Pharmacokinetics of high-dose methotrexate in adults and children
ÅM Johansson (1), N Mayne (2), S Hennig (1), M Perisoglou (2), J Whelan (2), MO Karlsson (1) , JF Standing (1)
(1) Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden; (2) University College London Hospital, London, UK.
Objective: Adults and children with osteosarcoma receive high dose (12 g/m2) intravenous chemotherapy with the dihydrofolate reductase inhibitor methotrexate; daily TDM is performed until concentrations fall below 0.2 µmol/L. The aim of this study was to develop a population pharmacokinetic methotrexate model, and investigate opportunities to improve TDM.
Methods: Patients undergoing treatment following the EURAMOS 1 osteosarcoma protocol were prospectively recruited, and records made of methotrexate dosing, TDM sampling times, and clinical and demographic details. A population pharmacokinetic model was built in NONMEM VI (FOCE INTER), with 1-, 2- and 3-compartment structural models tested. Inter-occasion variability was tested on CL, V1 and F. CL was divided into two components: filtration and secretion [1, 2]. Filtration was scaled to expected GFR [3] multiplied by the fraction unbound (fu) that were assumed to be 0.4 [4], and secretion was estimated. Covariates tested were body weight, body mass index, creatinine clearance, serum creatinine (SeCr), and age. The final model was used to predict time to fall below 0.2 µmol/L using only the first 2 measurements. The prediction error were calculated using predicted minus “observed” where the “observed” time was the estimated time using all samples. Evaluation tools used were basic goodness of fit plots, stochastic simulations and estimations and visual predictive check.
Results: In total 943 plasma concentrations from 46 patients (4-51 yr) on up to 12 occasions were collected. A 2-compartment model described the data well, although a 3-compartment model gave a significantly lower OFV, possibly due to the effect of data being censored by the cessation of TDM after concentrations were less that 0.2 µmol/L. Inter-occasion variability was added to CL, all parameters were scaled with body weight and CL was additionally scaled with age and sex adjusted SeCr [5]. Parameters for a typical 70 kg male >18 yrs were: CLfilt 2.69 L/h/70kg, CLsec 10.9 L/h/70kg, V1 74.3 L/70kg, Q 0.110 L/h/70kg, V2 4.10 L/70kg, proportional residual variability was 29.3%. Using only the first 2 concentrations, the mean (SD) prediction error of time to reach 0.2 µmol/L was 0.34 (8.2) hours. Excluding the occasions with preceding concentrations above 0.8 µmol/L (20.6 %) the variability was reduced (1.88 (3.3) hours).
Conclusions: A 2-compartment model with body weight and SeCr scaling adequately described the data, and predicted the time for concentration to fall below 0.2 µmol/L. This model will now be used for Bayesian forecasting in TDM. Censoring of the data by patients being discharged after levels were below 0.2 µmol/L may affect structural model selection. Methods for handling this type of censoring in pharmacometric data warrant further investigation.
References:
[1] Christophidis N et al. Cancer Chemotherapy and Pharmacology, 1981; 6: p. 59-64.
[2] Winograd B et al. European Journal of Clinical Pharmacology, 1986; 30: p. 231-238.
[3] Maia MB et al. International Journal of Clinical Pharmacology and Therapeutics, 1996. 34: p. 335-341.
[4] Rhodin MM et al. Pediatric Nephrology, 2009. 24: p. 67-76.
[5] Ceriotti F et al. Clinical Chemistry, 2008. 54: p. 559-566.
