Model-informed precision dosing of protein kinase inhibitors: benefits and limits. The example of imatinib.
Félicien Le Louedec (1), Nicolas Boespflug (1), Thierry Lafont (1), Fabienne Thomas (1), Florent Puisset (1), Étienne Chatelut (1)
(1) Institut Claudius Regaud, Institut Universitaire du Cancer de Toulouse Oncopole, Centre de Recherche en Cancérologie de Toulouse, INSERM U1037, Université Paul Sabatier, Toulouse (France)
Objectives: In model-informed precision dosing (MIPD), therapeutic drug monitoring (TDM) concentration data are analyzed by means of a population pharmacokinetic (PK) model to derive maximum a posteriori Bayesian estimates (MAPBE) of parameters and to calculate optimal dosing. Since this approach has proved effective for anti-infectious drugs like vancomycin, there is a growing interest in applying MIPD to oral protein kinase inhibitors in oncology. For imatinib, the level of TDM evidence is high, since efficacy is related to trough concentrations at steady state (Cmin,ss > 1.0 µg/mL in hematological malignancies). MIPD could thus be useful in order to: 1) estimate Cmin,ss in case of inadequate sampling time, 2) propose an alternative dosing regimen based on a posteriori simulations.
Methods: Concentration data collected during TDM of imatinib were retrospectively analyzed with the R package mapbayr [3]. One-hundred and fifty patients provided at least two samples, 70 at least three, and 31 four samples. Five PK models [4–8] with various structures were coded. In addition, a model-averaging procedure based on maximum likelihood was applied. For each situation, four levels of IIV flattening were arbitrarily applied: 1, 0.3, 0.1, and 0.03 (i.e., 33-fold IIV increase). Either one, two or three concentrations were fitted with each model (MAPBE of PK parameters), and the capacity to fit the data, as well as to predict further concentrations (i.e., the second, third or fourth concentration, respectively) was quantified thanks to mean prediction error (MPE) and root-mean-square error (RMSE).
Results: When one concentration was analyzed, the goodness of fit (GOF) varied across models (RMSE range [0.197; 0.524 µg/mL], MPE range [-0.203; +0.143]), all being superseded by the model-averaging procedure (RMSE = 0.159, MPE = -0.078). Predicted Cmin,ss varied by 1.19- to 11-fold across models and patients. Prediction of the following observation was poor regardless of the model (RMSE [0.537; 0.681]), including the model-averaging procedure (RMSE = 0.591). Flattening IIV priors with the weight of 0.03 improved GOF proportionally (RMSE [0.019; 0.204], MPE [-0.151; 0.011]), but still gave various Cmin,ss (from 1.12- to 11-fold), and worsened predictions of the next concentration (RMSE [0.551; 0.694]). When parameters were estimated from two or three concentrations, GOF worsened, and models yielded more homogenous results among themselves (RMSE [0.307; 0.439] with two observations, RMSE [0.315; 0.332] with three observations). The ability to predict the next concentration slightly improved with the number of observations, but all RMSE > 0.500 µg/mL. Notably, with multiple concentrations to fit, the effects of model averaging and IIV prior flattening decreased and yielded GOF and future predictions similar to the original models. However, in terms of bias, results were not as homogeneous between models. Delbaldo et al.’s model yielded the least biased results for GOF (MPE = 0.024 µg/mL) and future concentration prediction (MPE = 0.003 µg/mL), while other models performed at a level around MPE = -0.1 µg/mL.
Conclusions: Our work shows that the choice of the population PK model affects individualized imatinib dosing. Prediction of the next concentration was poor with all methods, with only Delbaldo et al. yielding the least biased results. Flattening IIV priors did not improve the quality of future prediction despite a better GOF of current concentrations. As compared to vancomycin, for which MIPD approaches were successful and with renal elimination that can be predicted with the glomerular filtration rate, imatinib is metabolized by CYP3A4, resulting in less predictable PK variability. Secondly, the interval of time between imatinib samplings is in weeks, while anti-infectious drug concentrations are monitored by days. Moreover, since imatinib is given orally to outpatients, adherence must be considered as a source of variability. Lastly, model-averaging might perform poorly because the fit of a single concentration at steady state is less dependent on the structure of the model than are vancomycin peak-and-trough concentrations. Overall, our results indicate that non-validated model-based approaches are prone to erroneous results when applied in a clinical setting.
References:
[1] Hughes JH, Keizer RJ. A hybrid machine learning/pharmacokinetic approach outperforms maximum a posteriori Bayesian estimation by selectively flattening model priors. CPT Pharmacometrics Syst Pharmacol. 2021;10:1150–60.
[2] Uster DW, Stocker SL, Carland JE, Brett J, Marriott DJE, Day RO, et al. A Model Averaging/Selection Approach Improves the Predictive Performance of Model-Informed Precision Dosing: Vancomycin as a Case Study. Clinical Pharmacology & Therapeutics. 2021;109:175–83.
[3] Le Louedec F, Puisset F, Thomas F, Chatelut É, White-Koning M. Easy and reliable maximum a posteriori Bayesian estimation of pharmacokinetic parameters with the open-source R package mapbayr. CPT: Pharmacometrics & Systems Pharmacology. 2021;10:1208–20.
[4] Delbaldo C, Chatelut E, Ré M, Deroussent A, Séronie-Vivien S, Jambu A, et al. Pharmacokinetic-pharmacodynamic relationships of imatinib and its main metabolite in patients with advanced gastrointestinal stromal tumors. Clin Cancer Res. 2006;12:6073–8.
[5] Di Paolo A, Polillo M, Capecchi M, Cervetti G, Baratè C, Angelini S, et al. The c.480C>G polymorphism of hOCT1 influences imatinib clearance in patients affected by chronic myeloid leukemia. Pharmacogenomics J. 2014;14:328–35.
[6] Menon-Andersen D, Mondick JT, Jayaraman B, Thompson PA, Blaney SM, Bernstein M, et al. Population pharmacokinetics of imatinib mesylate and its metabolite in children and young adults. Cancer Chemother Pharmacol. 2009;63:229–38.
[7] Widmer N, Decosterd LA, Csajka C, Leyvraz S, Duchosal MA, Rosselet A, et al. Population pharmacokinetics of imatinib and the role of alpha-acid glycoprotein. Br J Clin Pharmacol. 2006;62:97–112.
[8] Yamakawa Y, Hamada A, Nakashima R, Yuki M, Hirayama C, Kawaguchi T, et al. Association of Genetic Polymorphisms in the Influx Transporter SLCO1B3 and the Efflux Transporter ABCB1 With Imatinib Pharmacokinetics in Patients With Chronic Myeloid Leukemia. Therapeutic Drug Monitoring. 2011;33:244.
