2.1 Study Design and Setting

A prospective, observational PK study was conducted in a referral hospital in Surabaya, Indonesia. Patients aged ≥ 18 years receiving either first or multiple doses of intravenous ampicillin-sulbactam in general wards (excluding the intensive care unit [ICU]) were enrolled in this study. Patients on, or planned for, renal replacement therapy (RRT) at the time of sampling were excluded. Pregnant women were also excluded from the study.

Ethical approvals for the study were granted by the Ethics Committee of Dr Ramelan Navy Hospital (approval number 76/EC/KERS/2019) and The University of Queensland Human Research Ethics Committee (approval number 2018001592). Before collecting blood samples, written informed consent was obtained from either the patients themselves or their legally authorised representatives.

2.2 Drug Administration, Sampling Procedure and Data Collection

Ampicillin-sulbactam dosing regimens were prescribed at the discretion of the treating team. At the time of this study, there was only one product of ampicillin-sulbactam at the research site which consisted of 1000 mg of ampicillin and 500 mg of sulbactam. Doses given, administration times, and number of doses before sampling were documented at the time of sampling.

Doses of ampicillin-sulbactam were diluted with 0.9% sodium chloride just before administration as a bolus injection (over approximately 3 min). Multiple blood samples per patient were collected during one dosing interval (5 min, 20 min, 120 min, 240 min after injection and just before the next dose) with 3 mL of venous blood collected per blood sample using lithium heparin tubes. Each blood sample was placed on ice and immediately transferred to in-house laboratory for centrifugation (3000 rpm for 15 min). After centrifugation, the aliquot was stored in a – 80 °C freezer until analysis.

Relevant patient characteristics (age, gender, body weight, and body mass index) and laboratory data (serum albumin and serum creatinine, SeCr) were collected from the medical records at the time of recruitment. The Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation was used to determine the estimated glomerular filtration rate (eGFRCKD-EPI) for each patient [21, 22]. All medications administered concomitantly with ampicillin-sulbactam were recorded and checked for potential clinically significant interactions with ampicillin-sulbactam [26]. Interactions classified as “avoid combination” and “usually avoid combination” in the reference used in our study were considered as clinically significant interactions.

2.3 Ampicillin-sulbactam assay

Total concentrations of ampicillin and sulbactam were measured simultaneously in plasma using a validated ultra-high-performance chromatography-tandem mass spectrometry (UHPLC-MS/MS) method on a Nexera UHPLC system coupled to a 8030+ triple quadrupole mass spectrometer (Shimadzu, Kyoto, Japan). Test samples were assayed in batches alongside plasma calibrators and quality controls, and results were subjected to batch acceptance criteria [27]. The assay methods were linear from 1 to 200 mg/L and 0.5 to 100 mg/L for ampicillin and sulbactam, respectively. The precision at lower limit of quantification (LLOQ) of 1 mg/L (ampicillin) and 0.5 mg/L (sulbactam) was reported as 2.4% and 0.3%, respectively, while the accuracy at the same LLOQ was reported as − 1.1% and 5.8%, respectively. For ampicillin, precision of 2.5%, 1.4%, and 1.9%, and accuracy of 1.5%, − 3.0%, and − 6.4% were reported at concentrations of 3, 30 and 150 mg/L, respectively. For sulbactam, precision of 5.1%, 0.7%, and 8.2%, and accuracy of − 6.7%, 6.3%, and − 1.2% was reported at concentrations of 1.5, 15 and 75 mg/L, respectively.

2.4 Population PK Modelling2.4.1 Structural Model

Ampicillin and sulbactam concentration data were fitted to generate a population PK model using Non-Parametric Adaptive Grid (NPAG) programme in Pmetrics version 1.9.7. (Laboratory of Applied Pharmacokinetics and Bioinformatics, Los Angeles, CA, USA) for R (version 4.0.1) [23]. A structural model of ampicillin and sulbactam was first developed by comparing the one- and two-compartment model without introducing covariates. Both lambda and gamma error models were tested for ampicillin and sulbactam PK models.

The best structural model was chosen according to the goodness-of-fit of both the population- and individual-observed versus predicted concentration plots, the value of the − 2*Log-likelihood (− 2LL), and the Akaike Information Criterion (AIC). The goodness-of-fit of the models was assessed by visual inspection of the observed versus predicted concentration plots both in population- or individual-scatter plots, the coefficient of determination (R2), slopes, intercept, and bias of the linear regression [28, 29]. For AIC and − 2LL, models with lower values were considered to be better than the comparator with a decrease of 3.84 unit of − 2LL considered statistically significant.

2.4.2 Covariate Analysis

The effect of several biologically plausible demographic and clinical characteristics on the PK of ampicillin and sulbactam was assessed. The variables tested for inclusion were age, gender, weight, body mass index, serum albumin, SeCR, and eGFRCKD-EPI. The final estimated PK parameters from the final model with covariates were presented as mean, standard deviation (SD), percentage coefficient of variation (%CV), median value, and the percentage of shrinkage. The %CV represented the inter-individual variability.

2.4.3 PK Model Diagnostics

The best structural model was chosen according to the goodness-of-fit of both the population- and individual-observed versus predicted concentration plots, the value of the − 2*Log-likelihood (− 2LL), and the AIC. The goodness-of-fit of the models was assessed by visual inspection of the observed versus predicted concentration plots both in population- or individual-scatter plots, the coefficient of determination (R2), slopes, intercept, and bias of the linear regression [23]. For AIC and − 2LL, models with lower values were considered to be better than the comparator with a decrease of 3.84 unit of − 2LL considered statistically significant. Once the structural model was chosen, each covariate was separately added to that particular model and only covariates resulting in a statistical decrease of − 2LL (a decrease of 3.84 units) and AIC, whilst improving the goodness-of-fit of the scatter plots, were retained in the final model. Internal validation was conducted via a visual predictive check (VPC) to evaluate the predictive performance of the final model with covariates using 1000 simulations. The distribution of the observed concentration in this simulation was plotted and visually inspected.

2.4.4 Monte Carlo Dosing Simulation

The final model with covariates was used in the Monte Carlo dosing simulations to identify the dosing strategy of ampicillin and sulbactam with the highest likelihood of achieving target drug exposures. Dosing regimens simulated were selected according to the approved product information that was based on creatinine clearance (CLCr) ranges [11, 12]:

1.

Ampicillin: 1000 mg (corresponds to 1500 mg ampicillin-sulbactam) and 2000 mg (corresponds to 3000 mg ampicillin-sulbactam) every 24 h (for CLCr 10 mL/min/1.73 m2), every 12 h (for CLCr 20 mL/min/1.73 m2), every 6 and 8 h for (for CLCr ≥ 30 mL/min/1.73 m2).

2.

Sulbactam: 500 mg (corresponds to 1500 mg ampicillin-sulbactam) and 1000 mg (corresponds to 3000 mg ampicillin-sulbactam) every 24 h (for CLCr 10 mL/min/1.73 m2), every 12 h (for CLCr 20 mL/min/1.73 m2), every 6 and 8 h for (for CLCr ≥ 30 mL/min/1.73 m2).

In order to better identify the influence of renal function on the achievement of PK/PD targets, the typical simulated patients with CLCr ≥ 30 mL/min/1.73 m2 were further divided into three (3) groups including: 30, 70, 100 mL/min/1.73 m2. Each dosing regimen was simulated in Pmetrics® (version 1.9.7), with 1000 subjects as either receiving a bolus injection or prolonged infusion over 4 h.

Fixed protein binding values of 28% and 38% for ampicillin and sulbactam, respectively, were used in the dosing simulations [24, 25]. In general, %fT>MIC value of ≥ 50% provided maximal bactericidal effect for ampicillin based on preclinical studies [13]. While for sulbactam, it was found in an in vivo murine thigh infection model that ≥ 60% of fT>MIC would result in maximal bactericidal effect (3 log10 kill against A. baumannii) [11]. However, a recently published consensus paper on antimicrobial therapeutic drug monitoring in critically ill patients recommended a PK/PD target of 100% fT>MIC [26]. It is worth mentioning that even though our study was conducted predominantly in a non-ICU setting, due to limited ICU capacity, critically ill patients were often also treated in general wards. Therefore, the achievement of 60% fT>MIC and 100 % fT>MIC were both used as a priori PK/PD targets for ampicillin-sulbactam in our study.

The percentage of simulated patients that could attain the PK/PD targets was used to calculate the probability of target attainment (PTA). Since adequate PK/PD exposures attained from the beginning of treatment would be important for therapeutic success, the PTAs in our study were calculated both after the first administration and at steady state (i.e., after the fifth administration) [27]. The optimal PTA for a specific MIC was defined as ≥ 90% [34]. To calculate the fractional target attainment (FTA), the PTA of each dosing regimen was compared against the MIC distribution of relevant pathogens. The FTAs of ampicillin were calculated against Escherichia coli, Klebsiella pneumoniae, and Streptococcus pneumoniae, while for sulbactam, the FTA was calculated against A. baumannii. Two FTA assessments were performed. First, FTAs of ampicillin and sulbactam were calculated against the whole MIC distribution of bacteria (ranging from 0.002 to 512 mg/L) derived from European Committee on Antimicrobial Susceptibility Testing (EUCAST) database [35]. The intention of evaluating PK/PD exposures against entire MIC distributions was to select optimal dosing regimens in such situations where the MIC of the pathogen was unknown (i.e., when used as empiric therapy). Second, FTAs of ampicillin were calculated against a fixed MIC range for susceptible strains of selected pathogens (i.e., directed therapy), including: ≤ 8 mg/L, ≤ 8 mg/L, and ≤ 0.5 mg/L for E. coli, K. pneumoniae, and S. pneumoniae, respectively [35]. The directed FTAs for sulbactam against A. baumannii were calculated against MIC ≤ 4 mg/L [28]. Findings from the second assessment are relevant where susceptibility reports are made without including the actual MIC value, as often seen in LMIC countries [18, 20]. Any dosing regimen that achieved optimal FTA (i.e., defined as ≥ 95%) was considered a successful dosing regimen either for directed or empirical based therapy.

2.4.5 Statistical Analysis

Descriptive analysis using frequencies (%) for categorical data and mean (± standard deviation; SD) for continuous data in the demographic of patients were conducted using Microsoft Excel v2016.