**1. Introduction**

Veno-arterial extracorporeal membrane oxygenation (V-A ECMO) provides mechanical circulatory support for patients with cardiopulmonary failure [1]. There have been exponential increases in ECMO use and survival rates since 2009 [2]. However, infection is still a common complication during ECMO because it requires the use of percutaneously inserted devices with large-diameter catheters, and critically ill patients themselves are generally

**Citation:** Kang, S.; Yang, S.; Hahn, J.; Jang, J.Y.; Min, K.L.; Wi, J.; Chang, M.J. Dose Optimization of Meropenem in Patients on Veno-Arterial Extracorporeal Membrane Oxygenation in Critically Ill Cardiac Patients: Pharmacokinetic/Pharmacodynamic Modeling. *J. Clin. Med.* **2022**, *11*, 6621. https://doi.org/10.3390/ jcm11226621

Academic Editors: Luca Brazzi and Giorgia Montrucchio

Received: 10 October 2022 Accepted: 2 November 2022 Published: 8 November 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

vulnerable to infection [2,3]. One observation study reported that 62.8% had a bloodstream infection within 2 weeks of V-A ECMO, and both gram-positive and gram-negative bacteremia commonly occurred [4]. Biaazrro et al. also reported that the prevalence of infection in adult patients with ECMO was 21%, which was higher than that of children (16%) and neonates (8%). Also, V-A ECMO has higher risk of infectious complications than V-V ECMO [5]. Therefore, successful prevention and treatment of infection by broad-spectrum antibiotics is necessary in patients receiving V-A ECMO is [2,6].

It is well known that V-A ECMO affects the pharmacokinetics (PK) of several drugs [7], altering their volume of distribution (Vd) and clearance (CL) because of inherent physiological changes associated with ECMO and critical illness [8–10]. Non-pulsatile blood flow from V-A ECMO reduces glomerular filtration rate, and consequently reduces the CL of drugs [11]. Patients with profound cardiogenic shock during V-A ECMO commonly need more aggressive volume support for hemodynamic stabilization [12], which widely alters the effect of ECMO treatment on PK parameters. In addition, PK changes in patients receiving ECMO are dependent on the physicochemical properties of the drugs [13]. Therefore, exact predictions of PK changes in V-A ECMO are difficult [14].

One of the commonly used broad-spectrum antibiotics, piperacillin-tazobactam, was studied, and ECMO and CRRT increased, with Vd and the use of ECMO reduced CL [15]. The other study reported that use of ECMO increased both CL and Vd of cefpirome, another broad-spectrum antibiotic. However, studies on the impact of ECMO on meropenem PK showed conflicting results [16,17]. Shekar et al. reported that the CL was reduced during ECMO, but Gijsen et al. said that the use of ECMO did not influence the PKs of meropenem.

Thus, the present study aims to describe the PK profiles of meropenem by comparing patients receiving V-A ECMO with patients after stopping ECMO treatment. In addition, optimal dosage regimens were determined according to individual characteristics by simulating various dosing scenarios in patients on both V-A ECMO and continuous renal replacement therapy (CRRT).

#### **2. Methods**

This prospective cohort study was conducted from May 2016 to January 2019 in the cardiac intensive care unit of Severance Hospital in Seoul, Korea. Adult patients (≥18 years) receiving V-A ECMO and concomitantly receiving meropenem were included in this study. Patients who were allergic to carbapenem or pregnant were excluded. Patients with normal kidney function received 1 g meropenem q8h as an intravenous injection over 20 min as per protocol. Patients with an estimated glomerular filtration rate (eGFR) of less than 30 mL/min/1.73 m2, as calculated by the Modification of Diet in Renal Disease (MDRD) study equation, or patients on CRRT, received 1 g meropenem q12h.

The ECMO system consisted of a centrifugal blood pump with a controller (Capiox® SP-101, Terumo Inc., Tokyo, Japan), a conduit tube (Capiox® EBS with X coating, Terumo Inc., Tokyo, Japan), and an air-oxygen mixer (Sechrist® Industries, Inc., Anaheim, CA, USA). It was connected percutaneously between the femoral vein and peripheral cannulation of the femoral artery. If needed, continuous venovenous hemodiafiltration (CV-VHDF) (Prismaflex®; Gambro Inc., Meyzieu, France) with a Prismaflex® ST100 filter was utilized for CRRT. The ECMO and CRRT settings were recorded.

Data associated with demographics, renal and hepatic functions, blood chemistry, vital signs, blood cell counts, and details of ECMO and CRRT were collected. As allowed by the clinical situation, blood samples were collected during ECMO (ECMO-ON group) through the existing radial arterial line at the following times: pre-dose (0 min); 0.5, 1, 3, and 6 h after meropenem administration; and immediately before the next dose, according to administration frequency (8 h or 12 h). If the patients were administered meropenem after them weaning off of ECMO (ECMO-OFF group), blood samples were collected at the aforementioned times. The actual sampling time was recorded. The blood samples were collected in EDTA-coated tubes and immediately centrifuged (1500× *g* at 4 ◦C for 10 min). The plasma samples were stored at −80 ◦C until analysis.

Meropenem concentrations were measured using liquid chromatography-mass spectrometry (LC-MS, Ultimate 3000 RS-Q-Exactive Orbitrap Plus; Thermo Fisher Scientific, Waltham, MA, USA) in the Yonsei Center for Research Facilities. The plasma samples were deproteinized using acetonitrile with sulfamethoxine as an internal standard. The mixture was vortexed for 10 s, and then centrifuged (10 min at 10,000× *g*), and the supernatant was filtered using a 0.45-μm syringe filter. LC-MS was performed on an Acquity UPLC BEH C18 column (1.7 μm, 2.1 mm × 100 mm; Waters, Milford, MA, USA) with a column temperature of 40 ◦C and a flow rate of 0.4 mL/min. The mobile phase was comprised of solvent A (0.1% formic acid in water) and solvent B (100% acetonitrile) with the following elution gradient maintained at 90% A for 4 min, reduced to 5% A over 10 min, maintained at 5% A for 1 min, increased to 90% A over 0.5 min, and maintained at 90% A for 1.5 min. The lower limit of quantification was 0.1 mg/L. The inter- and intra-assay coefficients of variation were <15%.

The population PK model was conducted based on non-linear mixed-effects modelling using NONMEM (version 7.4.1; ICON Development Solutions, Dublin, Ireland) and Pirana (version 2.9.7; Certara, Princeton, NJ, USA). The Xpose4 package (version 4.6.1; https: //github.com/UUPharmacometrics/xpose4/releases (accessed on 1 March 2019)) in R (version 3.5.3; http://www.r-project.org (accessed on 1 March 2019)) was used to visualize and evaluate the models.

The plasma concentration-time profiles for meropenem were fitted to one-, two-, or three-compartment models using the first-order conditional estimation method with the interaction estimation option. Interindividual variability (IIV) of the PK parameters was evaluated using an exponential variance model assuming a log-normal distribution. Residual unexplained variability (RUV) was tested using an additive, exponential, and combined random error model. The model was selected based on a minimum objective function value (OFV), validity of the estimated relative standard error (RSE), shrinkage of PK parameters, and visual inspection of the goodness-of-fit plot. The likelihood ratio test was performed in the NONMEM program to assess statistical significance between the nested models. A decrease in the OFV of at least 3.84 was judged statistically significant for an added parameter (*p* value < 0.05, χ<sup>2</sup> distribution, degree of freedom = 1). For visual inspection, the goodness-of-fit plot was expressed as the observed concentrations vs. population predictions (PRED) or individual predictions (IPRED), and conditional weighted residuals (CWRES) vs. PRED.

To evaluate the influence of covariates on the meropenem PK parameters, the following potential covariates were tested: demographic variables (sex, age, weight, and height), ECMO-associated variables (during ECMO or weaned off of ECMO and ECMO flow rate [LPM, liters per minute]), CRRT-associated variables (use of CRRT, blood flow rate, CRRT 6 h prior to urine output, dialysate flow rate), complete blood count (absolute white blood cells, red blood cells, hemoglobin, and platelets), renal function (serum creatinine [SCr], blood urea nitrogen [BUN], creatinine clearance (CrCL) estimated via the Cockcroft-Gault equation, and eGFR estimated via the MDRD equation), liver function (alanine transaminase, aspartate aminotransferase, and total bilirubin), biomarkers of inflammation (C-reactive protein and procalcitonin), blood pressure, tympanic body temperature, and social variables (smoking status and alcohol consumption). In addition, to reflect the inherent correlation between patient status and improvement in critical illness between the ECMO-ON and ECMO-OFF groups, we tested the time since ECMO initiation and ECMO termination as an individual covariate. Most data were tested as time-varying covariates, except fixed variables, such as sex, age, and smoking status, which were considered time-independent.

Covariates were evaluated using linear, exponential, power, and proportional models based on the stepwise covariate modelling (SCM) process. If needed, the continuous covariates were centered on their median values. For forward selection, a *p* value < 0.05 (OFV reduction of >3.84) and for backward elimination, a *p* < 0.01 (OFV increase of >6.64) were considered to measure significance. The final covariate model selection was based on biological or clinical plausibility, RSE, shrinkage of PK parameters, a condition number of <1000, and visual improvement in the goodness-of-fit plot.

To evaluate the precision and robustness of the final PK model, an automated sampling importance resampling (SIR) algorithm (sampling = 5000, resampling = 1000, five iterations) and a prediction-corrected visual predictive check (pcVPC) were carried out using the Perl Speaks NONMEM toolkit version 4.9.0. (Uppsala University, Uppsala, Sweden) [18,19]. The medians with 95% confidence intervals for the replicates from the SIR algorithm were compared with the estimated PK parameters from the final model. Furthermore, the simulated pcVPC results with the 5th percentile, median, and 95th percentile curves were visually assessed.

Monte Carlo simulations were performed using the estimated PK parameters to assess the effect of the screened covariates on the predicted meropenem concentrations (*n* = 10,000). Intravenous intermittent infusion (II) over 20 min. and intravenous extended infusion (EI) over 3 h and 6 h were simulated by the following dosage regimens: 1 g q12h, 2 g q12h, 0.5 g q8h, 1 g q8h, and 2 g q8h over a 24-h period since the first meropenem administration. In addition, intravenous continuous infusion (CI) over 8 h (q8h) of 0.5, 1, and 2 g were simulated. The % fT > MIC was determined for each simulated subject by linear interpolation. The PTA was calculated by counting subjects achieving more than 40% fT > MIC and 100% fT > MIC; the dosage scenario that achieved PTA above 90% was considered to be efficient. The MIC, the clinical breakpoint for meropenem, that was used was 2 mg/L for susceptible strains and 8 mg/L for resistant strains according to EUCAST (ver. 10.0, Växjö, Sweden, valid from 1 January 2020).
