Comparative Plasma and Interstitial Tissue Fluid Pharmacokinetics of Meropenem Demonstrate the Need for Increasing Dose and Infusion Duration in Obese and Non-obese Patients

Detailed information about study design, procedures, and data collection in this prospective, parallel-group, open-label, controlled single-center trial (EudraCT No. 2012-004383-22) have been described elsewhere [17]. Inclusion criteria were abdominal surgery, age ≥ 18 years, body mass index (BMI) = 18.5–30 kg/m² for non-obese and BMI ≥ 35 kg/m² for obese patients. Exclusion criteria comprised liver cirrhosis. Non-obese patients were age and sex matched to those in the obese patient group.

Dense blood sampling (pre-dose and after 0.5, 1, 2, 3, 4, 5, 6, and 8 h) and collection of microdialysate samples in ISF of subcutaneous adipose tissue (pre-dose and 0–0.5, 0.5–1, 1–1.5, 1.5–2, 2–3, 3–4, 4–5, 5–6, 6–7, and 7–8 h) of both upper arms (one catheter per arm) were performed. Catheters were inserted into both upper arms (one catheter per arm) to allow quantification of catheter-related variability [18] and to allow precise estimation of pharmacokinetic parameter values associated with the ISF. To derive drug concentration in ISF from microdialysate concentrations, the retrodialysis calibration method was used [18‐20]. Sampling, preparation, and storage are described by Simon et al. [17], and meropenem concentrations were quantified using validated high-performance liquid chromatography ultra-violet detection (lower limit of quantification = 0.1 mg/L in plasma and 0.02 mg/L in microdialysate) [21]. Based on in-process quality control samples (50, 15, and 0.5 mg/L in plasma or 15, 1.5, and 0.15 mg/L in 0.9% NaCl as the surrogate for microdialysate, respectively), the mean intra-/interassay imprecision and inaccuracy were < 4% coefficient of variation (CV). The autosampler stability (20–24 h/6°C) was 99.2 ± 3.6% (plasma) and 98.3 ± 4.1% (microdialysate/retrodialysate). Further details on the bioanalysis of meropenem are provided elsewhere [21].

Nonlinear mixed-effects model development and simulations were performed in NONMEM v7.4.3 (Icon Development Solutions, Ellicott City, MD, USA, FOCE Interaction) accessed with PsN v4.8.1 through Pirana v2.9.6 (Certara, Princeton, NJ, USA). RStudio v1.2.1335 (Boston, MA, USA) was used for dataset preparation, model evaluation, and post-processing of results. Stepwise covariate modeling was performed in PsN v4.8.1.

Meropenem standard (observations available in this clinical study) and alternative dosing regimens (Table S1 in the electronic supplementary material [ESM]) in both patient populations were evaluated via Monte Carlo simulations based on the developed population pharmacokinetic model. Two- and three-compartment (mammillary and serial) pharmacokinetic models with plasma data attributed to the central and ISF data to the central or peripheral compartments were evaluated using the integrated dialysate-based modeling approach [22, 23]. Elimination from the central compartment and intercompartmental flows were assessed using linear and nonlinear processes. Between-patient variability was implemented assuming a log-normal distribution of the individual parameters. Microdialysis technique-related variabilities between patients, within catheters, and between catheters were characterized assuming a logit-normal distribution to confine relative recovery values to 0–100%. Additive, proportional, combined additive and proportional, and log-normal models of residual unexplained variability were evaluated.

Statistical comparisons between nested models with additional covariates were made using the likelihood ratio test and based on the Akaike information criterion for non-nested models [24]. Model selection was based on statistical significance (p < 0.05) and the plausibility and precision of parameter estimates.

Population pharmacokinetic models were evaluated with standard goodness-of-fit plots (e.g., observed vs. predicted meropenem concentrations and conditional weighted residuals vs. population prediction/time). The predictive model performance was assessed with visual predictive checks (n = 1000), and the uncertainty, bias, and stability of parameter estimates was evaluated using the nonparametric bootstrap (n = 1000) [38]. Statistically significant differences (p < 0.05) in individual parameter estimates between obese and non-obese patients were investigated using the Mann–Whitney–Wilcoxon test.

To demonstrate the impact of ABW and CLCR_{CG_ABW} on meropenem exposure, simulations of meropenem plasma and ISF concentrations over 8 h following short-term, prolonged (both 1000 mg), and continuous (3000 mg/24 h) infusion were performed covering the study range of ABW (ABW = 60–120 kg) and CLCR_{CG_ABW} (CLCR_{CG_ABW} = 60–200 mL/min) covering the range of patient characteristics in the study cohort (Table 1).

The extent of meropenem penetration into ISF in obese versus non-obese patients was evaluated by the commonly evaluated ratio of the area under the unbound drug concentration–time curve (AUC) between 0 and 8 h (fAUC_0–8h) in ISF to plasma (i.e., [fAUC_{0–8h, ISF}] : [fAUC_{0–8h, plasma}]; penetration index). Given the negligible plasma protein binding (~ 2%) of meropenem, it was considered unbound in plasma [39].

Since the antibiotic effect of β-lactams such as meropenem has been demonstrated to be time dependent [5, 6], an AUC-related penetration index is irrelevant for the outcome of meropenem therapy, so we also evaluated the ratio of the effect-related %fT _{> MIC} in ISF to plasma (i.e., [%fT _{> MIC, ISF}] : [%fT _{> MIC, plasma}]; effective penetration index). Both were evaluated for (1) ABW = 60–120 kg at healthy renal function (CLCR_{CG_ABW} = 100 mL/min) and (2) CLCR_{CG_ABW} = 60–140 mL/min at ABW = 60.0 kg. To investigate whether PK/PD targets related to 4 × MIC for continuous infusions are suitable to avoid ISF concentrations below MIC for the entire dosing interval (i.e., effective penetration index = 0) %fT _{> MIC} was related to 4 × MIC in plasma and 1 × MIC in ISF.

To evaluate whether meropenem dosing regimens are sufficient to achieve effective concentrations in obese and non-obese patients when non-susceptible or resistant bacteria are encountered in the clinic, PTA analyses were performed for susceptibility and resistance breakpoints (EUCAST). To determine the PTA of various simulated dosing regimens, Monte-Carlo simulations (n = 1000 per covariate combination) of meropenem concentrations over time in plasma and ISF for the first day of treatment were performed using the developed population pharmacokinetic model and virtual patients without the effect of anesthesia. PTA for achieving a PK/PD target of %fT _{> MIC} = 95 [6] and %fT_{> MIC} = 40 [5] was calculated for MIC = 0.25 mg/L, 2 mg/L (EUCAST epidemiological cut-off value for P. aeruginosa and Enterobacteriaceae), and 8 mg/L (non-species-related resistance breakpoint). These PK/PD targets were selected to cover a broad range of PK/PD targets and to ensure comparability with other PK/PD evaluations of meropenem in (morbidly) obese patients [7, 8, 16]. The PK/PD target of %fT _{> MIC} = 95 was selected instead of %fT _{> MIC} = 100, since treatment at day 1 was evaluated during PTA analysis and meropenem concentrations in all matrices are zero before the first dose, preventing any concentration–time profile from attaining %fT  _{> MIC} = 100. For continuous infusions, stricter PK/PD targets were selected: %fT _{> MIC} was related to 4 × MIC [10].

To determine whether meropenem dosing regimens are sufficient to achieve effective concentrations in obese and non-obese patients in routine clinical settings, we calculated the sum of PTA weighted by the relative frequency of MIC values (cumulative fraction of response [CFR] [40]) for specific pathogens. Infections by pathogens commonly treated with meropenem [41] were selected (E. coli, C. freundii, K. pneumoniae, P. aeruginosa).

A dosing regimen was considered adequate if the PTA or CFR was ≥ 90% [15]. We also assessed whether the meropenem toxicity threshold concentration was reached, for which a 50% risk of developing a neurotoxicity event has been reported [42]. The dosing regimens evaluated via simulations were thrice-daily (TID) intravenous short-term (30 min) or prolonged (3 h) infusions of 1000 and 2000 mg meropenem over 24 h, and continuous infusions of 3000 and 6000 mg meropenem starting immediately after a 1000-mg 30-min infusion loading dose (Table S1 in the ESM). PTA was assessed for different combinations of ABW (60–120 kg) and CLCR_{CG_ABW} (60–200 mL/min).

In total, 30 patients (15 obese and 15 non-obese) scheduled for elective abdominal surgery were recruited according to study protocol [17]. The 15 obese (two class II and 13 class III) patients were scheduled for bariatric surgery, and the 15 non-obese patients were undergoing elective abdominal surgery, mainly tumor resection (11 patients underwent gynecological operations, and the remaining four operations involved the stomach, liver, kidneys, or appendix). The cohort was predominantly female (26/30 patients) and covered a wide BMI range (BMI_non–obese = 20.5–27.1 kg/m² and BMI_obese = 38.1–81.5 kg/m²). As expected, heart rate and blood pressure were reduced during anesthesia versus the post-anesthetic period, as indicated by the derived mean arterial blood pressure (Table 1).

In total, 915/932 (observed/planned) meropenem concentrations were available, comprising plasma concentrations (n = 239/240), microdialysate concentrations collected via two catheters (to investigate microdialysis-related variability) in the ISF of subcutaneous adipose tissue over 8 h (n = 292/300 + 293/300), and retrodialysate concentrations (n = 46/46 + 45/46).

The geometric mean, selected to account for the heteroscedasticity of observed maximum meropenem concentrations (C_max) in plasma and at target site were lower in obese than in non-obese patients (ΔC_{max, plasma} = − 19.6% and ΔC_{max, target site} = − 38.5 to − 40.9%; Fig. 1; Fig. S1 in the ESM), whereas observed minimum concentrations (C_min) were higher in obese than in non-obese patients (ΔC_{min, plasma} = + 49.7% and ΔC_{min, target site} = + 38.6 to + 54.3%).

The developed pharmacokinetic model was a serial three-compartment model (Table 2, Fig. 2) with a significantly higher total meropenem volume of distribution (V₁ + V₂ + V₃, median [range]) in obese (22.8 L [18.0–38.8]) than in non-obese patients (15.4 L [12.1–18.0]; p < 0.001) and similar clearance for obese (CL = 12.2 L/h [6.33–20.6]) and non-obese patients (CL = 10.5 L/h [5.38–14.6]; p = 0.250; Table 3). Individual volumes of distribution were higher in obese than in non-obese patients for V₁, V₂, and V₃ (V_1,obese = 10.1 L [6.09–17.5]; V_{1,non–obese} = 6.67 L [3.87–11.5]; p = 0.001; V_2,obese = 11.7 L [8.04–17.9]; V_{2,non–obese} = 7.49 L [5.28–9.95]; p < 0.001; V_3,obese = 3.06 L [2.28–4.41], V_{3,non–obese} = 2.07 L [1.77–2.51]; p < 0.001). Concentrations in ISF were related to the first peripheral compartment, which was best described by a tissue factor, scaling predicted meropenem concentrations in this peripheral compartment to ISF concentrations (Fig. 2).

The effect of ABW on central and peripheral volumes (V_1–3), CL, and intercompartmental flows (Q_1–2) was implemented using theory-based allometric scaling with fixed exponents [25] (i.e., positive relationships of pharmacokinetic parameters with body size with the exponents 1 for V_1–3 and 0.75 for CL and Q_1–2, respectively) based on likelihood ratio testing (Table S2 in the ESM). When, instead, exponents were estimated separately for flows and volumes, the decrease in objective function value (OFV) was not significant (p = 0.104, ΔOFV = − 2.64) and the 90% confidence interval (CI) of estimates included the theory-based allometric exponents of 0.75 for flows (0.505; 90% CI = 0.135–0.876) and 1 for volumes (0.757; 90% CI = 0.408–1.10).

Additionally, positive relationships between CLCR_CG (via total, lean, and adjusted body weight) and predicted glomerular filtration rate (via the MDRD and CKD-EPI equations de-indexed by individual body surface area; Fig. S3 in the ESM) on CL were evident. The very high concordance (Lin’s concordance correlation coefficient ≥ 0.962; Fig. S4 in the ESM) between CLCR_{CG_ABW} and de-indexed predicted glomerular filtration rate via the MDRD and CKD-EPI equations resulted in similar model performance (Table S2 in the ESM). Since ABW was selected during allometric scaling and to enhance simplicity, CLCR_{CG_ABW} was selected in the final model.

CL was successfully split into a renally filtered part related to CLCR_{CG_ABW} (CL_Rfilt = 5.48 L/h; 29% relative standard error [RSE]) and CL accounting for remaining elimination pathways scaling with ABW (CL_nonfilt = 4.99 L/h, 32.5% RSE).

Implementation of these two patient characteristics as covariates entirely explained differences in population parameter estimates between obese and non-obese patients.

Relative recovery estimates for catheters were variable and significantly lower for obese (31.0% [26.4–61.6]) than non-obese patients (55.4% [31.5–81.5]; p < 0.001). Additionally, the anesthesia status of the patient and mean arterial blood pressure impacted the tissue factor and CL_nonfilt, respectively: post anesthesia, the tissue factor was 19.0% (11.4% RSE) larger than during anesthesia, and a 10% increase of mean arterial blood pressure increased CL_nonfilt by 6.2% (37% RSE).

Inclusion of these covariates considerably decreased the between-patient and retrodialysis-related variability compared with the base model (15.1–31.0% relative reduction in CV). Microdialysis/retrodialysis data with repeated measurements from two catheters allowed quantification of microdialysis technique-related variabilities, which were large between patients (σ² = 0.388) and within catheters (σ² = 0.515) and comparably low between catheters (σ² = 0.157).

The key model development steps are detailed in Table S2 in the ESM. Adequate model prediction for all three matrices was demonstrated by visual predictive checks (Figs. S5–6 in the ESM): The final pharmacokinetic model adequately captured both the central tendency and the variability of meropenem pharmacokinetics in all investigated matrices. Additionally, all meropenem pharmacokinetic- and microdialysis technique-related parameters were precisely estimated (RSE ≤ 49.8%, Table 2), and the results of model evaluation demonstrated appropriate model performance (Figs. S7–8 in the ESM). A small trend of underpredicting low meropenem plasma concentrations in obese patients and overpredicting low concentrations in non-obese patients remained (Fig. S7 in the ESM, top panel). This trend remained when allometric exponents were estimated instead of using theory-based allometric exponents (abolsute average fold error [AAFE]: AAFE_obese = 0.996, AAFE_non–obese = 1.05 and slope of observed vs. predicted plasma concentrations: Slope_obese = 0.985; 95% CI = 0.938–1.03 and Slope_non–obese = 0.958; 95% CI = 0.910–1.01). No trends were found when inspecting random-effects variables for structural pharmacokinetic parameters versus the identified covariates (Figs. S9–10 in the ESM). Additionally, after implementation of the impact of body mass on pharmacokinetic parameters, no additional impact of obesity on any pharmacokinetic parameter remained in the final pharmacokinetic model (p > 0.05) as judged by likelihood ratio testing. Visual predictive checks stratified by obesity status demonstrated that, for low plasma concentrations, the trend in the median of observed values for obese and non-obese patients was also observable. However, for obese and non-obese patients, not only the median of observed values was within the 95% CI of the median of predicted concentrations (Fig. S6 in the ESM, top panel) [43]. This also applied to the 5th and 95th percentiles of meropenem plasma concentrations. This qualified the model application for the intended use of simulating with interindividual variability in both subpopulations.

Simulated maximum concentrations in patients with ABW = 120 kg (obese) compared with ABW = 60 kg (non-obese) were lower in plasma and in ISF (34.9–47.8% decrease throughout all dosing regimens, Fig. 3A). In the final descending phase within one dosing interval (8 h), an inverse trend was observed for short-term and prolonged infusion dosing regimens (Fig. 3A, left and middle panels), with patients of ABW = 120 kg having higher minimum plasma and ISF concentrations (50.0–332% increase). For continuous infusions (Fig. 3A, right panel), this inverse effect was not evident: Simulated meropenem plasma and ISF concentrations in patients with ABW = 120 kg were continuously below simulated concentrations in patients with ABW = 60 kg. Throughout all simulated dosing regimens, patients with CLCR_{CG_ABW} = 200 mL/min had lower minimum plasma and ISF concentrations than patients with CLCR_{CG_ABW} = 60 mL/min (from 10.8% reduction to even no remaining concentration; Fig. 3C).

The model-predicted, commonly evaluated ISF to plasma ratio of fAUC_0–8h (penetration index) was similar over the ABW range: 30.7% (95% CI = 25.0–37.8) for 120 kg vs. 30.2% (95% CI = 24.7–37.2) for 60 kg. Similarly, for simulated short-term and prolonged infusions (Fig. 3B; Fig. S11 in the ESM, left and middle panels), the ISF to plasma ratio of %fT _{> MIC} (effective penetration index) was relatively similar over the investigated ABW range for MIC ≤ 2 mg/L (≤ 11.9% relative reduction). In contrast, the effective penetration index for MIC = 8 mg/L decreased from 60 to 120 kg ABW (≥60.3% relative reduction, Fig. 3B; Fig. S11 in the ESM, upper panels). Similar trends were observed for varying CLCR_{CG_ABW}, yet relative differences were lower than for ABW (Fig. 3D; Fig. S11 in the ESM, lower panels).

For simulated continuous infusions (%fT _{> MIC} related to 4 × MIC in plasma to avoid suspected ISF concentrations constantly below the MIC and 1 × MIC in ISF; Fig. 3B, right panel) ISF:plasma %fT _{> MIC} ratios were ≥0.983 over the entire ABW and CLCR_{CG_ABW} range (Fig. 3B, D; Fig. S11 in the ESM, right panels). With increasing MIC values, the ISF:plasma %fT _{> MIC} ratio also increased since %fT _{> MIC} in plasma (related to 4 × MIC) decreased faster than %fT _{> MIC} in ISF (compare Fig. S11 in the ESM). Simulated unbound steady-state concentrations were 2.68 times (95% CI = 2.11–3.37) higher in plasma than in ISF over the whole ABW range (Fig. 3A, right panel).

When evaluating susceptibility and resistance breakpoints, for short-term and prolonged infusions, PTA (%fT _{> MIC} = 95) increased with increasing ABW because of the aforementioned inversion of concentration at the end of the dosing interval (Figs. S12–13 in the ESM). Consequentially, at ABW > 90 kg and CLCR_{CG_ABW} ≤ 100 mL/min for %fT _{> MIC} = 95 and MIC = 2 mg/L, PTA was adequate with 2000 mg TID prolonged infusions (Table 4, Fig. 4E). However, for the continuous infusion, higher body mass resulted in decreased PTA, leading to adequate PTA only for ABW < 90 kg (%fT_{> MIC} = 95) with a 3000 mg continuous infusion for MIC = 2 mg/L (inadequate PTA for all patients at MIC = 8 mg/L) and patients with CLCR_{CG_ABW} ≤ 100 mL/min (Table 4, Fig. 4C).

Overall, PTA analysis showed lower PTA at higher CLCR_{CG_ABW} (Table 4). In all patients for %fT _{> MIC} = 95, a TID prolonged infusion of 1000 mg was sufficient to achieve adequate PTA for MIC = 0.25 mg/L, whereas only a continuous infusion of 6000 mg sufficed for MIC = 2 mg/L. None of the investigated dosing regimens was adequate for MIC = 8 mg/L. For the lower target, %fT _{> MIC} = 40, the TID short-term infusion of 2000 mg and both prolonged infusion dosing regimens were adequate for MIC ≤ 2 mg/L, and a 2000 mg TID prolonged infusion was even sufficient for MIC = 8 mg/L. PTA results for additional covariate combinations are presented in Fig. S12–S14 in the ESM.

The CFR for the standard meropenem dosing regimen (short-term infusions of 1000 mg TID) was sufficient for CLCR_{CG_ABW} ≤ 130 mL/min and commonly encountered MIC distributions for E. coli, C. freundii, and K. pneumoniae (Fig. 4, top panel). For P. aeruginosa, the CFR of the meropenem standard dosing regimen was inadequate for all virtual patients (Fig. 4, top panel).

In addition to efficacy, the reported neurotoxicity threshold [42] was not reached for the scenario with the highest meropenem exposure, i.e., the highest investigated daily dose (meropenem 6000 mg continuous infusion with 1000 mg loading dose) and the lowest investigated CLCR_{CG_ABW} and ABW after 24 h (Fig. S15B in the ESM).