Anna Chan Kwong (1,2), Amaury O’Jeanson (1), Patrick Nolain (3), Florence Gattacceca (2) and Sonia Khier (1)
(1) Montpellier University, IMAG (Alexander Grothendieck Institute), (2) Aix-Marseille University, INSERM, CNRS, CRCM SMARTc, (3) University of Limoges
Objectives: Population pharmacokinetic (popPK) models can support the optimization of dosing regimens in therapeutic drug monitoring (TDM): PK parameters of each patient are estimated with their first concentration(s) to predict further individual concentrations (also called Bayesian feedback TDM). For most of the drugs measured in TDM, popPK models are reported in literature. It may be of interest to adapt the literature models to the target population with the “prior approach” [1] (tweaked models). We compared the predictive ability of both literature and tweaked models on TDM concentrations of meropenem.
Methods: Blood samples of meropenem were collected from 2017 to 2019 in patients of the intensive care unit of the university health center of Montpellier (France). The study protocol was approved by the Ethics Committee (2019_IRB-MTP_03-01). The total dataset was split into an “estimation” and a “prediction” dataset. The “estimation” dataset was composed of the patients having only one reported meropenem concentration (mono-point), as they were not eligible for predicting further concentrations, and half the patients having more than one reported meropenem concentration (multi-points). The remaining multi-points patients composed the “prediction” dataset. As in a cross-validation approach, multi-points patients were randomly split six times (stratified on dialysis) into the “estimation” and the “prediction” dataset.
PopPK models for meropenem were selected from literature. These models were run on the “estimation” dataset with the $PRIOR NWPRI subroutine in NONMEM. Then, the initial estimates of the control file were updated to their estimated values. Prior information was removed from this “tweaked model” control file and MAXEVAL=0 was set in $ESTIMATION to perform Bayesian predictions of the “prediction dataset”, based on the first observation of each patient (further observations were set at MDV=1). Meanwhile, the literature models were also run with MAXEVAL=0 directly on the “prediction dataset”. This procedure was repeated for each of the six splits. Differences between observation and prediction (prediction errors) were computed for both tweaked and literature models: the results were summarized using relative mean prediction error (MPE(%)) and root mean square error (RMSE(%)), reflecting respectively the bias and the imprecision as a percentage of the mean of the observations.
Results: The total dataset was composed of 115 concentrations from 58 patients (31 mono-point and 27 multi-points with two to seven observed values). For each of the six splits, the “estimation” and the “prediction” datasets were respectively composed of 44 and 14 patients or 45 and 13 patients. Six popPK models were selected in literature [2–7]. MPE(%) and RMSE(%) were smaller with tweaked than with literature models in 61% and 75% of the cases, respectively. On average, MPE(%) and RMSE(%) were respectively 0.8% and 1.8% lower with tweaked than with literature models.
Conclusions: For these sparse data from clinical practice, there was a tendency for tweaked models to better predict individual concentrations than literature models. Thus, the “prior approach” could be a valuable tool to improve the predictive ability of literature models. Sharing model codes would facilitate the use of published models and the implementation of innovative methods to improve Bayesian feedback TDM.
References:
[1] Gisleskog PO, Karlsson MO, Beal SL (2002) Use of Prior Information to Stabilize a Population Data Analysis. J Pharmacokinet Pharmacodyn 29:473–505 . https://doi.org/10.1023/A:1022972420004
[2] Dhaese SAM, Farkas A, Colin P, Lipman J, Stove V, Verstraete AG, Roberts JA, De Waele JJ (2019) Population pharmacokinetics and evaluation of the predictive performance of pharmacokinetic models in critically ill patients receiving continuous infusion meropenem: a comparison of eight pharmacokinetic models. J Antimicrob Chemother 74:432–441 . https://doi.org/10.1093/jac/dky434
[3] Ulldemolins M, Soy D, Llaurado-Serra M, Vaquer S, Castro P, Rodríguez AH, Pontes C, Calvo G, Torres A, Martín-Loeches I (2015) Meropenem population pharmacokinetics in critically ill patients with septic shock and continuous renal replacement therapy: influence of residual diuresis on dose requirements. Antimicrob Agents Chemother 59:5520–5528 . https://doi.org/10.1128/AAC.00712-15
[4] Li C, Kuti JL, Nightingale CH, Nicolau DP (2006) Population pharmacokinetic analysis and dosing regimen optimization of meropenem in adult patients. J Clin Pharmacol 46:1171–1178 . https://doi.org/10.1177/0091270006291035
[5] Mattioli F, Fucile C, Del Bono V, Marini V, Parisini A, Molin A, Zuccoli ML, Milano G, Danesi R, Marchese A, Polillo M, Viscoli C, Pelosi P, Martelli A, Di Paolo A (2016) Population pharmacokinetics and probability of target attainment of meropenem in critically ill patients. Eur J Clin Pharmacol 72:839–848 . https://doi.org/10.1007/s00228-016-2053-x
[6] Roberts JA, Kirkpatrick CMJ, Roberts MS, Robertson TA, Dalley AJ, Lipman J (2009) Meropenem dosing in critically ill patients with sepsis and without renal dysfunction: intermittent bolus versus continuous administration? Monte Carlo dosing simulations and subcutaneous tissue distribution. J Antimicrob Chemother 64:142–150 . https://doi.org/10.1093/jac/dkp139
[7] Jaruratanasirikul S, Thengyai S, Wongpoowarak W, Wattanavijitkul T, Tangkitwanitjaroen K, Sukarnjanaset W, Jullangkoon M, Samaeng M (2015) Population pharmacokinetics and Monte Carlo dosing simulations of meropenem during the early phase of severe sepsis and septic shock in critically ill patients in intensive care units. Antimicrob Agents Chemother 59:2995–3001 . https://doi.org/10.1128/AAC.04166-14
Reference: PAGE () Abstr 9581 [www.page-meeting.org/?abstract=9581]
Poster: Clinical Applications