Tranexamic Acid Dosing for Cardiac Surgical Patients With Chronic Renal Dysfunction: A New Dosing Regimen

Angela Jerath, FRCPC, FANZCA, MBBS, BSc; Qi Joy Yang, MSc; K. Sandy Pang, PhD; Nikita Looby, MSc; Nathaly Reyes-Garces, MSc; Tijana Vasiljevic, BSc; Barbara Bojko, PhD; Janusz Pawliszyn, PhD; Duminda Wijeysundera, PhD, FRCPC; W. Scott Beattie, PhD, FRCPC; Terrence M. Yau, PhD, FRCSC, MD, MSc, BA; Marcin Wąsowicz, FRCPC, PhD, MD


Anesth Analg. 2018;127(6):1323-1332. 

In This Article


Study Design

This single-center prospective cohort study enrolled patients undergoing cardiac surgery with CPB from August 2011 to April 2015. The article adheres to the appropriate Enhancing the QUAlity and Transparency Of health Research (EQUATOR) guidelines. We received Health Canada, Institutional Ethics Board approval and written patient consent. Study eligibility was adult patients (>18 years of age) with stages 1–5 CRD defined by the Kidney Disease Outcome Quality Initiative classification (KDOQI).[19] Exclusion criteria were TXA allergy, preexisting coagulopathy, pregnancy, renal transplant recipients, advanced liver disease (2-fold or greater elevation in liver enzymes), and taking contraceptives or tretinoin (thrombotic risks are raised with TXA). In keeping with our institutional practice, we defined 2 study populations based on the postoperative bleeding risk that received separate dosing regimens. A "low-risk" group underwent aortocoronary bypass or single-valve repair/replacement. A "high-risk" group underwent complex redo, multiple valve and combined aortocoronary bypass with valve repair/replacement or aortic procedures. We aimed to recruit up to 6 patients in each CRD stage for both study groups. Given that we previously conducted and reported the pharmacokinetic analysis in high-risk stage 1 CRD patients, we therefore recruited stages 2–5 CRD high-risk patients during this study.[10]

Study Conduct

As per our institutional practice, the low-risk group received a single 50 mg/kg TXA bolus after induction of anesthesia. The high-risk group received Blood Conservation Using Anti-fibrinolytics Trial (BART) TXA regimen, consisting of 30 mg/kg bolus infused over 15 minutes after induction, followed by 16 mg/kg/h infusion until chest closure with a 2 mg/kg load within the pump prime. The BART TXA infusion regimen was developed in 2002 by Dowd et al.[15] This study assessed 3 TXA dosing regimens and subsequently developed a TXA dosing infusion regimen to achieve a plasma TXA concentration of 100 mg/L and maximal antifibrinolysis. This dosing regimen was generated using a computer-simulated algorithm that was never tested in vivo. This TXA dosing regimen was subsequently used in the BART in 2008, which was designed to compare the safety and efficacy of TXA, aprotinin, and ε-aminocaproic acid.[1] This study resulted in Food and Drug Administration removal of aprotinin for cardiac surgical patients. Perioperative care was standardized for all patients. Anesthesia was induced by using fentanyl (10–20 μg/kg), midazolam (0.05–0.1 mg/kg), rocuronium (0.6–1 mg/kg), and propofol (<1 mg/kg) and maintained with sevoflurane or isoflurane (0.6–1 minimum alveolar concentration [MAC]) and propofol infusion at 50 μg/kg/min during CPB. Patients were anticoagulated by using heparin (400 U/kg) to achieve an activated clotting time >480. CPB circuit was primed using 1.1 L Ringer's lactate with 50 mL of 20% mannitol. Standard CPB management included hypothermia drift to 34°C, mean arterial pressures between 50 and 70 mm Hg, hematocrit >0.24, and α-pH stat. Post-CPB, anticoagulation was reversed with 1-mg protamine per 100 U of heparin. Patients with severe (stage 4 or 5) CRD received dialysis on bypass and the intensive care unit (ICU) to manage hyperkalemia and/or volume excess.

Serial blood samples were drawn at defined time points for measuring plasma TXA concentrations to facilitate pharmacokinetic modeling (see below for details). Samples were collected in standard citrate tubes and blinded to the analytical laboratory. The citrate tubes were placed on ice, centrifuged at 2000g for 15 minutes at 4°C, and the supernatant stored at −70°C until analysis. TXA was extracted and measured using solid-phase microextraction and liquid chromatography-tandem mass spectroscopy/tandem mass spectroscopy. This methodology was previously described and validated.[11]


Primary outcome measured serial plasma TXA concentrations to enable construction of pharmacokinetic profiles for all 5 CRD stages in low- and high-risk groups. Previous data identified that plasma concentration of 100 mg/L provides near 100% inhibition and maximal antifibrinolysis.[10,11] This level constituted the therapeutic threshold for TXA dose adjustment.

Descriptive clinical outcomes were postoperative seizures, blood loss, bleeding requiring chest reexploration or tamponade release, transfusion, ischemic-thrombotic complications (deep venous thrombosis, pulmonary embolism, myocardial ischemia, and stroke), ventilation duration, in-hospital mortality, and length of ICU and hospital stay. Routine electroencephalography (EEG) is not performed in our institution to monitor subclinical seizure activity and was not the study's primary focus. All clinical seizure activity was documented, followed by a neurological assessment, brain computed tomography scan, and EEG at the discretion of the specialist team. Blood hematological and renal (glomerular filtration rate, creatinine) markers were measured plus postoperative creatinine clearance (CL) using 24-hour urine collections when permissible. Patients were followed until hospital discharge.

Pharmacokinetic and Statistical Analysis

To assess the pharmacokinetic profile of TXA in cardiac surgical patients, serial blood samples were drawn at baseline (pre-TXA administration) and then in 5 distinct phases of the cardiac surgical procedure, which matched the pharmacokinetic analysis. Phase I reflects the loading dose, phase II is sternotomy to pre-CPB, phase III is on-CPB period, phase IV is post-CPB to chest closure, and phase V reflects the washout period (samples taken up to 12 hours, for details, please refer to Yang et al,[18] 2015). Blood samples were taken for a total of 12 hours, spanning the post chest closure duration that more than doubled the half-life of TXA.

Compartment Model Fitting to Data. Pharmacokinetic analysis, estimation of CLs, and volumes of distribution were performed using the previously developed 2-compartment model (see Supplemental Digital Content, Figure 1, The extracorporeal (CPB) compartment was turned on during CPB (phase III) only, with the assumption that fluid in the extracorporeal compartment (1.1 L) was returned to patients at the end of CPB. The flow rate between the CPB circuit and the central compartment was assigned as 5 L/min.[18] Data from each CRD stage (stages 1–5) and risk group (low versus high risk) were fitted individually using ADAPT5 (BMSR version 5; USC, Los Angeles, CA); mass balance equations were previously reported to examine whether pharmacokinetic parameter estimates are different for various CRD stages and risk groups.[18] The computational algorithm used was the maximum likelihood solution via the expectation/maximization algorithm (MLEM), and the error variance (VAR) function used was VAR i =[S 1+S 2×g(S,ti )]2, where S 1 and S 2 are the intercept and slope of standard deviation of variance function, respectively. The difference in the estimates was compared using 5 × 2 two-way analysis of variance, with main effects defined as CRD stages 1–5 and risk groups (low versus high). Additionally, the correlation between total TXA plasma CL and estimated glomerular filtration rate (eGFR) was studied, and the coefficient of determination (r 2) and the corresponding P value were calculated by Sigma Plots (Systat Software, San Jose, CA).

Simulations. A series of simulations were performed for each CRD stage and dosing groups according to our developed 2-compartmental model and reported pharmacokinetic parameters.[18] The simulated profiles were based on the fitted parameters (Table 1) and suggested bolus loading dose and maintenance infusion rate (Table 2) for the high-risk patient group among stages 2–5. The 2-compartmental model was used, and the area under the curve to time infinity (AUC) was calculated using trapezoidal rule and added to the extrapolated area (Clast/terminal slope). The plasma CL (dose/AUC) was plotted against the baseline eGFR using scatterplot. The coefficient of determination (r 2) and the corresponding P value were calculated by Sigma Plots.

NONMEM Population Modeling and Covariates. The composite data (with all 5 CRD stages in the 2 risk groups) were used for population analysis to evaluate the important covariate that can explain the between-subject variability (BSV) using NONMEMVII (version 7.3; ICON Development Solution, Ellicott City, MD). The base population model was obtained from a previous study.[18] The baseline eGFR was assigned as a continuous covariate and expressed in the form of (eGFR/ )θ, where is the median value.[18] The risk group (low versus high), as a binary or dichotomous covariate, was assigned to be either 1 (high-risk group) or 0 (low-risk group) and can be expressed in form of CLi or , Vi = (1 − RISK)·CLiL or ViL + RISK·CLiH or ViH, where CLiL or ViL and CLiH or ViH represent pharmacokinetic parameters for the low- and high-risk group, respectively. Because the weight-normalized dose was used, body weight was not considered as a covariate. Model I is the basic model with consideration of covariates, then eGFR on CL, or CL and V1, or CL and CL12 (same as CL21), then eGFR on CL and risk category (low versus high) on V1, or eGFR on CL and risk category on V1 and CL12, and finally, eGFR on CL and risk category on V1, CL12, and V2. Improvement of the fit was examined after the incorporation of each covariate combination to certain pharmacokinetic parameters. Decrease in the objective function values was compared using −2 times the log-likelihood ratio test with a stringent α level of .005 for significance. The model with the best improvement in goodness of fit was selected as the final model. The model fit was examined with diagnostic plots (observation versus prediction), and unexplained variability (ETAs) in CLi or Vi versus covariates (baseline eGFR and risk groups) plots were used to guide the covariate model-building process.

Descriptive clinical data were reported using median (interquartile range) for continuous variables. Clinical categorical variables were reported as frequencies (percentages). In keeping with other pharmacokinetic studies, the sample size was deemed sufficient for modeling.[15,20–22] Analyses were conducted using SAS v9.4 (SAS Institute Inc, Cary, NC). To control for 5 comparisons, a P value <.01 (conventional P value .05 divided by 5) was considered statistically significant after Bonferroni correction.