Successful validation of a model-informed precision dosing instrument for meropenem in critically ill patients, the DoseCalculator, against NONMEM®
Franz Eric Weber (1,2), Ferdinand Anton Weinelt (1,2), Christin Nyhoegen (1), Frieder Pfäfflin (3), Anja Theloe (4), Ulrike Trost (3), Peggy Kießling (5), Wilhelm Huisinga (6), Sebastian Georg Wicha (7), Robin Michelet (1), Stefanie Hennig (1,8,9), Miriam Songa Stegemann (3), Charlotte Kloft (1)
(1) Freie Universitaet Berlin, Institute of Pharmacy, Dept. of Clinical Pharmacy & Biochemistry - Berlin (Germany), (2) Graduate Research Training program PharMetrX, Berlin/Potsdam (Germany), (3) Department of Infectious Diseases and Respiratory Medicine, Charité-Universitaetsmedizin Berlin, Corporate Member of Freie Universitaet Berlin, Humboldt-Universitaet zu Berlin, and Berlin Institute of Health, Berlin (Germany), (4) Pharmacy, Charité-Universitaetsmedizin Berlin, Corporate Member of Freie Universitaet Berlin, Humboldt-Universitaet zu Berlin, and Berlin Institute of Health, Berlin (Germany), (5) Labor Berlin-Charité Vivantes GmbH, 13353 Berlin (Germany), (6) Institute of Mathematics, University of Potsdam (Germany), (7) Institute of Pharmacy, University Hamburg (Germany), (8) School of Clinical Sciences, Faculty of Health, Queensland University of Technology, Brisbane, QLD 4000 (Australia), (9) Certara, Inc., Princeton, New Jersey (USA)
Objectives: Model-informed precision dosing (MIPD) instruments can optimise antibiotic therapy by translating available clinical data to individualised dosing recommendations but often lack target-user-friendliness and integrability of specific local clinical data. Thus, the DoseCalculator, an easy-to-use MIPD instrument for dosing regimen optimisation of meropenem in critically ill patients, was developed based on the algorithms of TDMxR within an interprofessional, prospective observational study at Charité - Universitätsmedizin Berlin (Charité) [1,2]. The DoseCalculator allows for including Therapeutic Drug Monitoring, dosing, patient and local bacterial susceptibility data when available to predict the best dosing regimen for an individual. Within the DoseCalculator, by leveraging the underlying pharmacokinetic (PK) model, (i) estimation of maximum a posteriori (MAP) parameters and (ii) simulations with these MAP parameters and their posterior distribution approximated by the variance-covariance matrix of the individual ETAs (VCOV) can be performed [3,4]. For clinical implementation, we aimed to validate the TDMxR algorithm within DoseCalculator for (i) and (ii) against the current reference software in population PK modelling and simulation, NONMEM® (NM), for meropenem with regard to clinical comparability.
Methods: A validation dataset based on patient records collected within our clinical study (Charité Ethics Committee, EA4/053/19) between 2019 and 2020 was used (npatients=53, nsamples=181) [5]. First, MAP parameters and their VCOV were estimated in the DoseCalculator ('TDMxR::estimate.map()') and in NM ('MAXEVAL=0'). Second, based on the estimation results, a standard 4 h infusion of 2 g meropenem every 8 h was simulated for each patient in the DoseCalculator using the entire VCOV ('TDMxR::simulate.scenario(type = "ind_var_vcov")'), and in NM, using the diagonal elements of the VCOV. The MAP estimation was deemed clinically equivalent if the DoseCalculator MAP parameters accuracy, described as relative Bias (rBias) and median absolute relative error (MARE), was <±10% and <10%, respectively, compared to NM [6]. To assess precision, the relative root mean squared error (rRMSE) values were calculated for all MAP parameters. rRMSE should be as low as possible (ideally 0%) [7]. To assess the agreement between the MAP estimation in the DoseCalculator and NM, MAP parameter (DoseCalculator) vs MAP parameter (NM) plots were generated, and Bland-Altman analyses were performed. For the simulation results, rBias and MARE of the median, 5th (P0.05) and 95th (P0.95) percentile of the individually predicted meropenem concentrations (IPRED) were to be <±20% and <20%, respectively [7,8,9]. rRMSE values were also calculated for the median, P0.05 and P0.95 of IPRED, which should be, again, ideally, 0% [7]. The IPRED median, P0.05 and P0.95, were plotted against time and depicted in the same graph to compare the predictions directly between both software.
Results: The analysis demonstrated high accuracy and precision for the MAP parameter estimation within the DoseCalculator: Individual clearance values exhibited an rBIAS of -0.294%, MARE of 0.0674%, and rRMSE of 1.07%; the central volume of distribution an rBIAS of 0.191%, MARE of 0.272%, and rRMSE of 0.990%, and the peripheral volume of distribution an rBIAS of 0.0168%, MARE of 0.201%, and rRMSE of 0.517%. The MAP parameter (DoseCalculator) vs MAP parameter (NM) and Bland-Altman plots confirmed the results by showing high concordance for the individually generated parameters. The median of the IPRED showed a rBIAS of 0.145% (P0.05: 18.2%, P0.95: -12.8%), MRE of 0.188% (P0.05: 9.68%, P0.95: 11.4%), and rRMSE of 0.452% (P0.05: 25.0%, P0.95: 14.6%) compared to NM. Visually, the IPRED revealed high agreement, whereas slightly higher deviations were apparent for P0.05 and P0.95. In all cases, the prespecified acceptance criteria were met.
Conclusions: Overall, our validation showed a high level of agreement between the TDMxR algorithm within the DoseCalculator compared to NM for the MAP estimation and simulated concentration-time profiles of individual patients. This comparability further supported the utilisation of the DoseCalculator within clinic practice. Next, the DoseCalculator will be implemented into our study's antimicrobial stewardship program to test its benefits in terms of user-friendliness, the probability of target attainment increase, and daily dose reduction.
References:
[1] Wicha SG et al. Int. J. Antimicrob. Agents 45: 442–444 (2015).
[2] Weber FE , Weinelt FA et al. ECCMID 2023 Abstr E1325 [https://elibrary.escmid.org/]
[3] Weinelt FA et al. Pharmaceutics, 13, 2128 (2021).
[4] Ehmann L et al. Int. J. Antimicrob. Agents 54: 309–317 (2019).
[5] Weinelt FA et al. Antibiotics Vol. 11, Page 758 11: 758 (2022).
[6] Le Louedec F et al. CPT Pharmacometrics Syst. Pharmacol. 10: 1208–1220 (2021).
[7] Cunio CB et al. Clin. Microbiol. Infect. 27: 783.e7-783.e14 (2021).
[8] Sheiner LB Biopharm. 9: 503–512 (1981).
[9] Sheiner LB et al. Clin. Pharmacol. Ther. 26: 294–305 (1979).
