Population Pharmacokinetic and Pharmacokinetic/Pharmacodynamic Analyses of Cefiderocol, a Parenteral Siderophore Cephalosporin, in Patients with Pneumonia, Bloodstream Infection/Sepsis, or Complicated Urinary Tract Infection

Cefiderocol is a novel siderophore cephalosporin with antibacterial activity against Gram-negative bacteria, including carbapenem-resistant strains. The standard dosing regimen of cefiderocol is 2 g administered every 8 hours over 3 hours infusion in patients with creatinine clearance (CrCL) of 60 to 119 ml/min, and it is adjusted for patients with <60 ml/min or ≥120 ml/min CrCL.


RESULTS
A population PK model was developed using data of 3,427 plasma cefiderocol concentrations from 516 subjects (Fig. S1 in the supplemental material). A summary of background characteristics for the analysis population is shown in Table 1.
Plasma cefiderocol concentrations were adequately described by a 3-compartment model with a proportional error model for the intraindividual variability. The developed final model contained the effects of CrCL and infection sites (pneumonia [CREDIBLE-CR and APEKS-NP studies], BSI/sepsis, cUTI [CREDIBLE-CR study], or cUTI/ AUP [APEKS-cUTI study]) on CL, body weight on the volume of distribution in the central and peripheral compartments (V 1 and V 2 , respectively), albumin concentration (ALB), and any of the infection sites on V 1 . The model code and parameter estimates are shown in Table S2 and Table 2, respectively. Goodness-of-fit (GOF) plots for the final model demonstrated good fitting to the data without any bias (Fig. S2). Prediction-corrected visual predictive check (pcVPC) indicated that the model well captured the central tendency and variability of the observed data (Fig. 1). The parameter estimates for the final model were comparable to the median of bootstrap estimates ( Table 2), suggesting the robustness of the final model.
CrCL was the most significant covariate on cefiderocol PK, as expected from the previous analysis (12). The CL of cefiderocol was assumed to increase following the power model for up to 150 ml/min CrCL with constant CL for $150 ml/min CrCL. The CrCL cutoff value of 150 ml/min was selected based on visual inspection of the relationship between CL and CrCL (Fig. S3). Then, the appropriateness of the selected value (150 ml/min) was confirmed by testing three cutoff values of 120, 150, and 180 ml/min and comparing their model fitting based on the values of objective function (OBJ). A negative correlation between ALB and V 1 was observed. The CL in patients with pneumonia, BSI/sepsis, and cUTI (CREDIBLE-CR study) was comparable to that in subjects without infection. In contrast, the CL in patients with cUTI/AUP in the APEKS-cUTI study was 27% higher than that in subjects without infection, which was consistent with the previous analysis (12). The V 1 in infected patients was suggested to be 39% higher than that in subjects without infection.
The maximum concentration (C max ) and daily area under the concentration-time curve (AUC) calculated using empirical Bayesian estimation overlapped among infection sites ( Fig. 2A). For pneumonia patients, the estimated C max and AUC were similar between the patients with and without mechanical ventilation, as shown in Fig. 2B.
A PK/PD analysis was conducted using the data from 60 patients in the CREDIBLE-CR study and 97 patients in the APEKS-NP study. Total numbers of isolated pathogens   at baseline in the CREDIBLE-CR and APEKS-NP studies were 77 and 122, respectively, and approximately 30% of the patients were infected by more than one Gram-negative pathogen. The range (median) of MIC of the isolated Gram-negative pathogens was #0.03 to 64 mg/ml (0.25 mg/ml) in both studies (Table S3). The %fT .MIC was 100% in 97% of the patients in both studies. No clear PK/PD relationship was found for any of the outcomes or vital status. This is because the %fT .MIC was 100% in most of the patients in the phase 3 studies (Fig. 3). The PTAs for 75% fT .MIC and 100% fT .MIC were calculated in the simulated patients with different infection sites and renal function groups. The PTA for 75% fT .MIC was .95% against MICs #4 mg/ml regardless of infection site or renal function group (Tables 3 and 4). The PTA even for 100% fT .MIC was .90% against MICs #4 mg/ml for all of the infection sites and renal function groups except for the normal renal function in BSI/sepsis patients (85%). The PTA integrated with all renal function groups is shown in Fig. 4 with the MIC distributions combined from 3 consecutive (2014 to 2016) multinational surveillance studies (20). Regarding the integrated PTA, the highest MIC value achieving .90% PTA was 8 mg/ml for 75% fT .MIC regardless of infection site, and it was 4 mg/ml even for 100% fT .MIC .

DISCUSSION
This is an updated report to the previously published research (10,12) for population PK and PK/PD analyses of cefiderocol by integrating the newly available data in patients with pneumonia, BSI/sepsis, and cUTI caused by Gram-negative pathogens from two phase 3 studies. The updated results based on the target patient population would provide useful information to understand the PK and PK/PD characteristics of cefiderocol in clinical practices.
In the population PK analysis, CrCL was the most significant covariate for cefiderocol PK, as expected, since cefiderocol is mainly excreted via the kidneys (9, 10, 12). The developed model suggested that cefiderocol CL increases following the power model for up to 150 ml/min CrCL with a constant for $150 ml/min CrCL. For another cephalosporin, ceftazidime of ceftazidime-avibactam, the relationship between CL and CrCL was assessed using CrCL cutoff value of 100 ml/min in the population PK analysis, and the slope of CL to CrCL for $100 ml/min CrCL was found to be much smaller than that for ,100 ml/min CrCL (21), which is consistent with the results in this research. The predictability of the Cockcroft-Gault equation for high CrCL (i.e., $120 ml/min) is considered low since the number of data points for CrCL $120 ml/min was limited for the equation development (8), and the Cockcroft-Gault equation was reported to overestimate renal function for a high CrCL range (22). The low predictability for high CrCL is one of the possible reasons for the relationship between cefiderocol CL and CrCL with two slopes. However, it could be concluded that the cefiderocol PK could be successfully modeled with the developed population PK model, which could be used for subsequent simulation works since the pcVPC plots (as shown in Fig. 1) suggested that the FIG 1 Prediction-corrected visual predictive check for final model by study and renal function group. Results for 500 simulations. Renal function groups defined by CrCL were as follows: augmented, $120 ml/min; normal or mild, 60 to ,120 ml/min; moderate, severe, or end-stage renal disease (ESRD), 5 to ,60 ml/min. Semilog scale. Solid line, observed median; dashed line, observed 2.5th and 97.5th percentiles; dark-gray shaded area, model-predicted 95% confidence interval of median; gray shaded area, model predicted 95% confidence intervals of 2.5th and 97.5th percentiles. model adequately described the PK profiles of cefiderocol even in patients with augmented renal function.
Augmented renal function, which leads to faster elimination of antibiotics, is observed especially in critically ill patients, e.g., trauma, sepsis, burns, or pancreatitis (23)(24)(25). In the phase 3 studies, augmented renal function (defined as CrCL $120 ml/ min in this study) was observed in 20% of the patients with comparable proportions for CrCL of 120 to ,150 and $150 ml/min. The estimated free trough concentrations in the phase 3 studies at 2 g every 6 hours (q6h) dosing regimen for the augmented renal function group were higher than 4 mg/ml (minimum, 4.28 mg/ml; geometric mean, 12.7mg/ ml). The geometric mean of estimated daily AUC at 2 g q6h (1,365 mg·h/ml) was similar to that in patients with normal renal function in the phase 3 studies (1,494mg·h/ml) at 2 g q8h. In addition, Monte-Carlo simulations for augmented renal function at 2 g q6h demonstrated that the PTA for 75% fT .MIC was .90% against MICs #8 mg/ml regardless of infection site, and even for 100% fT .MIC , it was .90% against MICs #4 mg/ml. These results suggest that the recommended dosing regimen of 2 g q6h over 3 hours infusion would provide sufficient exposures in patients with CrCL of 120 ml/min or greater.
A negative relationship between V 1 and ALB was incorporated in the final model, suggesting larger V 1 in patients with lower ALB. The increase in volume of distribution with hypoalbuminemia is consistent with the report for PK of antibiotics in critically ill patients (23,24,26). The estimated C max and daily AUC at steady state for the patients in the phase 3 studies were similar between ALB groups (ALB of ,2.8 or $2.8 g/dl) ( Fig. S4 in the supplemental material). Since protein binding data were not available in the subjects used for the analyses, an effect of ALB on the unbound fraction of cefiderocol could not be directly assessed. However, if the unbound fraction was changed depending on albumin concentrations, the total CL of cefiderocol would be changed depending on albumin concentrations since the unbound fraction of cefiderocol is readily excreted via kidney. The fact that there is no clear difference in total CL depending on albumin concentrations in this study suggested the effect of ALB would not be clinically relevant to the exposure to cefiderocol. The effects of infection sites on CL and V 1 were assessed in the population PK analysis. The CL in patients with cUTI/AUP in the APEKS-cUTI study was 27% higher than that in subjects without infection, which was consistent with the previous analysis (12), while that in patients with pneumonia, BSI/sepsis, and cUTI (CREDIBLE-CR study) was comparable to that in subjects without infection. The V 1 of cefiderocol in the patients with any infection site was 39% higher than that in subjects without infection. These results were consistent with the reports suggesting increased CL and volume of distribution of ceftolozane in cUTI patients (27) and increased volume of distribution of antibiotics in critically ill patients (23,24,26). Although the AUC for patients with cUTI/AUP in the APEKS-cUTI study was slightly lower than that for the other patients, including cUTI patients in the CREDIBLE-CR study, the estimated C max and AUC overlapped among the infected patients ( Fig. 2A). Therefore, the effect of infection sites was considered not to be clinically relevant on the exposure to cefiderocol. The patients' background characteristics were different among the APEKS-cUTI, CREDIBLE-CR, and APEKS-NP studies, including disease severity and renal/hepatic function as well as the

Cefiderocol Population PK in Infected Patients
Antimicrobial Agents and Chemotherapy selection of carbapenem-resistant infections in the CREDIBLE-CR study. The effect of infection site could not be distinguished explicitly from these factors in the population PK analyses since they were confounded. In addition, the pathophysiological reason for the effect of the infection site has not been identified, which is the limitation of the developed population PK model. There have been reports for a changed volume of distribution of antibiotics in patients with mechanical ventilation, although the estimated influences were variable (21,23,28). The volume of distribution of ceftazidime in nosocomial pneumonia patients with ventilation (NPv) was 30% higher than that in non-NPv patients (21), while the V 1 of ceftazidime in intensive care unit patients with ventilation was about half of that in patients without ventilation (28). As for cefiderocol, mechanical ventilation was not a significant covariate on CL or V 1 . In addition, the estimated C max and AUC were similar between the pneumonia patients with and without mechanical ventilation in the phase 3 studies (Fig. 2B). Therefore, it could be concluded that the effect of ventilation would not be clinically relevant to cefiderocol PK. The %fT .MIC was 100% in 97% of the patients in the phase 3 studies, suggesting adequate exposure to cefiderocol was achieved against MIC of causative Gram-negative pathogens (#0.03 to 64 mg/ml; MIC 90 of 2 mg/ml in both CREDIBLE-CR and APEKS-NP studies). The geometric means (range) of estimated free trough concentrations were 16.3 (2.91 to 84.8) mg/ml in the CREDIBLE-CR study and 12.7 (0.856 to 89.5) mg/ml in the APEKS-NP study. Based on the Monte-Carlo simulation, the PTA for 75% fT .MIC was .95% against MICs #4 mg/ml regardless of infection site or renal function. The PTA even for 100% fT .MIC was .90% against MICs #4 mg/ml for all of the infection sites and renal function groups except for the normal renal function in BSI/sepsis patients (85%). For PTA calculations, the target 75% fT .MIC was selected as the mean value achieving a bactericidal effect (1 log 10 reduction) in animal infection models, and 100% fT .MIC was used as a very conservative target in consideration with variations in the estimated %fT .MIC among pathogens in animal infection models (5) (20). These studies also support that cefiderocol has antibacterial activity against more than 90% of meropenem-nonsusceptible strains with MICs of #4mg/ml, and the recommended dose regimens would provide sufficient exposure against their causative pathogens. No clear PK/PD relationship was found for any of the outcomes or vital status, which was because the %fT .MIC was 100% in most of the patients in the phase 3 studies. The eradication rates were 33% to 44%, even at 100% fT .MIC in the phase 3 studies. Most microbiological outcomes were indeterminate. Microbiological eradication in a population as complex as that enrolled in the phase 3 studies is often confounded by nonstudy antibiotics, missing data, and the continued presence of foreign body devices such as endotracheal tubes. For critically ill patients, higher target concentrations (e.g., 4-fold MIC) were considered a PK/PD index as reported for b-lactams (15). Even for 4fold MIC as a target, %fT .MIC (%fT .4ÂMIC ) was 100% in 83% of the patients in the phase 3 studies and no PK/PD relationships with %fT .4ÂMIC were found (data are not shown).
In the CREDIBLE-CR study, the estimated C max and daily AUC of cefiderocol in death cases (n = 18) were 1.4-fold of those in survival cases (n = 54) (Fig. S5). Preclinical concentration-dependent toxicology studies suggest that the no-observed-adverse-effect level of exposure is 9-fold of that achieved at the standard dosing regimen (2 g q8h) (29). The AUC values in two death cases were less than 2-fold of the maximum AUC in survival cases in the CREDIBLE-CR study, and they did not reach the level that might be associated with increased risk of toxicity based on toxicology studies (29). The causes of death in the two patients were considered to be related to exacerbation of the underlying illness and infection and/or were complicated with a history of shock within 31 days at the time of randomization (18).
Epithelial lining fluid (ELF) is an important consideration for the treatment of patients with pneumonia. In a very recent study, ELF concentrations were determined from mechanically ventilated patients with bacterial pneumonia (30), and ELF PTA was calculated based on these data as well as ELF concentrations in healthy subjects (31) using a PK steady state was assumed. Shaded area indicates .90%. There were 1,000 simulated patients in each simulation scenario. Body weight was assumed to be log-normal distributed with mean of 72.6 kg and CV of 30%. Albumin was assumed to be log-normal distributed with mean of 2.8 g/dl and CV of 30%. Augmented, CrCL of .120 ml/ min (120 to ,150 = 50%; .150 = 50%); normal, CrCL of 90 to ,120 ml/min; mild, CrCL of 60 to ,90 ml/min; moderate, CrCL of 30 to ,60 ml/min; severe, CrCL of 15 to ,30 ml/min; ESRD (end-stage renal disease), CrCL of 5 to ,15 ml/min. intrapulmonary PK modeling (32). The results suggested that sufficient drug exposures could be achieved in ELF in all renal function groups for 100% fT .MIC for an MIC of #4mg/ml. The breakpoints of cefiderocol are inconsistent among the agency/organization (FDA, CLSI, and EUCAST) (33; https://www.fda.gov/drugs/development-resources/ antibacterial-susceptibility-test-interpretive-criteria; http://www.eucast.org/clinical _breakpoints/). For example, the breakpoints for Enterobacterales determined by FDA and CLSI are 4/8/16 mg/ml (susceptible/intermediate/resistant), while those by EUCAST are 2/4 mg/ml (susceptible/resistant). More details were discussed by Simner et al. and Yamano et al. (34,35).
In summary, the developed population PK model adequately described plasma cefiderocol concentrations in subjects without infection and patients with pneumonia, BSI/sepsis, and cUTI/AUP. CrCL was the most significant covariate on cefiderocol PK. In the phase 3 studies, the %fT .MIC was 100% in almost all of the patients (97%), including the patients with augmented renal function, ventilation, and/or were critically ill in the intensive care unit. Adequate plasma exposure to cefiderocol can be achieved at the recommended dosing regimen of 2 g q8h over 3 hours infusion and the regimens adjusted based on renal function in patients with pneumonia, BSI/sepsis, or cUTI caused by Gram-negative pathogens, including carbapenem-resistant strains.

MATERIALS AND METHODS
Data for analyses. Plasma cefiderocol concentration data were collected from 115 pneumonia patients in the APEKS-NP study (https://www.clinicaltrials.gov identifier NCT03032380) (19), 72 patients with pneumonia, BSI/sepsis, or cUTI in the CREDIBLE-CR study (https://www.clinicaltrials.gov identifier NCT02714595) (18), 238 patients with cUTI/AUP in the APEKS-cUTI study (https://www.clinicaltrials.gov identifier NCT02321800) (17), and 91 subjects without any infection in phase 1 studies (9, 36) as shown in Table S1 in the supplemental material. The pneumonia patients enrolled in the APEKS-NP and CREDIBLE-CR studies included patients with hospital-acquired pneumonia, ventilator-associated pneumonia, and health care-associated pneumonia. Population PK models of cefiderocol were previously developed using the data without phase 3 studies (10,12). In this study, a population PK model was developed using the updated data set with additional data from patients with pneumonia, BSI/sepsis, or cUTI caused by Gram-negative pathogens, including carbapenem-resistant pathogens.
Plasma concentration data for 32 patients who received hemodialysis in the phase 3 studies were excluded from the analysis. Six concentration data values in the phase 3 studies were considered to be anomalous and excluded from the analysis since they were approximately 10-fold higher than the C max following a 2-g single dose infused over 3 hours in a phase 1 study. There were 363 plasma concentrations below the limit of quantification (BLQ) excluded from the analysis. In phase 1 studies, most BLQ data (334 out of 353 plasma concentrations) were predose or $24 hours postdose when plasma concentrations had been expected to be zero or very low. The rest of the 19 BLQ data were 12 to 24 hours postdose at lower doses of 100, 250, and 500 mg. In the CREDIBLE-CR and APEKS-cUTI studies, data of 9 plasma concentrations in 3 patients (data from 3 plasma concentrations each) were BLQ at all sampling points, which were considered to be anomalous. In the APEKS-NP study, BLQ data were observed at 1 point (just prior to the start of infusion). This exclusion of BLQ data would not affect the results of modeling because the developed model predicted plasma concentrations reasonably for any study and renal function group (Fig. 1). The detail for the excluded data in the APEKS-cUTI and phase 1 studies were described in the previous reports (10,12).
Consequently, the population PK model was developed using a total of 3,427 plasma concentrations from 516 subjects, 1,861 plasma concentrations from 91 uninfected subjects, and 1,566 plasma concentrations from 425 patients with infection (Fig. S1). The PK/PD analysis was conducted in the patients who had data for MIC of causative Gram-negative pathogens and microbiological or clinical outcomes following cefiderocol dosing, 60 patients in the CREDIBLE-CR study, and 97 patients in the APEKS-NP study. The total numbers of isolated pathogens at baseline in the CREDIBLE-CR and APEKS-NP studies were 77 and 122, respectively.
Bioanalytical method. A bioanalytical method for the determination of plasma total cefiderocol concentrations was validated where the lower limit of quantification was 0.1 mg/ml (9). Composite plasma samples mixed with 0.2 mol/liter ammonium acetate (pH 5) in a 1:1 volume ratio were prepared and analyzed by a validated liquid chromatography-tandem mass spectrometry assay. The assay was linear from 0.1 to 100 mg/ml, and the precision and accuracy levels were 1.2% to 13.4% and 27.0% to 7.0%, respectively.
Population pharmacokinetic analysis. A 3-compartment model was initially tested as a structural PK model based on the previous analysis (10,12). An interindividual variability for PK parameters was assumed to follow a log-normal distribution and could be modeled with an exponential error model. A model for intraindividual variability was selected from a proportional error model or a combined error model (additive error plus proportional error model).
A covariate model was developed to identify influencing covariates on cefiderocol PK. The effects of the following covariates on CL were tested; CrCL was calculated by the Cockcroft-Gault equation (8), body weight, age, sex, aspartate aminotransferase, alanine aminotransferase, total bilirubin, ALB, race, infection site (no infection, infection with pneumonia in the APEKS-NP and CREDIBLE-CR studies, infection with BSI/sepsis, infection with cUTI in the CREDIBLE-CR study, or infection with cUTI/AUP in the APEKS-cUTI study), and ventilation (mechanical ventilation during PK sampling). Age, sex, ALB, race, infection site, and ventilation were also tested as covariates on V 1 , and body weight was tested as a covariate on V 1 , V 2 , and intercompartmental clearance (Q 2 ).
The effect of CrCL on CL was initially tested using a power model, a piecewise linear model, and a power plus linear combination model with a CrCL cutoff value of 150 ml/min based on the visual inspection of the relationship between CL and CrCL (Fig. S3). The power plus linear combination model was selected based on OBJ, and the slope of CL to CrCL for CrCL of $150 ml/min was extremely small (,0.0001). Therefore, the models in which CL was assumed to be constant for CrCL values of $120, 150, or 180 ml/min were tested, and the CrCL cutoff value of 150 ml/min was selected based on the model fitting assessed by OBJ. Next, the effect of body weight on the PK parameters was tested based on the physiological aspect. After incorporating CrCL and body weight into the model, the other covariates were tested using a univariate regression analysis as screening. The significance level of 0.01 based on x 2 test (P , 0.01) was used for inclusion of covariates into the model.
After incorporating all covariates which were statistically significant in the screening, an inferential assessment and stepwise backward deletion were performed to refine the model. In the inferential assessment, the ratio of parameters and the 95% confidence interval were calculated based on the parameter estimate and standard error and compared with a clinically insignificant range, 0.80 to 1.25, to evaluate the impact of covariate effect. In the stepwise backward deletion, the significance level of 0.001 based on x 2 test (P , 0.001) was used for construction of a final model.
The developed population PK model was evaluated based on GOF plots. The predictive performance was also evaluated by the pcVPC (37) with 500 simulation runs. In addition, the model robustness was evaluated by a bootstrap technique (38). Resampling from the original data set was conducted for generating 300 bootstrap data sets, and PK parameters were estimated for each of the data sets using the final model. The median and 95% CI of the bootstrap estimates were compared to the parameter estimate for the final model.
Pharmacokinetic/pharmacodynamic analysis. The C max and daily AUC for infected patients were calculated using post hoc PK parameters for the final model. Individual %fT .MIC was calculated based on the MIC of causative Gram-negative pathogens, and the simulated steady-state free plasma concentrations were calculated using an unbound fraction of 0.422 (16).
Relationships of %fT .MIC with microbiological outcome and clinical outcome at test of cure and vital status on day 28 from the start of treatment were evaluated using data in the phase 3 studies. The data of "eradication" for microbiological outcome, "clinical cure" for clinical outcome, and "survival" for vital status were treated as positive outcomes. The data of "persistence" and "indeterminate" for microbiological outcome, "clinical failure" and "indeterminate" for clinical outcome, and "death" for vital status were treated as negative outcomes. In cases where more than one causative pathogens were detected, the pathogen with highest MIC was used to evaluate the relationships with clinical outcome and vital status, while the MICs of each pathogen were used to evaluate the relationship with microbiological outcome.
Monte-Carlo simulations. Monte-Carlo simulations were performed to calculate PTA for 75% fT .MIC and 100% fT .MIC for patients with pneumonia, BSI/sepsis, and cUTI. The simulation for cUTI patients was performed using the parameters for cUTI in the CREDIBLE-CR study to assess the PTA for the target % fT .MIC in critically ill patients with infection caused by carbapenem-resistant Gram-negative pathogens. A thousand virtual patients for each infection site (pneumonia, BSI/sepsis, or cUTI) were generated by simulating CrCL, body weight, and ALB, which were significant covariates in the population PK analysis. The PTA was calculated by infection site and renal function group. The integrated PTA was also calculated by weighting proportions of patients in each renal function group based on the distribution of CrCL in the phase 3 studies (CrCL of $120 ml/min, 20.3%; CrCL of 90 to ,120 ml/min, 15.0%; CrCL of 60 to ,90 ml/min, 24.6%; CrCL of 30 to ,60 ml/min, 32.6%; CrCL of 15 to ,30 ml/min, 4.8%l CrCL of 5 to ,15 ml/ min, 2.7%). The %fT .MIC was calculated against an MIC range of 0.25 to 16 mg/ml. The dose regimen for the simulation was set as follows: 2 g q6h for augmented renal function with $120ml/min CrCL, 2 g q8h for normal renal function and mild renal impairment with CrCL of 60 to ,120 ml/min, 1.5 g q8h for moderate renal impairment with CrCL of 30 to ,60 ml/min, 1 g q8h for severe renal impairment with CrCL of 15 to ,30 ml/ min, and 0.75 g q12h for end-stage renal disease (ESRD) with CrCL of 5 to ,15 ml/min.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, PDF file, 1.2 MB.