Comparison of performances of the open-source R package “nlmixr” vs. Monolix for population pharmacokinetics of continuous infusion meropenem in patients with onco-hematological malignancies
Chun Liu(1), Pier Giorgio Cojutti(2), Linda Bussini(3), Elena Rosselli Del Turco(3), Michele Bartoletti(1,3), Federico Pea(1,2)
1. Department of Medical and Surgical Sciences, Alma Mater Studiorum, University of Bologna, Bologna, Italy. 2. Clinical Pharmacology Unit, IRCCS Azienda Ospedaliero Universitaria di Bologna, Bologna, Italy. 3. Infectious Diseases Unit, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Bologna, Italy. Acknowledgement: This project has received funding from European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No 861323.
Introduction
Meropenem is a carbapenem used for empirical escalation therapy of onco-hematological patients with febrile neutropenia who are at risk of developing severe infections caused by multi-drug resistant Gram-negative pathogen (1). Its time-dependent anti-bacterial activity is optimized when meropenem is administered by 24h continuous infusion (CI).
Population pharmacokinetics of 24h CI meropenem has never been investigated by parametric modelling. One of the most established parametric software is Monolix. Its high-reliability estimation performances coupled with strong covariate analysis statistics and easy-to-use environment for plotting have made it one of the most used for population pharmacokinetic/pharmacodynamic analysis. “Nlmixr” is an open-source package of R for non-linear mixed-effects modelling. It is largely built on a previously developed RxODE package and has no external dependencies that require licensing.
The study aimed to compare the performances of “nlmixr” and “Monolix” (2021R1) in estimating the population pharmacokinetics of CI meropenem from real-life patients with onco-hematological malignancies.
Material&Method
This retrospective study was conducted among onco-hematologic patients who were admitted at the Division of Hematology of IRCCS Azienda Ospedaliero Universitaria di Bologna, Italy, from Jan 2021 to Feb 2022 and received 24h CI meropenem for febrile neutropenia treatment. Due to stability restrictions, 24h CI was granted by reconstituting each meropenem aqueous solution every 6h and by infusing it over 6h. Therapeutic drug monitoring of meropenem steady-state concentration (Css) was applied for dose adjustment. Demographic and clinical data (meropenem dose and serum creatinine) were retrieved from medical records. Creatinine clearance (CLCR) was estimated by CKD-EPI formula.
As meropenem was administered by CI, a 1-compartmental model with linear elimination was chosen for model fitting. A proportional error model was used in the base model. Covariate analysis was performed and covariate was retained only if decreases of the Akaike information criteria (AIC), Bayesian information criteria (BIC), and objective function value (OFV) of ≥3.84 were observed compared to the base model. Model performances were assessed by regression analysis of observed values compared with either population or individual predictions, the distribution of the weighted residuals and the visual predictive check (VPC) plot.
Results
76 patients (52 males, 68.4%) with a median (range) age, height, weight, and CLCR of 62 (17-83) years, 1.70 (1.48-1.95) m, 71 (45-107) kg, and 103 (16-211.9) mL/min/1.73m², respectively, were included in this study. Overall, 111 meropenem Css were used for model building.
The R package “nlmixr” and Monolix gave similar CL estimates (12.30 and 12.45 L/h, respectively) but quite different V estimates (28.5 and 13.82 L, respectively). Between subject variability (CV%) was high both for CL (90.7 and 72.5%, respectively) and V (47.1 and 940%, respectively).
The R-squared of the regression between observation-vs.-population predictions and observation-vs.-individual predictions were available in Monolix, and the R² values were 0.20 and 0.5 respectively. Covariate analysis was possible only in Monolix, but no covariates were retained in the final model (ΔOFV, ΔAIC, and ΔBIC from the base model of 3.4, 1.4, and -0.91, respectively). The distribution of residuals and the VPC graphs were directly provided only by Monolix, while “nlmixr” requires another package (nlmixr.xpose) for plotting outputs.
Conclusion
Our analysis showed that both “nlmixr” package of R and Monolix may adequately fit the pharmacokinetics of CI meropenem from real-life patients, as the CL estimates were very close to each other and similar to the values previously reported among 61 oncohematological patients with febrile neutropenia (13.04 L/h) (2). As far as V estimation is concerned and with the limitation of estimating it with CI data, “nlmixr” performed better than Monolix, as the V was closer to that previously reported in the literature (21.88 L) (2) and with a much lower CV with respect to Monolix.
However, the fact that all the diagnostics (regression of the observed vs. population and individual predictions, distribution of residuals and VPC) were not directly accessible with “nlmixr”, constitutes an important limitation that should be addressed for more widespread use of this software in clinical pharmacology.
References
- Blennow, O. & Ljungman, P. Infections in Hematology Patients. Concise Guide to Hematology 503–518 (2018) doi:10.1007/978-3-319-97873-4_38.
- Cojutti P et al. Population Pharmacokinetics of Continuous-Infusion Meropenem in Febrile Neutropenic Patients with Hematologic Malignancies: Dosing Strategies for Optimizing Empirical Treatment against Enterobacterales and P. aeruginosa. Pharmaceutics 2020, 12, 785; doi:10.3390/pharmaceutics12090785