both a determinant of high-risk status and an outcome of interest, respectively.

• ▪

  Restricting the study cohort to 2013 to 2016 (previously 2006 to 2016).

• ▪

  Providing a preliminary “dose-response” analysis modeling tranexamic acid by dose categories and their association with “composite complications” among patients with a history of VTE, MI, seizures, or ischemic stroke/transient ischemic attacks receiving tranexamic acid. Here, the presence of a dose-response pattern of higher complication odds with a higher tranexamic acid dose would inform potential dose adjustment strategies in high-risk patients, whereas the absence of such a pattern would argue against higher risks associated with tranexamic acid use in these high-risk patients.

• ▪

  Separating out renal comorbidities and renal complications from the rest of the preexisting comorbidities/complications.

• ▪

  Application of Bonferroni adjustments.

Overall, 765,011 (representing 589 hospitals) lower-extremity joint arthroplasties were included: 265,158 total hip arthroplasties and 499,853 total knee arthroplasties with an overall blood transfusion rate of 8.1% (n = 62,190). Tranexamic acid was utilized in 52.9% (n = 404,974) of procedures; of note, tranexamic acid was not used at all in 73 hospitals representing 37,148 cases. The majority of tranexamic acid (n = 381,943; 94.3%) was administered intravenously with billing for 2,000 mg the most common dose (n = 207,907 [51.3%]; n = 119,275 [29.5%], n = 54,761 [13.5%], and n = 23,031 [5.7%] for 1,000 mg, 3,000 mg or more, and unspecified dose categories, respectively). Tranexamic acid utilization, indicated by row totals, was particularly higher in more recent years, in total knee arthroplasties, and among those receiving most nonopioid analgesic modes (table 1). Interestingly, tranexamic acid use was lower among those receiving patient-controlled analgesia, and as expected, use of anticoagulants was highly associated with tranexamic acid use (standardized differences of more than 0.1; table 1). Tranexamic acid use differed minimally between patients in the various high-risk groups compared to those in the non–high-risk groups (table 1: indicated by boldface text; all standardized differences less than 0.1).

Tables 2 through 4 show dosing characteristics and outcomes by tranexamic acid use, separately for the three high-risk groups: those with a history of VTE, MI, seizures, and ischemic stroke/transient ischemic attack (n = 27,890) in group I, renal disease (n = 44,608) in group II, or atrial fibrillation (n = 45,952) in group III. Of note, numbers for the non–high-risk groups (i.e., patients without a history of the aforementioned comorbidities) can be found in the Supplemental Digital Content (supplemental tables 2 through 4, https://links.lww.com/ALN/C584). Minimal differences in tranexamic acid dosing were seen between high-risk and non–high-risk patients. Across high-risk groups, generally lower blood transfusion rates, complication rates, and resource utilization were observed in patients receiving tranexamic acid (compared to those that did not).

This pattern was further corroborated in the adjusted analyses presented in table 5; tranexamic acid use was associated with decreased odds of blood transfusion in high-risk group I (721 of 13,004 [5.5%] vs. 2,293 of 14,886 [15.4%]; odds ratio, 0.307; 99.9% CI, 0.258 to 0.366); group II (2,045 of 22,424 [9.1%] vs. 5,159 of 22,184 [23.3%]; odds ratio, 0.315; 99% CI, 0.263 to 0.378); and group III (1,325 of 22,379 [5.9%] vs. 3,773 of 23,573 [16.0%]; odds ratio, 0.321; 99% CI, 0.266 to 0.389); all adjusted comparisons P < 0.001. No increased odds of composite complications were observed: high-risk groups I (129 of 13,004 [1.0%] vs. 239 of 14,886 [1.6%]; odds ratio, 0.89; 99.9% CI, 0.49 to 1.59); group II (238 of 22,424 [1.1%] vs. 369 of 22,184 [1.7%]; odds ratio, 0.98; 99% CI, 0.58 to 1.67); and group III (187 of 22,379 [0.8%] vs. 290 of 23,573 [1.2%]; odds ratio, 0.93; 99% CI, 0.54 to 1.61); all adjusted comparisons P > 0.999. In addition, no increased odds were observed for the other complication outcomes when applying both Bonferroni-adjusted and -unadjusted (not shown in table 5) P values. To the contrary, tranexamic acid use was associated with decreases in length of stay (−3.2% to −4.4%) and cost of hospitalization (−2.8% to −4.2%); all P < 0.01. A post hoc analysis evaluating a potential dose–response effect (table 6) did not demonstrate a pattern of higher complication odds with a higher tranexamic acid dose in high-risk patients. All model c-statistics were 0.74 or higher (the majority was 0.80 or higher), indicating sufficient discrimination.

Finally, when evaluating tranexamic acid utilization trends, the overall increased utilization was observed with somewhat lower tranexamic acid use in high-risk (compared to non–high-risk) patients (fig. 1). Sensitivity analyses including only hospitals with at least 75% tranexamic acid use (to address potential selective tranexamic acid use) found results similar to those in our main analyses (i.e., no increased odds of complications with tranexamic acid use among high-risk patients; Supplemental Digital Content (supplemental table 4, https://links.lww.com/ALN/C584).

In this cohort of 27,890 to 45,952 high-risk patients undergoing elective lower-extremity joint replacement surgery, we found an increase in tranexamic acid use over time. The utilization rate and dosing schemes did not meaningfully differ between high-risk and non–high-risk patients. Although consistently associated with decreased odds of blood transfusions, tranexamic acid use was not associated with increased odds of a variety of thromboembolic and ischemic complications among high-risk patients, using varying definitions of high risk. To the contrary, somewhat lower length and cost of hospitalization was observed with tranexamic acid use among high-risk patients. In addition, higher doses of tranexamic acid in high-risk patients did not demonstrate a pattern of higher complication odds; conversely, there was a trend toward lower odds of complications.

These results target a clinical challenge and prominent evidence gap on tranexamic acid use in high-risk patients, as previously noted.¹³  Adding to this challenge is the fact that tranexamic acid is still considered “off label” for use in total hip and knee arthroplasty surgery, and safety concerns remain,^29,30  despite the availability of high-quality evidence supporting the use of tranexamic acid in this surgical cohort in terms of its beneficial impact on blood loss and transfusion risk.^2–6,13,29  Among several large meta-analyses and observational studies, none have demonstrated increased risks associated with the use of tranexamic acid,^{2–6,18,22,31}  for which main concerns include thromboembolic risk and renal failure caused by acute renal cortical necrosis.^7–9,30  These concerns may be exacerbated in patients considered at high risk because of preexisting comorbidities, a subgroup generally excluded from trials.^7,13,32 

Despite these concerns, our results demonstrate a minimal difference in tranexamic acid use among high-risk and non–high-risk patients, using various definitions, suggesting that high-risk status does not appear to be a main driver behind tranexamic acid use. Moreover tranexamic acid was consistently associated with decreased blood transfusion odds, whereas no increased odds of complications were observed among high-risk patients. These findings are in line with a recently published meta-analysis and an institutional study, although smaller sample sizes were assessed.^22,32  In the former, Fillingham et al.³²  used the American Society of Anesthesiologists (Schaumburg, Illinois) Physical Status score in definitions of high risk and found that patients with a higher ASA score receiving tranexamic acid were not at increased risk of VTE. Likewise, using single-institution data, Porter et al.²²  subgrouped 38,220 lower-extremity joint replacement patients based on the presence of prothrombotic conditions before surgery and found no statistically significant increased odds of complications among those with and without a high-risk status. The current study supplements the analysis by Porter et al.²²  by using a larger sample size and thus has the ability to detect smaller differences between groups regarding these rare complications. Moreover, the current study provides information on additional definitions of high risk, i.e., renal disease and history of seizure, and additional outcomes such as new-onset seizures given the existence of official warnings.³³  The data used in the current study also allowed an assessment of various tranexamic acid doses in which we did not find a dose-response pattern of higher complication odds with a higher tranexamic acid dose, thus further supporting our main conclusions regarding the absence of an association between tranexamic acid use and complications in high-risk patients. Finally, we were able to address potential selective use of tranexamic acid in sensitivity analyses. Although these studies add to the existing body of literature suggesting no increased complication risk after tranexamic acid use among high-risk patients,^22,32,34,35  important future work may follow from prospective multi-institutional registries specifically geared toward tracking outcomes after administration of tranexamic acid, especially given the expected increased use in other surgeries and expected increases in joint arthroplasty demand.^36,37  Of note, even in the large sample size used in the current study, specific preexisting comorbidities represent relatively small samples, for example, 575 patients with a history of VTE. Thus, given that tranexamic use continues to increase, future work may also need to be focused on even more specific populations than the current study.

We observed somewhat lower length and cost of hospitalization with tranexamic acid use among high-risk patients, whereas higher doses of tranexamic acid in high-risk patients also demonstrated a pattern toward lower odds of complications; some of these results mirror previous findings.^18,38  Although speculative, selective avoidance of tranexamic acid may occur in patients perceived to be at high risk for complications. This is, however, not supported by our results, because minimal differences were observed in tranexamic acid use across high-risk and non–high-risk groups. Other potential mechanisms may stem from the potential beneficial effects of avoidance of blood loss/transfusion,^39,40  which may exert a stronger impact than any potential negative effect of tranexamic acid. Given the seriousness of potential harm, we acknowledge the statements of Fillingham et al.³²  in that vigilance should be exercised, and each patient’s individual risk profile should be considered together with the potential benefits of tranexamic acid.

Our study is burdened by various limitations. Despite the large overall sample size and generalizability, our high-risk groups were relatively small compared to those categorized as non–high-risk. Moreover, complications such as our composite complication variable are rare in this high-volume surgical cohort, further limiting reproducibility and reliability of observations; continuing monitoring efforts will therefore be crucial. We also lacked information on clinical variables such as hemoglobin values, blood loss, and local protocols on blood transfusion and tranexamic acid use. Particularly, missing information on local protocols may be partially addressed by the mixed-effects statistical modeling approach, which accounts for unspecified hospital-level effects. Additionally, no information was present on drivers of tranexamic acid use, for example, physician preferences or local protocols, specifically if avoidance was purposeful in high-risk patients. However, observed tranexamic acid utilization patterns and confirmatory results from sensitivity analyses suggest a minimal role of this limitation. Additionally, the majority of our subgroups and outcomes of interest were defined based on ICD-9 codes, which may have underestimated true comorbidity burden or complication risk. This may have been additionally burdened by potential inaccuracies or underestimations in “present-on-admission”⁴¹  coding. However, we do not expect ICD-9 and present-on-admission coding to be dependent on billing for tranexamic acid. Last, although our data provide a unique look into tranexamic acid use in daily clinical practice, this remains an observational study, and therefore we can only infer associations and not causation. In conclusion, among 27,890 to 45,952 high-risk patients undergoing elective lower-extremity joint replacement surgery, we found that tranexamic acid use was not associated with increased odds of thromboembolic and ischemic complications.

Support was provided solely from institutional and/or departmental sources.

The authors declare no competing interests.