Model-Informed Ceftazidime-Avibactam Dosing Stratified by Renal Function and Renal Replacement Therapy Status in Critically Ill Patients: A Prospective Population Pharmacokinetic and Monte Carlo Simulation Study

Abstract

Introduction

Carbapenem-resistant organisms (CROs) threaten critically ill patients, requiring model-informed ceftazidime-avibactam (CAZ-AVI) dosing by renal function and renal replacement therapy (RRT) status.

Materials and Methods

Thirty-three patients were prospectively enrolled at one center (18 RRT; 15 non-RRT). RRT settings were recorded. Samples were collected at 0, 1, 2, 4, 6, and 8 h after first-dose and steady-state infusions. Population pharmacokinetic modeling used NONMEM. Monte Carlo simulations evaluated joint probability of target attainment (PTA) for minimum inhibitory concentration (MIC)-directed therapy and pathogen-specific cumulative fraction of response (CFR) for empirical therapy using BRICS MIC distributions under two targets: (1) CAZ 100% fT > MIC and AVI 100% fT > 1 mg/L; and (2) CAZ 50% fT > 4MIC and AVI 50% fT > 1 mg/L. Robust model-informed regimens required joint PTA ≥90% under both targets for MIC-directed therapy and CFR ≥90% under both for empirical therapy; lower values were exploratory.

Results

We analyzed 211 blood samples (120 RRT; 91 non-RRT). Creatinine clearance (CrCL) was retained in the final non-RRT models, whereas no covariate was retained in the final RRT models. RRT-model clearance (CL)/volume of distribution (V_d) estimates were 2.03 L/h/17.0 L for CAZ and 1.77 L/h/14.0 L for AVI; non-RRT estimates were 3.36 L/h/20.5 L and 4.27 L/h/16.0 L, respectively. At MICs ≤8 mg/L, robust model-informed regimens included 2000/500 mg q6h for augmented renal clearance and 2000/500 mg q12h during RRT. For CFR-based empirical therapy during RRT, robust model-informed regimens were 2000/500 mg q12h for carbapenem-resistant Klebsiella pneumoniae (CRKP) and 2000/500 mg q8h for carbapenem-resistant Pseudomonas aeruginosa (CRPA).

Conclusion

Our study describes the pharmacokinetics of CAZ-AVI in critically ill patients and provides model-informed dosing regimens stratified by renal function and RRT status for both empirical and targeted therapy. Larger, multicenter prospective studies are warranted to validate these findings.

Introduction

Infections caused by carbapenem-resistant organisms (CROs) have emerged as a major global health threat over the past two decades.1,2 CROs comprise mechanistically and clinically heterogeneous pathogens, including carbapenem-resistant Enterobacterales (CRE), such as carbapenem-resistant Escherichia coli (CREC) and carbapenem-resistant Klebsiella pneumoniae (CRKP), and non-fermenting Gram-negative bacilli, particularly carbapenem-resistant Pseudomonas aeruginosa (CRPA) and carbapenem-resistant Acinetobacter baumannii (CRAB).3 Ceftazidime-avibactam (CAZ-AVI), a β-lactam/β-lactamase inhibitor (BL/BLI) combination, has demonstrated clinical effectiveness against susceptible carbapenem-resistant gram-negative infections.4–6 Its spectrum depends on the underlying resistance mechanism: avibactam inhibits class A enzymes, including KPCs, class C β-lactamases, and certain class D enzymes, particularly OXA-48-like carbapenemases, but not class B metallo-β-lactamases such as NDM, VIM, or IMP.7 Despite favorable real-world effectiveness and tolerability, microbiological failure and the emergence of CAZ-AVI resistance remain concerns,8 and pneumonia and renal replacement therapy (RRT) have also been associated with CAZ-AVI treatment failure and resistance in critically ill patients.9 Recent clinical and molecular evidence suggests that CAZ-AVI resistance involves host-, treatment-, and pathogen-related factors, including the within-host replacement, amplification, mutation, or altered expression of bla_KPC variants.10,11 Avoiding inadequate exposure is therefore biologically plausible as one component of resistance mitigation; however, resistance emergence was not evaluated as an outcome in this cohort.

Attainment of appropriate pharmacokinetic/pharmacodynamic (PK/PD) exposure is closely associated with antibacterial efficacy.12 Like traditional β-lactam antibiotics, the percentage of the dosing interval during which the free concentration exceeds the minimum inhibitory concentration (%fT > MIC) has been described as the optimal PK/PD index associated with the efficacy of novel BL/BLI combinations. Preclinical studies support a conventional joint target requiring simultaneous attainment of CAZ 50% fT>MIC and AVI 50% fT>1 mg/L.13–15 Previous investigations reported adequate target attainment with the standard regimen of 2000/500 mg q8h administered as a 2-h infusion.16,17

However, critically ill patients exhibit substantial pathophysiological alterations that may markedly affect volume of distribution (V_d) and clearance (CL) of CAZ and AVI.16,18 Both agents are hydrophilic, have low molecular weights and low protein binding, and are predominantly eliminated unchanged in the urine; their pharmacokinetics are therefore particularly sensitive to changes in fluid distribution, native renal function, and extracorporeal removal.16 Capillary leakage and fluid resuscitation may expand V_d and reduce plasma concentrations, whereas augmented renal clearance (ARC), often associated with hyperdynamic circulation, may accelerate renal elimination and increase the risk of underexposure.19–21 Conversely, acute kidney injury or renal failure may reduce native renal clearance and cause accumulation.16,22 Continuous renal replacement therapy (CRRT) adds a potentially time-varying extracorporeal clearance pathway,16,22,23 whose magnitude may vary with modality, membrane, delivered effluent rate, dilution configuration, operating duration, interruptions, and residual renal function.16,22,24 These opposing factors place critically ill patients, particularly those receiving RRT, at risk of inadequate or excessive CAZ-AVI exposure.

Given this PK variability and the potential need for greater exposure in severe infections, expert guidance and clinical studies have proposed β-lactam exposure targets including 100% fT>MIC and, in selected clinical circumstances, 100% fT>4MIC.25,26 However, evidence supporting intensified exposure remains limited, and its safety implications remain uncertain.27,28 Against this background, we evaluated two joint PK/PD target pairs informed by published β-lactam exposure targets and the established AVI threshold concentration of 1 mg/L: CAZ 100% fT>MIC plus AVI 100% fT>1 mg/L, and CAZ 50% fT>4MIC plus AVI 50% fT>1 mg/L.13,29 The former emphasizes sustained coverage, whereas the latter emphasizes greater exposure intensity over half of the dosing interval. Both were efficacy-oriented simulation assumptions rather than validated safety thresholds and were evaluated in parallel, with regimen classification based on integrated attainment across the two targets.

Evidence on CAZ-AVI PK/PD during CRRT remains limited.23,30–34 A recent prospective PK/PD study of 21 critically ill adults receiving CRRT reported high CAZ-AVI exposure and modality-related differences in extracorporeal clearance between continuous venovenous hemofiltration and continuous venovenous hemodialysis.23 Nevertheless, prospective studies combining population pharmacokinetics (PPK) models from patients with and without RRT with Monte Carlo simulation (MCS) across renal-function and RRT strata remain scarce.

MCS uses PPK models and estimated variability to generate virtual patient populations, extending model-based evaluation beyond the observed sample without increasing the empirical sample size.35 By comparing candidate regimens and PK/PD targets in virtual clinical trials, MCS is particularly useful when large trials are difficult to conduct, such as in patients receiving CRRT. In antimicrobial analyses, probability of target attainment (PTA) is estimated at specified MICs, whereas cumulative fraction of response (CFR) for empirical therapy is calculated by weighting MIC-specific PTA across a pathogen-specific MIC distribution. Current local MIC data are preferable for institution-specific decisions, whereas multicenter or global surveillance data provide broader context but may not reflect local susceptibility patterns; their source and representativeness should therefore be considered when interpreting CFR estimates.36

Therefore, we conducted a prospective study to characterize the population pharmacokinetics of CAZ and AVI in critically ill patients with and without RRT. The resulting PPK models were subsequently used in MCS to identify candidate model-informed dosing regimens across renal-function and RRT strata.

Materials and Methods

Study Design and Patient Enrollment

This was a prospective, single-center cohort study conducted at the First Affiliated Hospital of the Medical College, Zhejiang University, a tertiary teaching hospital, from June 1, 2021 to January 1, 2024. Institutional ethics approval was obtained (No. 2022IIT468). The study was registered at ClinicalTrials.gov (NCT05413343) after the initiation of patient enrollment. Critically ill patients (≥18 years) receiving CAZ-AVI as a 2-h intravenous infusion for clinically diagnosed, culture-confirmed CRO infections susceptible to CAZ-AVI were enrolled. Eligibility criteria included an expected intensive care unit (ICU) stay ≥48 h and a hemoglobin concentration >70 g/L to minimize the risk associated with serial pharmacokinetic blood sampling; pregnant women were excluded. The study was conducted in accordance with the Declaration of Helsinki, and all participants provided written informed consent. Arterial blood (3 mL) was collected at 0, 1, 2, 4, 6, and 8 h after the first and steady-state infusions of CAZ-AVI. Anticoagulated blood was centrifuged (4°C, 4000 rpm, 10 min), and the plasma was stored at −80°C for drug analysis.

Patient Data Collection

Patient data were extracted from electronic medical records and included (1) demographic information, such as sex, age, body mass index (BMI), body weight; (2) disease severity scores, including Acute Physiology and Chronic Health Evaluation II (APACHE II) score and Sequential Organ Failure Assessment (SOFA) score; (3) laboratory parameters, such as plasma albumin, hematocrit, lactate and creatinine clearance (CrCL); (4) infection-related data, such as site of infection, pathogen type and clinical outcomes; (5) the CAZ-AVI treatment regimen; and (6) cumulative fluid balance, vasopressor use, and mechanical ventilation. Moreover, the following RRT parameters were recorded: machine/dialyzer type, RRT mode, pre-dilution, post-dilution, filter type, blood flow rate, dialysate flow rate, total replacement-fluid rate, mean net ultrafiltration rate, weight-normalized total ultrafiltration rate, treatment dose, filtration fraction and daily residual urine volume.

Drug Concentration Analysis

Plasma concentrations of CAZ and AVI were quantified using ultrahigh-performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry (UHPLC-Q-TOF). Reference standards were dissolved in ultrapure water and serially diluted to prepare working solutions. Matrix-matched calibration standards and quality-control (QC) samples were prepared by spiking the working solutions into blank plasma and were processed identically to the study samples.

For sample preparation, one volume of plasma was mixed with three volumes of methanol for protein precipitation. After centrifugation, the resulting supernatant was further diluted 10-fold with ultrapure water, corresponding to an overall 40-fold dilution relative to the original plasma sample. The calibration standards corresponded to original-plasma concentrations of 10–640 mg/L for CAZ and 2–160 mg/L for AVI, yielding final prepared-sample calibration ranges of 0.25–16 μg/mL and 0.05–4.0 μg/mL, respectively. The corresponding final-sample lower limit of quantification (LLOQ)/ upper limit of quantification (ULOQ) values were 0.25/16 μg/mL for CAZ and 0.05/4.0 μg/mL for AVI. Quantification was performed using matrix-matched external-standard calibration without an internal standard.

The analytical procedure was evaluated for linearity, selectivity, matrix effect, extraction recovery, accuracy, intra-day and inter-day precision, and carryover. QC samples were evaluated at final prepared-sample concentrations of 0.4, 2.0, and 8.0 μg/mL for CAZ and 0.1, 0.5, and 2.0 μg/mL for AVI. All study samples were quantified within the established calibration ranges; no additional dilution of above-range samples or extrapolation beyond the calibration curves was required. No below limit of quantification (BLQ) samples were included in the pharmacokinetic analysis.

Population Pharmacokinetics

Base Model Development

The PPK model for CAZ and AVI was developed using NONMEM (version 7.2) with parameter estimation via the first-order conditional estimation method with interaction (FOCE-I). Separate models were developed for the RRT and non-RRT populations to account for their different clearance determinants. First-dose and steady-state concentrations were jointly analyzed in the same PPK model. For each patient, the complete dosing history from treatment initiation through steady-state sampling was entered chronologically, including the dose amount, administration time, and infusion rate for every dosing event. One- and two-compartment structural models with linear elimination were evaluated. The model with the smallest Akaike information criterion (AIC) was selected as the optimal model. The base model was established by incorporating a random-effects model into this structural model. Inter-individual variability was characterized using an exponential model (Equation 1): P_ij=P_pop×e^ηi. (1) where P_ij and P_pop are the individual pharmacokinetic and typical values of the population pharmacokinetic parameters, respectively, and ηi is a random variable that is normally distributed with a mean of zero and variance of ω². Residual variability was evaluated using three different error models: an additive error model (Equation 2), a proportional error model (Equation 3), and a combined error model (Equation 4). Cobs=Cpred+ε. (2) Cobs=Cpred×(1+ε). (3) Cobs=Cpred×(1+ε)+ε’. (4) Cobs and Cpred represent the observed and predicted concentrations, respectively, whereas ε and ε’ are random error terms with means of zero and variances of σ² and σ’², respectively.

Covariate Model Development

Stepwise covariate selection, combining forward inclusion and backward elimination, was performed to identify significant covariates. Candidate covariates evaluated in both the non-RRT and RRT models included sex, age, BMI, body weight, SOFA score, APACHE II score, serum albumin, serum lactate, CrCL, cumulative fluid balance, mechanical ventilation, vasopressor use, and infection site. The RRT models additionally evaluated RRT mode, actual operating duration, weight-normalized total effluent rate, filtration fraction, mean net ultrafiltration rate, and daily residual urine output. During forward inclusion, a covariate significantly reduced the objective function value (OFV = −2log-likelihood; ΔOFV > 3.84, P<0.05) and was retained. This yielded a full model. Subsequent backward elimination removed covariates only if their exclusion increased the OFV by >10.83 (P<0.001). A covariate was ultimately retained in the event of the following: (1) The lowest OFV was obtained and goodness-of-fit plots improved. (2) A clinically interpretable effect could be ascertained. (3) Its associated pharmacokinetic parameter estimate had a 95% confidence interval excluding zero.

Model Validation

Model validation was achieved through goodness-of-fit (GOF) plots, including observed concentrations (Cobs) versus population (PRED) and individual predictions (IPRED), and conditional weighted residuals (CWRES) versus time or PRED. Parameter-estimation precision was assessed using relative standard errors (RSEs). Model stability was assessed using bootstrap analysis (1000 replicates), and parameter estimates as well as 95% CIs were compared. The predictive performance was further evaluated by numerical predictive check (NPC) with 1000 simulations.

Monte Carlo Simulation (MCS)

Monte Carlo simulations (1000 replicates)37,38 using the final PPK models evaluated the steady-state performance of CAZ-AVI dosing regimens in critically ill patients with and without RRT. All simulated regimens were administered as 2-h intravenous infusions. For non-RRT patients, six renal function categories based on CrCL were simulated: 131–160 mL/min (ARC39), 81–130 mL/min, 51–80 mL/min, 31–50 mL/min, 16–30 mL/min, and 6–15 mL/min. Following previously published CAZ-AVI and BL/BLI MCS studies, joint probability of target attainment (PTA) was calculated as the product of the component-specific PTAs for CAZ and AVI.38–40 This product-based calculation assumes independence of component-specific target attainment, and cross-component correlation between CAZ and AVI was not explicitly modeled in the present simulations. Pathogen-specific CFR was calculated by weighting joint PTA across the corresponding 2019–2021 Blood Bacterial Resistant Investigation Collaborative System (BRICS) MIC distribution for CRKP, CRPA, CRE, CREC and CRO. The free fractions of CAZ and AVI used in the simulations were 85% and 92%, respectively.41

Two PK/PD targets were set: (1) CAZ 100% fT > MIC and AVI 100% fT > 1 mg/L; and (2) CAZ 50% fT > 4MIC and AVI 50% fT > 1 mg/L.29 Joint PTA informed MIC-directed definitive therapy, whereas pathogen-specific CFR informed empirical therapy when the MIC was unavailable. For pathogens with MICs ≤8 mg/L, definitive-therapy regimens were evaluated using joint PTA at an MIC of 8 mg/L. Within each PK/PD target, the target-attainment threshold was a joint PTA ≥90% at an MIC of 8 mg/L for definitive therapy or a pathogen-specific CFR ≥90% for empirical therapy.38 Final regimen classification was based on integrated attainment across both PK/PD targets.

Regimens were first evaluated separately under each PK/PD target. When multiple regimens achieved ≥90% within a target, they were ranked by the lowest total daily dose, followed by the lowest dosing frequency and then the higher PTA or CFR. A regimen achieving ≥90% under both targets was classified as a robust model-informed regimen. If no robust regimen was identified, regimens attaining ≥84% under both targets were evaluated exploratorily: ≥90% under one target and 84% to <90% under the other defined a target-dependent simulation-supported alternative, whereas 84% to <90% under both defined a near-threshold exploratory regimen. Multiple exploratory candidates were ranked by the highest minimum attainment across the two targets, followed by the lowest total daily dose and dosing frequency. Regimens attaining <84% under either target were not selected. The predefined 84% cutoff represented an exploratory margin and was not equivalent to the 90% target-attainment threshold.

Statistical Analysis

All statistical analyses were performed using SPSS version 27.0. Categorical variables are presented as counts and percentages. Continuous variables in the baseline patient characteristics are expressed as medians and interquartile ranges (IQRs). PK parameters are expressed as the mean ± standard deviation (SD).

Results

Patient Characteristics

The clinical characteristics of the 33 enrolled patients are summarized in Table 1. The cohort predominantly consisted of middle-aged and elderly males (78.78%), with a median age of 70 years (IQR 61–75.5). The median APACHE II and SOFA scores were 15 (IQR 12–22) and 7 (IQR 4–10), respectively. Eighteen patients received RRT with prescribed modalities of continuous venovenous hemofiltration (CVVH) or continuous venovenous hemodiafiltration (CVVHDF). The actual operating duration on the days of first-dose and steady-state pharmacokinetic sampling varied across patients, with some sessions lasting less than 24 hours. These shorter durations reflected the delivered operating time on the respective sampling days and were not classified as separate intermittent modalities based solely on duration. Patient-specific RRT settings and actual operating durations are provided in Supplementary Tables 1 and 2. Fifteen patients without RRT were stratified by CrCL: 2 had ARC, 9 had normal renal function, and 4 presented with moderate renal impairment. The causative pathogens were CRKP (63.63%) and CRPA (36.36%), and the infection sites included lung (69.70%), bloodstream (9.09%), urinary tract (6.06%) and multisite infections (15.15%). The majority of patients (30, 90.90%) received 2.5 g of CAZ-AVI via a 2-h infusion every 8 h. In the overall population, the rates of microbiological clearance and clinical improvement were 51.51% and 81.81%, respectively. The 14-day infection-related mortality rate was 15.15%, and the 30-day all-cause mortality rate was 30.30%.

Table 1.

Demographic and Clinical Characteristics of Patients

Characteristics	Total (n=33)	RRT (n=18)	Non-RRT (n=15)
Age (years), median (IQR)	70 (61.0, 75.50)	72 (59.50, 75.25)	67 (62.0, 76.0)
Male, n (%)	26 (78.78)	13 (72.22)	13 (86.67)
BMI, median (IQR)	21.3 (19.25, 26.55)	22.6 (20.05, 28.60)	20.5 (18.40, 26.0)
Weight (Kg), median (IQR)	60 (54.50, 70.50)	64 (57.25, 77.75)	59 (55.0, 63.0)
SOFA, median (IQR)	7 (4.00, 10.00)	8.5 (6.75, 12.50)	4 (2.0, 5.0)
APACHE II, median (IQR)	15 (12.0, 22.0)	17.5 (13.50, 22.50)	15 (10.0, 19.0)
Albumin (g/L), median (IQR)	33.90 (30.30, 37.70)	34.10 (30.7, 37.5)	33.70 (28.70, 38.40)
CrCL (mL/min), median (IQR)	80.41 (47.16, 114.77)	68.62 (48.46, 103.52)	100.48 (43.89, 122.32)
Lactate (mmol/L), median (IQR)	1.00 (0.80, 1.80)	1.00 (0.83, 1.77)	1.10 (0.65, 1.70)
Cumulative fluid balance (mL), median (IQR)	947.65 (67.35, 2129.93)	1105.00 (−496.70,1994.80)	853.40 (406.50,2089.50)
Mechanical ventilation, n (%)	26 (78.79%)	16 (88.89%)	10 (66.67%)
Vasoactive drug use, n (%)	14 (42.42%)	11 (61.11%)	3 (20.00%)
Pathogens
CRKP, n (%)	21 (63.63)	9 (50.0)	12 (80.0)
CRPA, n (%)	12 (36.36)	9 (50.0)	3 (20.0)
Infected sites
Lungs, n (%)	23 (69.70)	12 (66.67)	11 (73.33)
Bloodstream, n (%)	3 (9.09)	2 (11.11)	1 (6.67)
Urinary tract, n (%)	2 (6.06)	2 (11.11)	0 (0)
Lungs + Bloodstream, n (%)	4 (12.12)	2 (11.11)	2 (13.33)
Lungs + Urinary tract, n (%)	1 (3.03)	0 (0)	1 (6.67)
Treatment regimen
1.25g q8h, 2-h infusion	2 (6.06)	0 (0)	2 (13.33)
2.5g q12h, 2-h infusion	1 (3.03)	1 (5.56)	0 (0)
2.5g q8h, 2-h infusion	30 (90.90)	17 (94.44)	13 (86.67)
Clinical outcomes
Microbiological clearance, n (%)	17 (51.51)	10 (55.56)	7 (46.67)
Treatment outcome (Improved, n (%)	27 (81.81)	13 (72.22)	14 (93.33)
14-day infection-related mortality, n (%)	5 (15.15)	4 (22.22)	1 (6.67)
30-day all-cause mortality, n (%)	10 (30.30)	8 (44.44)	2 (13.33)
Total ICU length of stay, median (IQR)	33 (13.50, 71.50)	32 (20.75, 73.75)	33 (10.00, 70.00)
CrCL-Based Grouping in patients not receiving RRT
ARC (CrCL>130 mL/min), (%)			2 (13.33)
Normal renal function (81–130 mL/min), n (%)			9 (60.0)
Mild renal impairment (51–80 mL/min), n (%)			0 (0)
Moderate renal impairment (31–50 mL/min), n (%)			4 (26.67)
Severe renal impairment (16–30 mL/min), n (%)			0 (0)
Profound severe renal impairment (5–15 mL/min), n (%)			0 (0)

Note: Creatinine clearance (CrCL) was calculated using the Cockcroft–Gault equation.

Abbreviations: BMI, body mass index; SOFA, Sequential Organ Failure Assessment; APACHE II, Acute Physiology and Chronic Health Evaluation II; CrCL, creatinine clearance; ARC, augmented renal clearance; RRT, renal replacement Therapy; CRKP, carbapenem-resistant Klebsiella Pneumoniae; CRPA, carbapenem-resistant Pseudomonas Aeruginosa; IQR, interquartile range; q8h, every 8 hours; q12h, every 12 hours.

PPK Analysis of CAZ and AVI in RRT and Non-RRT Patients

A total of 211 blood samples were included in the population pharmacokinetic analysis (120 RRT and 91 non-RRT) (Detailed participant and sample flow diagram was in Supplementary Figure 1). The pharmacokinetics of CAZ and AVI in patients with and without RRT were adequately described by a one-compartment model with first-order elimination. Compared with the corresponding two-compartment models, the one-compartment models consistently yielded lower AIC values for both CAZ and AVI in the RRT and non-RRT populations (ΔAIC, 1.797–3.862), supporting selection of the simpler structural model for the present dataset. Using stepwise covariate selection, CrCL was retained as a covariate on CL in the final non-RRT models (ΔOFV = −21.338 for CAZ and −25.960 for AVI), the final covariate relationships for the non-RRT models were CL_CAZ = CL_POP×(CrCL/100.45)^0.792 and CL_AVI = CL_POP×(CrCL/100.45)^0.906, respectively. In contrast, none of the evaluated candidate covariates met the predefined criteria for retention in the final RRT models. In the final model, the typical estimates of CL and V_d for CAZ in the RRT group were 2.03 L/h and 17 L, respectively, while those for AVI were 1.77 L/h and 14 L, respectively; the corresponding estimates in the non-RRT group were 3.36 L/h and 20.5 L for CAZ and 4.27 L/h and 16 L for AVI, respectively (Table 2).

Table 2.

Population PK Parameter Estimates of CAZ and AVI for the Final PK Model and Bootstrap Results

RRT or not	Parameter	CAZ			AVI
		Bootstrap			Bootstrap
		Value (%RSE)	Median	95% CI	Value (%RSE)	Median	95% CI
RRT	CLPOP(L/h)	2.03(9)	2.02	1.68–2.38	1.77(11)	1.75	1.39–2.16
	VPOP(L)	17(9)	16.9	14.01–19.51	14(10)	13.97	11.29–16.67
	IIV CL	0.134(33)	0.123	0.051–0.228	0.196(30)	0.184	0.078–0.299
	IIV V	0.0645(67)	0.062	0.008–0.195	0.0814(60)	0.075	0.007–0.245
	RV (proportional)	0.0799(23)	0.078	0.045–0.116	0.0857(22)	0.086	0.048–0.123
non-RRT	CLPOP(L/h)	3.36(7)	3.33	2.82–3.99	4.27(7)	4.31	3.71–5.07
	VPOP(L)	20.5(20)	20.71	13.81–30.26	16(21)	15.62	10.89–24.64
	α, the ratio of CL for CrCL (EXP)	0.792(17)	0.79	0.35–1.05	0.906(12)	0.9	0.33–1.16
	IIV CL	0.0415(40)	0.03	0.00–0.07	0.0223(48)	0.02	0.00–0.06
	IIV V	0.368(35)	0.36	0.07–0.72	0.293(54)	0.24	0.07–0.57
	RV (proportional)	0.0559(21)	0.05	0.03–0.08	0.0909(30)	0.09	0.05–0.16

Note: The final-model OFV values were 979.211 for CAZ in RRT patients, 683.751 for CAZ in non-RRT patients, 696.703 for AVI in RRT patients, and 428.615 for AVI in non-RRT patients.

Abbreviations: CAZ, ceftazidime; AVI, avibactam; PK, pharmacokinetics; CL_POP, population clearance; VPOP, population volume of distribution; IIV, interindividual variability; RV (proportional), random variability proportional; RSE, relative standard error; CI, confidence interval; CrCL, creatinine clearance; RRT, renal replacement therapy; OFV, objective function value.

The fixed-effect parameter estimates showed RSEs ranging from 7% to 21%, compared with 30% to 67% for the interindividual variability estimates (Table 2). The GOF plots for CAZ and AVI in patients with and without RRT are shown in Figures 1 and 2. The observed concentrations were generally captured by the population and individual predictions, although the LOESS trends indicated some systematic deviation. Most CWRES were within ±2. The NPC results are presented in Supplementary Tables 3–6. In the bootstrap analysis, the final parameter estimates were close to the bootstrap medians and fell within the corresponding 95% confidence intervals, indicating generally consistent parameter recovery across successful bootstrap replicates.

Figure 1.

Goodness-of-fit plots of final population pharmacokinetic model of CAZ and AVI in patients receiving RRT. (A–D) Goodness-of-fit plots of CAZ, (E–H) Goodness-of-fit plots of AVI. (A) and (E) observed values vs individual predicted values. (B) and (F) observed values vs population predicted values. (C) and (G) weighted residuals vs population predicted values. (D) and (H) weighted residuals vs time after initial dose and at steady state. Black solid lines are reference lines, and red solid lines are locally weighted regression (LOESS) trend lines.

Figure 2.

Goodness-of-fit plots of final population pharmacokinetic model of CAZ and AVI in patients not receiving RRT. (A–D) Goodness-of-fit plots of CAZ, (E–H) Goodness-of-fit plots of AVI. (A) and (E) observed values vs individual predicted values. (B) and (F) observed values vs population predicted values. (C) and (G) weighted residuals vs population predicted values. (D) and (H) weighted residuals vs time after initial dose and at steady state. Black solid lines are reference lines, and red solid lines are locally weighted regression (LOESS) trend lines.

PK Parameters

Bioanalytical Method Performance

Calibration curves showed good linearity over the established concentration ranges, with coefficients of determination (r²) >0.99 for both analytes. No significant interfering peaks were observed in blank plasma. For CAZ, matrix effects ranged from 93.16% to 101.22%, extraction recoveries from 97.59% to 99.73%, accuracy from 92.40% to 99.57%, intra-day precision from 3.13% to 9.17%, and inter-day precision from 6.18% to 8.40%. For AVI, the corresponding values were 98.74%–105.78%, 97.06%–102.91%, 101.61%–103.79%, 3.22%–5.22%, and 3.75%–5.47%, respectively. Carryover following analysis of the highest calibration standard was <20% of the analyte response at the LLOQ. All study-sample concentrations were within the established calibration ranges after routine sample preparation; no concentrations were obtained by extrapolation beyond the calibration curves, no additional dilution of above-range samples was required, and no BLQ samples were included in the pharmacokinetic analysis.

Pharmacokinetic Parameters

All 33 enrolled patients (18 RRT and 15 non-RRT) were included in the PPK analysis. Table 3 presents a descriptive comparison restricted to the 30 patients who received 2000/500 mg q8h as a 2-h infusion (17 RRT and 13 non-RRT); the remaining three patients received different regimens and were excluded only from this descriptive comparison. Compared with the non-RRT group, the RRT group had numerically higher C_max values for both CAZ (194.80 ± 82.01 vs 131.34 ± 52.94 mg/L) and AVI (57.76 ± 25.42 vs 35.41 ± 15.07 mg/L). The elimination half-lives were also numerically longer in the RRT group for CAZ (5.86 ± 2.13 vs 4.81 ± 2.44 h) and AVI (5.76 ± 2.40 vs 3.06 ± 1.76 h). In addition, CL and V_d were numerically lower in the RRT group than in the non-RRT group.

Table 3.

Descriptive Pharmacokinetic Parameters of CAZ-AVI in Patients Receiving 2000/500 mg q8h as a 2-h Infusion (n=30)

PK Parameters	CAZ		AVI
	RRT (n=17)	no-RRT (n=13)	RRT (n=17)	no-RRT (n=13)
Cmax (mg/L)	194.80 ± 82.01	131.34 ± 52.94	57.76 ± 25.42	35.41± 15.07
t1/2 (h)	5.86 ± 2.13	4.81 ± 2.44	5.76 ± 2.40	3.06 ± 1.76
AUC0-24 (mg.h/L)	3036.33 ± 1126.24	1982.33 ±875.89	914.11 ± 419.31	411.12 ± 235.89
CL (L/h)	2.21 ± 0.70	3.42 ± 1.05	1.95 ± 0.79	4.34 ±1.44
Vd (L)	16.96 ± 2.83	23.77 ±14.70	14.02 ± 2.69	17.96 ±10.34

Notes: Between-group comparisons of PK parameters are presented for descriptive purposes only; no formal statistical testing was performed.

Abbreviations: CAZ, ceftazidime; AVI, avibactam; PK, pharmacokinetics; RRT, renal replacement therapy; C_max, peak concentration; t_1/2, half-life; AUC_0-24, area under the plasma concentration-time curve at 0–24h; CL, clearance; V_d, volume of distribution.

MIC Distribution of Pathogens

This study incorporated data on the MIC distribution of CAZ-AVI for CRKP, CRPA, CRE, CREC and CRO strains from BRICS between 2019 and 2021. The detailed MIC distributions of the bacterial strains are presented in Supplementary Table 7.

PTA Analysis

In patients receiving RRT, when CAZ 100% fT > MIC and AVI 100% fT > 1 mg/L were used as the PK/PD targets (Figure 3A), all the simulated dosing regimens except 1000/250 mg q12h achieved ≥90% PTA at MICs ≤8 mg/L. Under the targets of CAZ 50% fT > 4MIC and AVI 50% fT > 1 mg/L (Figure 3B), the regimens 2000/500 mg q8h, 2000/500 mg q12h and 1000/250 mg q6h maintained a ≥90% PTA at an MIC of 8 mg/L, but the 1000/250 mg q8h and q12h regimens did not.

Figure 3.

Joint probability of target attainment (PTA) of different CAZ-AVI dosing regimens in patients receiving RRT. (A) The PK/PD target was CAZ 100% fT > MIC and AVI 100% fT > 1 mg/L. (B) The PK/PD target was CAZ 50% fT > 4MIC and AVI 50% fT > 1 mg/L.

In patients not receiving RRT, when CAZ 100% fT > MIC and AVI 100% fT > 1 mg/L were used as the PK/PD targets (Supplementary Figure 2), for pathogens with an MIC ≤ 8 mg/L, the standard dosing regimen (2000/500 mg q8h) achieved a PTA less than 90% in both the ARC and CrCL 81–130 mL/min groups, with values of 77.1% and 88.8%, respectively. The 2000/500 mg q6h regimen met the predefined target-attainment threshold in the ARC group, whereas in the CrCL 81–130 mL/min group, the 1000/250 mg q6h regimen achieved a PTA of 93.99%. When the PK/PD target of CAZ was 50% fT > 4MIC and AVI 50% fT > 1 mg/L (Supplementary Figure 3), the 2000/500 mg q6h regimen achieved a PTA of 94.15% in ARC patients, and the 2000/500 mg q8h regimen achieved a PTA ≥90% in the CrCL 81–130 mL/min group. Furthermore, compared with the 100% fT > MIC target, fewer dosing regimens achieved PTA ≥90% under the 50% fT > 4MIC target in the other renal-function strata.

CFR Analysis

As shown in Figures 4 and 5, CFR values were generally lower under the PK/PD targets of CAZ 50% fT > 4MIC and AVI 50% fT > 1 mg/L than under those of CAZ 100% fT > MIC and AVI 100% fT > 1 mg/L. In patients receiving RRT, under the PK/PD target of 100% fT > MIC and AVI 100% fT > 1 mg/L, CFRs exceeding 90% were observed with the 2000/500 mg q8h, 2000/500 mg q12h, 1000/250 mg q6h, and 1000/250 mg q8h regimens in cases of CRKP and CRPA infection. For CRE or CRO infection, the 2000/500 mg q8h and 1000/250 mg q6h regimens achieved CFRs greater than 90%, whereas for CREC infection, only the 2000/500 mg q8h regimen exceeded this threshold. Under the other PK/PD criteria (CAZ 50% fT > 4MIC and AVI 50% fT > 1 mg/L), only specific regimens achieved CFRs ≥90%: 2000/500 mg q8h for both CRKP and CRPA, and 1000/250 mg q6h and 2000/500 mg q12h for CRKP alone.

Figure 4.

Monte Carlo simulation-derived CFRs for CAZ-AVI against different carbapenem-resistant Gram-negative bacterial populations under the PK/PD target of CAZ 100% fT > MIC and AVI 100% fT > 1 mg/L: (A) carbapenem-resistant Klebsiella pneumoniae (CRKP); (B) carbapenem-resistant Pseudomonas aeruginosa (CRPA); (C) carbapenem-resistant Enterobacterales (CRE); (D) carbapenem-resistant Escherichia coli (CREC); and (E) carbapenem-resistant organisms (CRO).

Figure 5.

Monte Carlo simulation-derived CFRs for CAZ-AVI against different carbapenem-resistant Gram-negative bacterial populations under the PK/PD target of CAZ 50% fT > 4MIC and AVI 50% fT > 1 mg/L: (A) carbapenem-resistant Klebsiella pneumoniae (CRKP); (B) carbapenem-resistant Pseudomonas aeruginosa (CRPA); (C) carbapenem-resistant Enterobacterales (CRE); (D) carbapenem-resistant Escherichia coli (CREC); and (E) carbapenem-resistant organisms (CRO).

For the CFR analysis in non-RRT patients, under the target of CAZ 100% fT > MIC and AVI 100% fT > 1 mg/L, the following results were observed for CRKP or CRPA infection. In patients with ARC, only the 2000/500 mg q6h regimen achieved a CFR ≥90%, in the CrCL 81–130 mL/min group, the 2000/500 mg q8h regimen did not achieve CFR ≥90% for either CRKP or CRPA (88.01% and 87.31%, respectively), and only the 1000/250 mg q6h regimen resulted in a CFR ≥90% for CRKP infection. In the CrCL 51–80 mL/min group, the 2000/500 mg q8h regimen achieved CFRs > 90% for both CRKP (94.86%) and CRPA (95.48%) infections. Under the PK/PD target of CAZ 50% fT > 4MIC and AVI 50% fT > 1 mg/L, no regimen achieved a CFR ≥90% for CRKP or CRPA infection in patients with ARC or CrCL 81–130 mL/min. Similarly, for CRE, CREC and CRO infections, no regimen achieved a CFR > 90% across evaluated non-RRT renal-function strata.

Integrated PTA- and CFR-Based Regimen Evaluation

Across both PK/PD targets, PTA-based evaluation at MIC = 8 mg/L identified the following robust model-informed regimens: 2000/500 mg q6h for patients with ARC, 1000/250 mg q6h for patients with CrCL 51–80 mL/min, and 2000/500 mg q12h for patients receiving RRT. For patients with CrCL 81–130 mL/min, 2000/500 mg q8h achieved PTAs of 88.8% and 93.0% under 100% fT > MIC and 50% fT > 4MIC, respectively, and was classified as a target-dependent simulation-supported alternative rather than a robust regimen. Reduced-dose regimens were identified for lower renal-function strata, as summarized in Supplementary Table 8.

For empirical treatment during RRT, robust CFR-based regimens were 2000/500 mgmg q12h for CRKP and 2000/500 mg q8h for CRPA. In patients with ARC, 2000/500 mg q6h was a target-dependent simulation-supported alternative for both CRKP and CRPA. For patients with CrCL 81–130 mL/min, 2000/500 mg q8h was a near-threshold exploratory regimen for CRKP (88.01%/89.77%) and CRPA (87.31%/84.49%). For patients with CrCL 51–80 mL/min, 1000/250 mg q6h was robust for CRKP, whereas 2000/500 mg q8h was a target-dependent simulation-supported alternative for CRPA. Results for the remaining renal-function strata and organisms are presented in Supplementary Table 9.

Discussion

Diminished pathogen susceptibility and marked pathophysiological variability in critically ill patients complicate antibiotic dose optimization in the ICU.42 In this prospective study of critically ill patients with and without RRT, we characterized the population pharmacokinetics of CAZ and AVI. Using the final PPK models, we identified exploratory, model-informed candidate CAZ-AVI dosing regimens according to renal function and RRT status, providing a population-level framework for dose optimization in this clinically heterogeneous population.

Compared with those of previous studies, our data showed numerical differences in CAZ-AVI PK parameters between the RRT and non-RRT groups. Elimination half-lives were numerically longer, whereas CL and V_d were numerically lower, in the RRT group. CL estimates in both groups were also numerically lower than values reported in healthy volunteers. Wenzler et al reported that in patients receiving CVVH who were given a 2-h infusion of CAZ-AVI at a dosage of 1.25 g every 8 h, the total CL was 2.87 and 2.95 L/h, respectively, which was lower than that in healthy volunteers, with higher V_d values (27.23 and 30.81 L, respectively).8 Another pharmacokinetic study of CAZ-AVI in critically ill patients revealed that the conventional dosing regimen (2.5 g every 8 h infused continuously) was suitable for patients receiving CVVHDF. In this case, the CL and V_d of CAZ and AVI were lower than those observed in healthy volunteers, accompanied by a longer half-life.32 The PK parameters in the RRT patients in our study are similar to those in a recent case report by Zhang et al, where the CL and V_d of CAZ and AVI were 2.03 L/h, 1.01 L/h, and 15.25 L, 14.82 L,31 respectively, both in Chinese populations, indicating consistency. Kline et al reported CL values for CAZ and AVI in CRRT patients of 4.0 and 4.2 L/h, respectively, which were numerically higher than our estimates, with V_d values of 43.2 and 56.1 L, respectively, which were higher than our data (17 and 14 L).43 Stein et al studied 10 critically ill patients who were not receiving RRT and reported that although the CLs of CAZ and AVI in these patients were similar to those in healthy volunteers (6.1±3.8 L/h and 11.1±6.8 L/h, respectively), higher V_d values were reported (35±10.5 L and 50.81±14.3 L, respectively),44 differing from our results. The reason for the difference in V_d may be the different study populations, but we could not obtain more V_d-related information, such as patient albumin levels, from Stein et al’s study.

Previous studies have indicated that CRRT membranes, CRRT modes and treatment doses may influence the PK/PD of antimicrobial agents.45 The drug removal ability of CRRT can also be affected by the properties of the drug itself. Molecular weight, protein binding, drug elimination pathways and V_d are considered the most important factors affecting PK parameters during CRRT.45 In the present study, RRT mode, actual operating duration, weight-normalized total effluent rate, filtration fraction, mean net ultrafiltration rate, and daily residual urine output were evaluated as potential covariates, but none met the predefined criteria for retention in the final RRT models. Failure to retain an RRT-related covariate in this cohort of 18 clinically heterogeneous patients should be interpreted in the context of limited statistical power and limited identifiability of the independent effects of individual RRT factors, rather than as evidence that RRT modality or dose has no PK effect. Delivered RRT parameters, including operating duration, weight-normalized total effluent rate, mean net ultrafiltration rate, and residual urine output, varied across patients, whereas the cohort provided limited contrast for some other RRT characteristics because the same machine and filter were used and CVVHDF predominated (14/18 patients). Together with the small sample size and potential interrelationships among RRT parameters, these features limited the ability of the models to distinguish their independent contributions. Based on PK data from previous clinical trials, covariate effects were broadly consistent throughout the iterations, and the main factors explaining the variability in exposure to both CAZ and AVI were patient population (patients vs healthy subjects) and CrCL.41,46,47 A study evaluating CAZ-AVI population pharmacokinetics and pharmacodynamic target attainment across adult indications and patient subgroups revealed that CrCL was a key covariate determining clearance variability.41 In our study, CrCL was incorporated as a covariate in the covariate model‑building process in the non‑CRRT group, which is in line with the findings of previous studies.16,41,48,49

In this study, separate PPK models were developed for CAZ and AVI in 18 RRT and 15 non-RRT patients using first-dose and steady-state concentrations. Some systematic LOESS trends were observed in the GOF plots, and occasion-specific variability was not separately estimated despite the joint modeling of both sampling occasions. In addition, joint PTA was calculated using the product-based approach without explicitly modeling cross-component correlation. Nevertheless, most CWRES remained within ±2, the final parameter estimates were generally consistent with the bootstrap results, and the NPC provided further support for the predictive performance of the models. Overall, these findings support the use of the current models for exploratory model-informed simulations. Future studies with larger independent cohorts and repeated PK sampling may further characterize occasion-specific variability, externally validate and refine the PPK models, and evaluate joint target attainment while accounting for potential correlations between CAZ and AVI exposures.

ARC and severe renal impairment represent opposite extremes of renal drug clearance in critically ill patients.50 However, limited data are available for CAZ-AVI dosing during CRRT. According to our results, at MICs ≤8 mg/L, the simulations identified 2000/500 mg q12h as a robust model-informed candidate regimen during RRT. For empirical treatment during RRT, the CFR analysis supported exploratory model-informed regimens of 2000/500 mg q12h for CRKP and 2000/500 mg q8h for CRPA. A patient with multidrug-resistant P. aeruginosa pneumonia receiving CVVHDF was treated with 2.5 g of CAZ-AVI via 2-h infusion every 8 h. This regimen achieved PK/PD targets of 100% fT>4MIC for CAZ and 100% fT>1 μg/mL for AVI.32 Wenzler et al30 reported a critically ill patient undergoing CVVH who received CAZ-AVI at 1.25 g every 8 h for multidrug-resistant P. aeruginosa bacteremia; the regimen also reached the target (CAZ: 100% fT>MIC up to 6 mg/L; AVI: 100% fT>1 mg/L). Another study demonstrated that in patients receiving CRRT, the 2.5 g every 8 h dosing regimen achieved the 100% ƒT > MIC target for CAZ in 90% of patients and the 100% ƒT > 4MIC target in 55% of patients.43 Furthermore, a recent PK/PD study of CAZ-AVI in patients with severe intra-abdominal infections treated with CVVH demonstrated that both the 1,000/250 mg every 8 h and the 2,000/500 mg every 8 h regimens achieved a PTA >90% for strains with an MIC ≤8 mg/L.51 Another study from China revealed that the standard dosing regimen of 2.5 g of CAZ-AVI every 8 h may increase the risk of excessive drug exposure and that CL for CAZ and AVI were higher during CVVH than during CVVHD.23 Optimizing population-based and individualized dosing strategies for CAZ-AVI during CRRT is a clinical challenge. For ARC patients (CrCL: 131–160 mL/min), the proposed regimen (2,000/500 mg q6h) aligns with that in previous studies suggesting that a strategy of increased dosing frequency combined with prolonged infusion may represent the optimal approach for achieving PK/PD targets for β-lactam antibiotics in this population.52 However, because all regimens in the present simulations were modeled as 2-h infusions, whether extending the infusion to 3–4 h could preserve joint target attainment with a lower total daily dose remains uncertain. Overall, the regimens proposed in our study should be regarded as exploratory, population-level, model-informed candidates requiring further validation. In clinical practice, dosing regimens should be individualized according to renal function, CRRT settings, pathogen MICs, and drug exposure. Therapeutic drug monitoring (TDM) is recommended whenever available.

For empirical therapy, pathogen-specific model-informed candidate regimens were identified for CRKP and CRPA, whereas no robust regimen was identified for CREC, CRE, or CRO under both PK/PD targets. The observed differences in CFR outcomes may be attributed to variations in carbapenem resistance mechanisms. Indeed, resistance mechanisms differ among CRKP, CRPA and CREC, with CREC strains predominantly producing metallo-enzymes. CAZ-AVI is effective against most strains producing KPC (mainly KPC-2) or OXA-48 but is not active against strains carrying NDM, VIM, or IMP genes or enzymes.53,54 This study did not differentiate bacterial enzymatic types, and further investigation is warranted to inform clinical treatment decisions.

The following limitations should be considered when interpreting our findings. First, this was a prospective, single-center study with a small sample size. Patient distribution across renal-function strata was uneven, and some simulated strata included no observed patients. Second, the RRT subgroup was small and included patients who received RRT for less than 24 h. Although relevant RRT-related variables were evaluated during covariate modeling, none were retained in the final model. Moreover, some patients receiving RRT had residual urine output, but residual renal clearance was not quantitatively assessed and urinary drug concentrations were not measured. Third, although the analytical assay demonstrated acceptable performance with respect to the evaluated parameters, including accuracy, precision, recovery, and matrix effects, several aspects of comprehensive bioanalytical validation were not assessed, including additional stability and dilution-integrity studies, and quantification was performed without an internal standard. Nevertheless, all study samples were quantified within the established calibration ranges without extrapolation or additional above-ULOQ dilution. Fourth, the MIC distributions used to calculate CFR were obtained from the BRICS dataset; therefore, the findings should be interpreted in the context of current local pathogen epidemiology and antimicrobial susceptibility patterns. Finally, drug exposure was not prospectively linked to clinical efficacy, safety, or resistance emergence; upper exposure thresholds and exposure-toxicity relationships were not evaluated; prolonged or continuous infusion strategies were not simulated; and the models were not externally validated. The proposed regimens should therefore be regarded as exploratory, population-level, model-informed candidate regimens that require confirmation, preferably with TDM where feasible and through larger prospective multicenter studies, before wider clinical adoption.

Conclusions

In this study, the pharmacokinetics of CAZ and AVI were characterized in critically ill patients. Both agents showed prolonged elimination half-lives and numerically lower CL values than those reported in healthy volunteers, while CL and V_d were numerically lower in patients receiving RRT than in those not receiving RRT. MCS based on the final PPK models evaluated candidate CAZ-AVI regimens according to renal function and RRT status. Integrated evaluation of both PK/PD targets supported 2000/500 mg q6h for ARC and 2000/500 mg q12h during RRT as robust model-informed regimens for definitive therapy at MICs ≤8 mg/L. In patients with CrCL 81–130 mL/min, 2000/500 mg q8h was a target-dependent simulation-supported alternative. For empirical therapy during RRT, robust model-informed regimens were 2000/500 mg q12h for suspected CRKP and 2000/500 mg q8h for suspected CRPA. The proposed regimens represent exploratory, population-level, model-informed strategies warranting confirmation in larger prospective multicenter studies; where feasible, TDM is recommended to support individualized dosing, particularly in patients with ARC or receiving RRT.

Funding Statement

The study was supported by the National Key R&D Program of China 2024YFC2311304.

Data Sharing Statement

The datasets supporting the findings of this study are available from the designated corresponding author, Yong-Hong Xiao, upon reasonable request.

Consent for Publication

All listed authors consent to the submission, and all data are used with the consent of the person generating the data.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare that they have no competing interests.