Population pharmacokinetic analysis of doripenem for Japanese patients in intensive care unit

We aimed to construct a novel population pharmacokinetics (PPK) model of doripenem (DRPM) for Japanese patients in intensive care unit, incorporating the clearance of DRPM by continuous renal replacement therapy (CRRT). Twenty-one patients treated with DRPM (0.25 or 0.5 g) by intravenous infusion over 1 h were included in the study. Nine of the 21 patients were receiving CRRT. Plasma samples were obtained before and 1, 2, 4, 6 and 8 h after the first DRPM administration. PPK analysis was conducted by nonlinear mixed effects modeling using a two-compartment model. Total clearance (CLtotal) in the model was divided into CRRT clearance (CLCRRT) and body clearance (CLbody). The final model was: CLtotal (L h−1) = CLbody(non-CRRT) = 3.65 × (Ccr/62.25)0.64 in the absence of CRRT, or = CLbody(CRRT) + CLCRRT = 2.49 × (Ccr/52.75)0.42 + CLCRRT in the presence of CRRT; CLCRRT = QE × 0.919 (0.919 represents non-protein binding rate of DRPM); V1 (L) = 10.04; V2 (L) = 8.13; and Q (L h−1) = 3.53. Using this model, CLtotal was lower and the distribution volumes (V1 and V2) tended to be higher compared to previous reports. Also, Ccr was selected as a significant covariate for CLbody. Furthermore, the contribution rate of CLCRRT to CLtotal was 30–40%, suggesting the importance of drug removal by CRRT. The population analysis model used in this study is a useful tool for planning DRPM regimen and administration. Our novel model may contribute greatly to proper use of DRPM in patients requiring intensive care.

Population pharmacokinetic (PPK) analysis provides PPK parameters consisting of average and variance for a population using the nonlinear mixed effect model (NONMEM) 1 that fits all the drug plasma concentrations of multiple patients 2 . Individual pharmacokinetic (PK) parameters such as clearance (CL) and distribution volume (V d ) are obtained by Bayesian method using PPK parameters 3 . Doripenem (DRPM) is a carbapenem antibacterial agent that has a broad antibacterial spectrum against Gram-positive bacteria, Gram-negative bacteria, and anaerobic bacteria 4 . Since DRPM shows a time-dependent effect based on the pharmacokinetic and pharmacodynamics (PK/PD) theory, bactericidal effect is greater when the blood concentration is maintained above the minimum inhibitory concentration (MIC) for 40% of the time or more 5 .
Several PPK analysis models for DRPM have been reported. The PPK model reported by Bhavnani et al. 6 was constructed using an intravenous 2-compartment model based on plasma DRPM concentrations obtained in a phase 1 DRPM clinical trial. The PK parameters they reported included total CL (CL total ), inter-compartmental clearance (Q), distribution volume of central compartment, and distribution volume of peripheral compartment. In addition, Lee et al. 7 succeeded to build a good PPK model for patients with sepsis by integrating creatinine clearance (Ccr) as a covariate in CL total using an intravenous 1-compartment model. In the PPK analysis reported by Nandy et al. 8 , covariates such as Ccr, body weight (BW), age and race were incorporated using an intravenous 2-compartment model for a wide range of subjects in phases 1-3 trials. They reported differences in DRPM clearance depending on race. Reports suggested that some degree of uniformity of the population including race may be necessary because racial variation may affect the PK of drugs such as distribution, metabolism, and elimination 9,10 . Matsuo et al. 11 reported a PPK model of DRPM targeting Japanese subjects by incorporating Ccr and age as covariates in CL total , using an intravenous 2-compartment model in phase 1 healthy subjects.
A retrospective cohort study conducted by Kumar et al. 12 showed that the death rate increased by 7.6% when administration of antibiotics was delayed by 1 h. Other reports [13][14][15] also suggested that the risk of death in septic shock patients decreased if antibiotics were administered within 1 h. There is a relationship between the time to start of antibiotic administration and death 16 , and international guideline has recommended to administer antibiotics within 1 h after diagnosis 17 . Carbapenem antibiotics are used empirically in the intensive care unit (ICU) for complex and severe infections. Water-soluble drugs with low molecular weight and low protein binding rate, such as carbapenems, are likely to show increased V d and higher CL in critically ill patients 18,19 . In critically ill patients in ICU, CRRT is often performed due to declined kidney function such as acute kidney injury and for removal of inflammatory cytokines. Therefore, clearance of DRPM from the body by CRRT should be considered in patients undergoing CRRT. Using an intravenous 2-compartment model that considers the clearance of DRPM by CRRT, Roberts et al. 20 constructed a PPK model without incorporating covariates for critically ill patients undergoing CRRT. They reported that it was necessary to consider the removal of DRPM by CRRT, since clearance by CRRT (CL CRRT ) contributed to 30-37% of CL total . In addition, Monte Carlo simulation using the PK parameters obtained suggested that the change in distribution volume substantially affected the time above MIC.
In this study, we aimed to construct a novel PPK model incorporating optimal covariates including CRRT in each PK parameter for Japanese ICU patients.

Material and methods
Patients. This study was conducted in accordance with the Declaration of Helsinki. The study was started after obtaining approval from the Ethics Committee in Oita University (Approval No. 613). The subjects in this study consisted of 21 inpatients (a total of 97 samples) in ICU treated with DRPM, who gave written informed consent obtained from either the patients or their legally authorized representatives. Nine of the 21 patients were receiving CRRT. Patients who were administered other carbapenem antibacterial agents before administration of DRPM were excluded.
DRPM at a dose of 0.25 or 0.5 g was given by intravenous infusion over 1 h. At the first administration, plasma samples were obtained from blood sampling before DRPM administration and at 1, 2, 4, 6 and 8 h after the start of infusion. Plasma DRPM concentrations were measured by high performance liquid chromatography (HPLC) according to the procedures we reported previously 21 . In brief, plasma samples were pre-treated by a solid-phase extraction method. The HPLC system (Waters 2695) was used with a Shiseido Capcell Pak C18 MGII column (5 μm, 250 mm × 4.6 mm; Shiseido Co., Tokyo, Japan) and ultraviolet absorbance detection (Waters 2489 UV/ Vis). Separation of DRPM and internal standard was satisfactory, and was free of interfering peaks from the plasma matrix. The limit of quantification (LOQ) for the DRPM assay was 0.5 μg mL −1 , and the calibration curve was linear from 0.5 to 100 μg mL −1 (r 2 = 0.999). Only one sample had a concentration below the LOQ.
Population pharmacokinetics. Analysis of population pharmacokinetics was conducted using nonlinear mixed effects modeling (NONMEM) version 7.3.0 22 . We selected the compartment model using the DRPM plasma concentration-time curve with log-transform, by comparing the 1-compartment and 2-compartment models using objective function value (OFV) and the Akaike criteria (AIC) 23 . Since a conventional 2-compartment model would not include the route of DRPM removal by CRRT, we modified the conventional model to include CRRT clearance. Thus, total clearance in the model was divided into CRRT CL (CL CRRT ) and body CL (CL body ) using the ADVAN6 subroutines from the NONMEM library. We initially regarded this as the base model ( Fig. 1). For each PPK parameter, inter-individual variability was evaluated by the exponential error model, and residual variability was evaluated using the additive error model. Clinically plausible covariates such as Ccr obtained by the Cockcroft-Gault equation 24 , BW, and albumin (Alb) were screened as PK parameters of DRPM. Screening of covariates was performed by addition of the candidate covariate to the base model. An eligible covariate should have a correlation with one PK parameter (correlation coefficient r > 0.6), and furthermore no correlation with other covariates; that is, no multiple collinearity. After addition of the covariate, a reduction in OFV of more than 2.71 for one degree of freedom was considered a statistically significant improvement www.nature.com/scientificreports/ (p < 0.10) based on the χ 2 test. After selection of the covariates by the above forward addition step, the model was refined by backward elimination step. A covariate was included in the model when the significance level p < 0.05 (a reduction in the OFV of more than 3.84 for one degree of freedom based on the χ 2 test) was obtained. This analysis was calculated by the first-order conditional estimation with interaction method.
Model evaluation. Validity of our model was examined using visual predictive check and goodness-of-fit plots 25 . Also, evaluation of reliability and stability of the final model was performed by the bootstrap method. A thousand bootstrap data sets were reconstructed by resampling the subjects from the original data set. The average and standard deviation of parameter estimates obtained from the bootstrap were compared to the estimates of parameters for the final model and standard error obtained from the original data set.

Results
Patients. Patient demographics and relevant clinical data are summarized in Table 1. Among 21 patients, 18 patients were males and three were females. The mean ± standard deviation of Ccr was 68.0 ± 33.4 (mL min −1 ). Nine patients were undergoing CRRT during DRPM treatment, eight of whom had renal indications for CRRT.
Population pharmacokinetics. As shown in Fig. 2, a semi-logarithm plot of DRPM plasma concentration versus time showed two phases (distribution phase and elimination phase). Comparisons of 1-compartment and 2-compartment models using OFV and AIC confirmed that both the OFV and AIC of the 2-compartment model were significantly smaller than those of the 1-compartment model (ΔOFV = 89.8, p < 0.05; ΔAIC = 36.3, p < 0.05). Based on this finding, we adopted the 2-compartment model in this study. In addition, comparison with the conventional 2-compartment model without including CRRT clearance confirmed that OFV decreased significantly in our modified 2-compartment model (ΔOFV = 29.8, p < 0.001). Thus, this model that considers removal of DRPM by CRRT was used as the base model in this study. After covariate selection by the forward addition step, models #8 was selected as the full model ( Table 2). The full model incorporated Ccr in CL body , Alb in CL CRRT , and BW in V 1 . Ccr and BW have been selected as covariates in previous reports 7,8,11,26 , and Alb has been reported to be removed by continuous hemofiltration using a cellulose triacetate membrane 27 . Next, at the backward elimination step, when Ccr was excluded from the full model (model #8), OFV increased significantly ( Table 2).
Model evaluation. The measured DRPM concentrations correlated well with the predicted concentrations both for population and individual predictions (Fig. 4). The conditional weighted residuals (CWRES) and individual weighted residuals (iWRES) distributed uniformly regardless of concentration and time. The above results suggested that the analyses of our model were valid. A comparison of each PPK parameter in the final model with each parameter obtained from 1000 bootstrap samples estimated no large error between the two, and the 95% confidence intervals were relatively small (Table 3). Visual predictive check was performed based on 1000 replicates. As shown in Fig. 5, almost all measured values were within the 95% confidence intervals estimated from our model. Thus, the reliability and stability of each PK parameter are proven and our model is valid.
Model simulation. Using our final model, Monte Carlo simulations were performed to evaluate the probability of target attainment (PTA) for targets of 40% and 100% of the time in which free drug concentrations Table 1. Demographics and relevant clinical data of all patients, those who underwent continuous renal replacement therapy (CRRT) and those who did not undergo CRRT. Data are expressed as number or mean ± S.D. Ccr, creatinine clearance; CRRT, continuous renal replacement therapy; APACHE II score, acute physiology and chronic health evaluation score II; SOFA score, sequential organ failure assessment score.   www.nature.com/scientificreports/ was adjusted so that the daily dose was equal). The PTA was estimated for Pseudomonas aeruginosa (MIC ≤ 2) according to the European Committee on Antimicrobial Susceptibility Testing (EUCAST) 31 . As shown in Table 4, PTA when administered by ExI and CI was higher than that when given by InI in all the simulations. For the target of 40% fT > MIC, all three infusion methods of all dosages achieved over 90% PTA regardless of renal function and CRRT.
For the targets of 100% fT > MIC and 100% fT > MIC × 4, however, InI achieved over 90% PTA only in patients with renal function 0 < Ccr ≤ 30 not using CRRT, while all other renal function and CRRT conditions required ExI or CI to achieve 90% PTA. For patients using CRRT, ExI of a higher dosage was needed to achieve over 90% PTA for 100% fT > MIC × 4 versus 100% fT > MIC when renal function was 0 < Ccr ≤ 30 (500 mg vs. 250 mg or above) and 30 < Ccr ≤ 60 (1000 mg vs. 250 mg or above). For patients not using CRRT, the two dosages administered by ExI for the targets of 100% fT > MIC × 4 versus 100% fT > MIC were either not different (250 mg or above for both in 0 < Ccr ≤ 30) or higher for 100% fT > MIC × 4 (500 mg vs. 250 mg or above in 30 < Ccr ≤ 60; 1000 mg vs. 250 mg or above for 60 < Ccr ≤ 90). Note that a similar trend was observed for InI in patients with 0 < Ccr ≤ 30 not using CRRT (100% fT > MIC × 4 vs. 100% fT > MIC: 1000 mg vs. 250 mg or above).
When administered by CI, all dosages (750-6000 mg) achieved over 90% PTA for 100% fT > MIC and doses of 3000 mg or above achieved over 90% PTA for 100% fT > MIC × 4, for all conditions regardless of CRRT or renal function.

Discussion
In our final PPK model for DRPM, CL tot [CL body(CRRT) and CL body(non-CRRT) ] tends to be lower and V 2 is higher compared to previous reports 6,8,11 . These differences probably arise because the subjects of previous studies included healthy individuals. PK of DRPM is expected to be greatly different between critically ill patients and healthy subjects, since a fluid retention tendency and augmented systemic inflammatory response are observed in the patients 32 .
The pathological conditions of patients in this study were diverse, including bacteremia, septic shock, infective endocarditis, pneumonia, intraperitoneal infection, urinary-tract infection, and poor infection control after surgery or transplantation. Previous reports on PPK models in patients undergoing CRRT focused on acute infections or sepsis 7,26 . Our PPK model was not developed targeting specific diseases and included patients using CRRT, which may more realistically reflect the critical clinical setting in which patients have diverse pathological conditions that could change rapidly. Thus this model may complement existing models, especially for disease conditions that change rapidly or for which PPK models have not been established.
In the final model, Ccr was a significant covariate for CL body . When CRRT is administered, serum creatinine is cleared via two routes: the kidney and CRRT. In the CRRT group, Ccr calculated from serum creatinine is considered inappropriate as an indicator of kidney function for the kidney alone, but appropriate as an indicator of overall renal function for the kidney combined with CRRT. Therefore, Ccr calculated from serum creatinine indicated renal function in different conditions depending on the presence or absence of CRRT. Hence, when used as a covariate in the model, Ccr has to be differentiated depending on the presence or absence of CRRT.   20 reported that the contribution of CL CRRT to CL total was 30-37%. The renal function of the patients undergoing CRRT in this study showed some variations [Ccr; 57.3 ± 8.85 (mean ± SE)]. Therefore, these patients showed CL total of 3.81 ± 0.31 (mean ± SE) and CL CRRT of 1.31 ± 0.19, and the contribution of CL CRRT to CL total was 40.0 ± 3.92%. Our result of the contribution of CL CRRT to CL total was the same as past report, indicating the importance to consider drug removal by CRRT.
Achievement of the maximal bactericidal effect requires at least 40% fT > MIC 33 . However, it would be necessary to aim for more aggressive exposure of 100% fT > MIC or 100% fT > MIC × 4 in critically ill patients 34,35 . Hence, we simulated the above three targets for four dosages using three infusion methods (InI, ExI and CI) for each condition of CRRT and renal function. As shown in Table 4, the PTA of InI was inferior to that of ExI and CI, and more than 90% PTA for 100% fT > MIC and 100% fT > MIC × 4 could not be achieved in most conditions. www.nature.com/scientificreports/     Our report had two limitations. First, we were not able to measure DRPM concentrations in the filtrate or plasma concentrations during blood removal and blood returning. Therefore, we were unable to calculate the CL CRRT from measured data. However, DRPM is removed by simple hemodiafiltration process, since it is unnecessary to consider the adsorption process of DRPM at the cellulose triacetate membrane 27 . Second, this study was conducted in a single facility, and the numbers of subjects and samples were not sufficient. A multicenter joint study is needed to validate the model.
In conclusion, PPK analysis was performed in patients who required systemic management at the ICU and received DRPM for severe infections. Because this population included patients undergoing CRRT, we constructed a model that incorporate the DRPM drug excretion pathway by CRRT. Using this model, Ccr was selected as a significant covariate for CL body . The contribution rate of CL CRRT to CL total was 40 ± 3.92%. Furthermore, the results of Monte Carlo simulation show a possibility that DRPM clearance may be significantly different depending on the presence or absence of CRRT, which may impact the therapeutic effect. The present findings thus suggest that drug removal by CRRT may be important. From the above, our novel model is a useful tool for deciding administration of DRPM and may contribute greatly to further proper use of DRPM in patients requiring intensive care.  www.nature.com/scientificreports/ Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.