Study design
In vitro CRRT clearance model
In vitro CRRT was simulated using a Prismaflex 7.2 control unit (Baxter Healthcare Corporation, Deerfield, IL, USA) in continuous veno-venous hemofiltration (CVVH) and continuous veno-venous hemodialysis (CVVHD) modes using fresh 1.4m2 polyarylethersulfone (PAES; Prismaflex HF1400) and 1.5m2 acrylonitrile (AN69; Prismaflex M150) hemofilter sets for each experiment. One liter of heparinized (20 units/mL) whole bovine blood (Densco Marketing Inc, Woodstock, IL, USA) was heated to 37°C in a water bath and stirred continuously. The Prismaflex circuit was initially primed with 186 mL (HF1400) or 189 mL (M150) of 0.9% sodium chloride per the manufacturer’s operating instructions [44, 45]. Prior to the start of each experiment, blood was then allowed to circulate throughout the system for at least 2 minutes to permit adequate exposure of the hemofilter to blood proteins. The blood flow rate was fixed at 200 mL/min for all experiments while CVVH replacement fluid (PrismaSOL® BGK 2/0; Baxter Healthcare Corporation, Deerfield, IL, USA) and CVVHD dialysate (PrismaSATE® BGK 2/0; Baxter Healthcare Corporation, Deerfield, IL, USA) rates of 2 L/h and 4 L/h were tested with each filter type. During CVVH at 2 L/h, replacement fluid was added 100% pre-filter, 100% post-filter, and at 50% pre-/50% post-filter. During CVVH at 4 L/h, replacement fluid was added at 50% pre-/50% post-filter. All experiments were performed in duplicate in each mode, at each rate, and with each filter for a total of 24 experiments (excluding adsorption experiments).
Apixaban (Eliquis®; Bristol-Myers Squibb, New York, NY, USA) was reconstituted per manufacturer’s instructions [46]. To account for measured bovine hematocrit of 36.9% (Biologic Resources Laboratory, University of Illinois at Chicago, Chicago, IL, USA), the dose of apixaban added to the central reservoir was adjusted a priori to simulate the mean peak serum concentration (Cmax) observed in healthy adult subjects following a single 5 mg dose of apixaban (~0.104 mg/L) [47]. Urea (Sigma-Aldrich, St. Louis, MO, USA) was also added at 75 mg/L to serve as the control solute.
After at least 1 minute of equilibration, serial pre-filter blood samples were collected in 3.2% sodium citrate tubes (Becton, Dickinson and Company, Franklin Lakes, NJ, USA) at 0, 10, 20, 30, 45, and 60 min post-dose with analogous post-filter blood and effluent samples collected at 10 and 30 min. Blood samples were centrifuged at 1,500 x g for 10 min and the resultant supernatant plasma and ultrafiltrate samples were frozen at -80°C within 30 min of collection until analysis.
Adsorption experiments
To evaluate potential adsorption of apixaban to the hemofilters, the initial CRRT model was modified to create a closed-circuit system. Effluent was rerouted to the central blood reservoir, and 0.9% normal saline was exogenously pumped into the effluent bag via a Masterflex® Peristaltic pump (Cole-Parmer, Vernon Hills, IL, USA) at the same rate to prevent the Prismaflex system from aborting due to the patient blood loss/gain alarm. Serial blood samples were drawn from the central reservoir at 0, 10, 20, 30, 45, 60, 90, 105, 120, 150, and 180 min, centrifuged at 1,500 x g for 10 min, and supernatant plasma was frozen at -80°C within 30 min of collection until analysis. This process was repeated twice in duplicate for a total of 4 experiments.
Protein binding determination
To assess apixaban protein binding in bovine plasma, 4 contrived samples were centrifuged at 2,000 x g for 30 minutes using a Centrifree® Ultrafiltration Device (Merck Millipore Ltd. Tullagreen, Carrigtwohill, Co. Cork, Ireland) with resulting bound and unbound plasma samples frozen at -80°C within 30 min until analysis. This process was repeated twice in duplicate for a total of 4 experiments.
Bioanalytical procedures
Concentrations of apixaban and urea in bovine plasma and effluent were quantified via liquid chromatography-tandem mass spectrometry (Keystone Bioanalytical, North Wales, PA, USA) as previously described [48]. The calibration range of the assay was linear from 0.001-0.2 mg/L (r ≥0.999). The precision and accuracy acceptance criteria for the quality control (QC) samples and calibration standards were ≤15% coefficient of variance (CV) and ±15% relative error (RE) determined at each concentration level.
Pharmacokinetic procedures
Pharmacokinetic parameters for apixaban were estimated from observed pre-filter plasma concentrations via noncompartmental analysis in Phoenix WinNonlin Version 8.1 (Certara USA Inc., Princeton, NJ, USA). Reported parameters included: Cmax, last observed plasma concentration (Clast), elimination rate constant (Ke), half-life (t1/2), apparent volume of distribution (Vd), clearance (CL), and the area under the concentration-time curve (AUC0-∞ and AUC0-last) as determined via the linear up-log down method. As in vitro experiments were performed over a period of one hour, AUC0-last was multiplied by 24 to demonstrate proportional AUC0-24. Calculations for the estimation of apixaban and urea removal from the CRRT circuit were as follows:
- sieving coefficient (SC) = (Cuf / Cpre)
- saturation coefficient (SA) = (2 * Cdialysate) / (Cpre + Cpost)
Where Cuf is the concentration in the ultrafiltrate, Cpre is the concentration from the pre-filter sampling port, Cdialysate is the concentration in the dialysate, and Cpost is the concentration from the post-filter sampling port [49-51].
Clearance by CRRT was then estimated by two distinct methods to ensure accuracy and allow for comparison. The primary method of estimating CLTM was estimated using the AUC0-24 determined via noncompartmental analysis (CLTM by AUC), as previously described. The secondary method, CLTM by sieving/ saturation coefficients (SC/SA), utilized the following equations:
- CLCVVH = (SC * Quf * Qb) / (Qb + Qrep)
- CLCVVHD = (SA * Qd)
Where Quf is the ultrafiltrate or replacement fluid rate flow rate and Qd is the dialysate flow rate. In experiments where replacement fluid was added pre-filter, a dilutional correction factor was incorporated into the clearance equation, with Qb representing blood flow rate and Qrep being the pre-filter replacement fluid rate [51, 52].
Adsorption was calculated as the difference between the total amount of apixaban added to the system and the total amount recovered in the dialysate and plasma after 180 min using the following equation at each sampling time point:
- Adsorption (%) = Ʃ1- [(dose of apixaban added at time zero) / (concentration of apixaban * measured volume in central reservoir)] [53].
Optimal dosing determination
Optimal dosing was calculated to provide a comparable mean AUC value to that achieved following the administration of apixaban 5 mg twice daily for 7 days in healthy subjects (2103.8 mg · h/L) via the equation AUC = Dose / CLT; where CLT = CLTM + CLNR [54]. Here, CLT represents total body clearance, CLTM represents CL via CRRT, and CLNR is non-renal clearance. The value for CLNR (2.52 L/h) was imputed from Phase 2 and 3 studies evaluating apixaban for the treatment or prevention of recurrent VTE and assumed to be constant [55, 56]. Additionally, residual renal function was assumed to be negligible as the majority of critically ill patients with AKI on CRRT have no appreciable residual renal function [52].
Statistical analysis
Data are presented as mean (±SD) or with 95% confidence intervals (95% CI). Continuous data were compared via Student’s t-test. Additionally, one- and two-way ANOVA models with Tukey’s post-hoc tests were built to evaluate statistically significant differences in mean CLTM according to CVVH point of dilution within and between each filter type, respectively. Then, three-way ANOVA models were fit using CLTM as the outcome to evaluate the interaction between CRRT mode, filter type, and flow rate. ANOVA-generated means and 95% CI of CLTM were then used to estimate AUC and optimal total daily doses (TDD) of apixaban during CRRT. Finally, multiple linear regression via backwards stepwise analysis was used to correlate flow rate with mean CLTM while adjusting for covariates (CRRT mode, filter type, point of dilution, and flow rate) and predict optimal dosing regimens across flow rates from 0.5-5 L/h. Model performance was assessed via the adjusted R2 value. Collinearity was assessed via tolerance and variance inflation factor. A P value of ≤0.05 was considered statistically significant in the final model. All statistical analyses were performed using SPSS® Version 26 (IBM Corp, Armonk, NY, USA).