Prediction of methotrexate elimination after high dose infusion in children with acute lymphoblastic leukaemia using a population pharmacokinetic approach

Abstract

High-dose methotrexate (HD-MTX) with leucovorin rescue is a component of therapy in children with acute lymphoblastic leukaemia. Since MTX toxicity is related to drug exposure, a monitoring of serum MTX concentrations at H24, H48, H72 and until the concentration is less than 0.2 micromol/L is commonly performed. However, a number of patients may reach concentrations of less than 0.2 micromol/L long before the next sampling is scheduled. The aim of our study was to develop a Bayesian method predicting the time at which MTX concentration reaches 0.2 micromol/L in order to decrease the number of samples drawn and to allow for a more rapid patient discharge. Methotrexate population parameters were estimated from a retrospective analysis of 60 infusions in 23 children and MTX concentrations were predicted from an independent set of 20 courses in 14 children with a Bayesian approach using either one (H48) or two (H24 and H48) samples. The following population parameters were obtained using a two-compartment model: CL = 3.51 L/h (inter-individual variability: 66%), Vd = 8.67 L (58%), k12 = 0.0044 h(-1)(105%), k21 = 0.039 h(-1)(25%). Clearance and Vd were found to increase with weight and age respectively. Both sampling schedules tested for the Bayesian estimation enabled accurate prediction of concentrations and provided satisfactory precision despite a small bias. When considering the ability to predict the time at which the threshold was reached, the one-sample (H48) schedule gave the best results. We conclude that a sampling schedule involving only one sample and Bayesian parameter estimation may be able to predict the delay necessary to reach 0.2 micromol/L in each individual.

Developing Population Pharmacokinetic Parameters for High-Dose Methotrexate Therapy: Implication of Correlations among Developed Parameters for Individual Parameter Estimation Using the Bayesian Least-Squares Method

“…This pair of parameters had been included in most of the previously reported population parameters. 9,[11][12][13][14][15][16][17][18][19] This result indicates that the reported population parameters may be inappropriate for the Bayesian least-squares method used in the present study. Thus, we compared correlations among the possible combinations of population parameters.…”

“…The former was the combination that had the weakest correlations in the present study, and the latter was the most frequently reported combination in previous studies. [9][10][11][12][13][14][15][16][17][18][19] As shown above, the sum of the absolute value of correlation coefficients among V 1 , k 10 , k 12 , and k 21 was 1.05 for regimen B, whereas V 1 , V 2 , CL, and Q was 3.29 for regimen B. Because of the relatively strong correlations among V 1 , V 2 , CL, and Q, this combination of the population parameters was considered to be a good target for a comparison with V 1 , k 10 , k 12 , and k 21 .…”

“…7,8) If Bayesian estimation makes it possible to predict individual parameters using only a few concentrations determined at an early stage after the administration of MTX, it is possible to predict a delay in the excretion of MTX. This estimation will provide a key for physicians to decide the duration of the rescue dosage using calcium folinate, which protects normal cells from the toxicity of MTX.…”

“…Several population parameters have already been developed for cases receiving HD-MTX, and the usefulness of these parameters has been tested. [9][10][11][12][13][14][15][16][17][18][19] Many of these studies were performed using NONMEM, which has been used as the standard method for population pharmacokinetic analyses including Bayesian estimation. However, this software is rarely used in clinical settings because of the difficulties associated with its use.…”

“…In addition, because an increase in the parameters included in the pharmacokinetic models increases the possible correlations among the parameters, this issue should be considered especially for medications in which the pharmacokinetic pro-file is a 2 or higher dimensional compartment model. MTX pharmacokinetics were previously shown to be accounted for by a 2-compartment model, [9][10][11][12][13][14][15][16][17][18][19] which suggests that this issue needs to be considered. However, to the best of our knowledge, correlations and/or covariance among population parameters have not been considered especially in the clinical use of Bayesian estimation performed using the Bayesian least-squares method.…”