Study design and settings

A multicentre, prospective pharmacokinetic study was conducted in three ICUs in Malaysia–(i) Sultan Ahmad Shah Medical Centre (SASMEC), (ii) Hospital Universiti Sains Malaysia (USM), and (iii) University of Malaya Medical Centre (UMMC) between March 2017 and March 2018. Eligible participants were adult (≥ 18 years old) ICU patients receiving meropenem who met the Sepsis-3 criteria³², had confirmed or suspected infections, and had a central venous or arterial line for blood collection. Patients suspected or known to be hypersensitive to beta-lactam antibiotics, pregnant patients, lactating mothers, or those with preexisting renal impairment defined as either receiving extra-corporeal renal support or having an estimated creatinine clearance of less than 30 mL/min or plasma creatinine concentration above 200 μmol/L were excluded from the study. The study was conducted in accordance with the Declaration of Helsinki, the International Council for Harmonization Good Clinical Practice (ICH-GCP) guidelines, and local regulatory requirements. Informed consent was obtained from all participants or a legally authorized representative before they participated in the study. Individual institutional ethical approvals were obtained according to local protocols (UMMC Medical Research Ethics Committee Reg. No 2016725-4020; USM Human Research Ethics Committee Reg. No USM/JEPeM/16110528; IIUM Research Ethics Committee Reg. No IREC 2017-062). The study was registered and also approved by the Medical Research Ethics Committee Malaysia (NMRR-16237231675).

Meropenem dosing and sample collection

Meropenem dosing regimens were determined by the treating physician based on standard prescribing practices and the clinical assessment of each patient. Meropenem was typically administered as an intermittent infusion (II) over 30–60 min, extended infusion (EI) over 3–4 h, or continuous infusion (CI) over the entire dosing interval (8 h). Pharmacokinetic sampling was conducted on day 1 (Occasion 1) and day 3 (Occasion 2) of antibiotic therapy. Each patient had up to seven blood samplings on each occasion from a central line into heparinized tubes. These sampling points were distributed evenly across the dosing interval. For patients receiving meropenem every 8 h, the sampling points were immediately before the antibiotic dose was administered (T₀), followed by 30, 60, 90, 120, 240, 360, and 480 min after the start of antibiotic infusion. For patients receiving meropenem every 12 h, the sampling points were immediately before the antibiotic dose was administered (T₀), followed by 30, 60, 180, 360, 540, and 720 min after the start of antibiotic infusion.

All samples were promptly refrigerated at 4 °C and centrifuged at 3000 rpm for 10 min within 1 h to separate plasma. Subsequently, these samples were frozen at − 80 °C within 24 h of collection and were stored locally at participating institutions. The frozen plasma samples were shipped on dry ice and were then assayed at the central bioanalysis laboratory at the University of Queensland Centre for Clinical Research (UQCCR), Brisbane, Australia.

Data collection

A standardized case report form was used to collect relevant clinical data including patients’ demography, baseline biochemistry results (e.g., plasma albumin and creatinine), estimated renal function (via creatinine clearance, CLcr and glomerular filtration rates, eGFR), infection site, baseline disease severity (assessed using the Acute Physiology and Chronic Health Evaluation II [APACHE II] and the Sequential Organ Failure Assessment [SOFA] scores)^24,33 and antibiotic dosing details (e.g., meropenem dosing regimen, administration method, and length of therapy). The Cockcroft-Gault CLcr³⁴, CKD Epidemiology Collaboration (CKD-EPI)³⁵, and Modification of Diet in Renal Disease (MDRD)³⁶ eGFR equations were used to estimate renal function. Inotrope use was defined as the administration of any inotropic agent (e.g., norepinephrine, dopamine) on the same day of meropenem PK sampling. “Surgery” in this study refers to any surgical procedures occurring within 24 h prior to the first PK sampling (Occasion 1).

Bioanalysis

Total meropenem plasma concentrations were measured using a validated ultra-high performance liquid chromatography-MS/MS (UHPLC-MS/MS) method on a Nexera UHPLC connected to an 8030 + triple quadrupole mass spectrometer (Shimadzu, Kyoto, Japan). The analysis was performed in batches, and samples were analysed concurrently with calibration standards and quality-control replicates at high (80 mg/L), medium (4 mg/L), and low (0.6 mg/L) concentrations. The assay limit for meropenem in plasma was 0.2 mg/L and linearity was established within a range of 0.2 to 100 mg/L. Precision and accuracy at three different concentrations were all within 10%. The analysis was performed in accordance with the U.S Food and Drug Administration’s guidance for industry on bioanalysis³⁷.

PK/PD target attainment analysis

In this study, a composite PK/PD target was used, aiming to achieve 100% of the time above the minimum inhibitory concentration (100%ƒT_>MIC) while ensuring that a predefined toxic trough concentration thresholds was not exceeded³⁸. An MIC value of 2 mg/L was chosen as this corresponds to meropenem susceptibility breakpoint for P. aeruginosa according to the 2024 guidelines from the Clinical and Laboratory Standards Institute (CLSI)³⁹ and the 2024 European Committee on Antimicrobial Susceptibility Testing (EUCAST)⁴⁰. Therapeutic meropenem trough concentrations were defined as those between 2 and 45 mg/L, with concentrations below 2 mg/L deemed suboptimal and those above 45 mg/L considered toxic. These thresholds were chosen because 2 mg/L is the MIC breakpoint for susceptibility, and concentrations exceeding 45 mg/L were associated with neurotoxicity and nephrotoxicity^31,38. Meropenem total concentrations were corrected for the protein binding value of 2% to determine its free concentrations⁴¹.

Population pharmacokinetic analysis

Structural model development

Population PK analysis was performed using the nonlinear mixed-effect modeling approach using the Monolix software (version 2021R2, LIXOFT, Antony, France). This software utilizes the stochastic approximation expectation maximization (SAEM) algorithm for parameter estimation⁴². Individual estimates for PK parameters were assumed to be log-normally distributed. The between-subject variability (BSV or ω) was described using an exponential model according to the equation \({\theta }_{j}={\theta }_{p}\times \text{exp}({\eta }_{j})\), where θ_j is the estimate for a PK parameter in the jth patient as predicted by the model, \({\theta }_{p}\) is the typical population PK parameter value, and \({\eta }_{j}\) is a random variable from a normal distribution with zero mean and variance ω², which is estimated. One- and two-compartment models with first-order elimination were compared. Several error models (constant, proportional, or combined error model) were tested to describe the residual variability (ε). An exponential variability model evaluated between-subject variability (BSV) and between-occasion variability (BOV). Model selection was based on visual inspection of goodness-of-fit (GOF) plots and numerical assessment of objective function value (OFV) and the corrected Bayesian information criteria (BICc). A reduction in the OFV of > 3.84 for one degree of freedom was considered a statistically improvement (p 

Covariate model development

From the structural model, the effects of fifteen potential covariates on meropenem PK parameters were evaluated. These covariates were age, gender, weight, body mass index, APACHE II and SOFA²⁴ scores on sampling occasions, plasma albumin, estimated renal function by Cockcroft-Gault, CKD-EPI, and MDRD equations, pre-ICU stay before sampling, mechanical ventilation, surgery in the previous 24 h before the first sampling, septic shock, and the use of inotropes during sampling.

The continuous covariates were centered on their median values and categorical covariates were introduced as \({\theta }_{j}={\theta }_{\text{jTPV}}*{e}^{{\theta }_{\text{COVi}}*COVi}\), where \({\theta }_{j}\) is the value of the PK parameter j, \({\theta }_{\text{jTPV}}\) is the median value of j, and \({\theta }_{\text{COVi}}\) is the parameter estimated to represent the effect of the ith covariate (COVi) when the value is 1. A stepwise covariate modelling approach was used, consisting of forward inclusion and backward elimination steps. A reduction in − 2LL of at least 3.84 (p 

Model evaluation

Evaluation of the model was based on goodness-of-fit (GOF) plots, including observations versus individual and population predictions, weighted individual residuals versus individual predictions and time (IWRES), and plots of normalized prediction distribution error (NPDE) versus population predictions and time. The visual predictive check (VPC) was performed using 500 simulations with the final model. This plot shows the time course of the simulated profiles’ 5th, 50th, and 95th percentiles and compares them with observed data. The accuracy of the final model was also examined using a bootstrap method. A 1000-run bootstrap resampling procedure was performed in Monolix using the Rsmlx (R Speaks ‘Monolix’, version 4.0.2) package in R software (version 4.1.3). The median, 2.5%, and 97.5% values obtained from the 1000 bootstrap runs for each parameter were calculated and compared to the estimates from the original data.

Dosing simulation

Based on the final population PK model parameter estimates, the probability of target attainment (PTA) for a meropenem dosing regimen to achieve 100% ƒT_>MIC were assessed on Occasion 1 and Occasion 2 using Monte Carlo simulations (n = 1000). Moreover, the simulation also examined the proportion of patients reaching potentially toxic meropenem concentrations, defined as 45 mg/L or higher.

Sixteen meropenem dosing regimens were simulated and these regimens ranged from intermittent infusion of 0.5 g meropenem every 8 h to extended infusion of 2 g meropenem every 12 h and continuous infusion of 6 g meropenem over 24 h. The PTA of each regimen was assessed for 60 possible scenarios across ten MIC values ranging from 0.125 to 64 mg/L and six levels of eGFR by the CKD Epidemiology Collaboration (CKD-EPI) at 30, 50, 70, 90, 110, and 130 mL/min. Additional dosing simulations were planned to incorporate covariates that improve the models during covariate testing. Similar to other relevant papers¹⁵, color-coded categories were applied to visualize PTA across dosing regimens: green (≥ 90%), yellow (80–89%), orange (50–79%), and red (

Statistical analysis

The data were presented as counts and percentages for categorical variables and as median values with interquartile ranges for continuous variables. Patient characteristics and treatment-related variables were evaluated for their impact on PK/PD target attainment using the chi-square test or Fisher’s exact test for categorical variables and the Mann–Whitney U test for continuous variables. A two-sided p value of