What is the between cycle variability in methotrexate clearance?


High dose methotrexate (MTX) is the mainstay of treatment for many cancers. It is typically administered on repeated occasions known as ‘cycles’.  Due to extensive variability in its pharmacokinetics (PK) and life-threatening toxicity, the use of repeated MTX concentration measurements to determine the duration of  leucovorin dosing has decreased the incidence of severe toxicity.  Individualization of future doses based on a target area under the curve (AUC) predicted from Bayesian estimation of PK parameters has been shown to improve survival (1) but it not widely used.  The usefulness of measuring concentrations during one cycle to predict the dose needed for subsequent cycles depends on the unpredictable between occasion variability (BOV) in PK. BOV determines the size of the unpredictable variation in PK from one cycle to the next.  In three previous reports, only the BOV for CL was estimated and in one BOV for V1 was also reported but without correlation with CL. The estimates are 14.6%, 13.3%, 13.3%, and 16.5% whereas the estimate of BOV for CL from our study is 56%. In our model, we included BOV on all key PK parameters. In this study, we have investigated the accuracy of BOV estimates without estimating BOV on the 4 key parameters of a 2 compartment model for MTX and without estimating the covariance when BOV has been estimated on all parameters.


Based on real data obtained from 56 children with cancer at the Children’s Hospital of Philadelphia, two models were compared and selected for the simulation study.  The two models are a model including BSV and BOV and their covariance (Full model) and a model including BOV only on CL and without any covariance (BOVCL model). Compared to the BOVCL model, the full model performed better and was selected for a simulation study. Details of the full PK model are described elsewhere (2). The full model MTX parameters including BSV and BOV and their correlations was used to simulate 100 data sets using the same covariates and sampling times as in the original data. A parametric bootstrap was performed with each of these data sets by fitting to the full model to two simpler models. The first of these estimated BOV for CL, V1, Q and V2 but without covariance between BSV or BOV parameters (model BOV4). The second did not include covariance between the BSV parameters and only estimated BOVCL (model BOVCL). In each case the bias was calculated and was expressed as a percentage of the final model true value or variance. Parameter estimation was performed using NONMEM 7.2 (Icon PLC) using the first-order conditional estimation method with interaction (NSIG=3; SIGL=9).


Using data simulated from the full model, the estimate of BOV for CL was accurately estimated (FULL=2% smaller variance than true value used for simulation) but seriously underestimated when BOV in the other PK parameters was ignored (BOVCL=91% smaller variance)  or if covariance in BOV was ignored (BOV4=71% smaller variance).  The residual error parameter was estimated with negligible bias with the full model but was overestimated with the simpler BOV4 (49% smaller variance) and BOVCL (75% smaller variance) models. When the BOV4 model was applied to the original data set the BOV in CL was estimated to be 16%.


We conclude that literature estimates of BOV of CL are possibly substantial underestimates of the true BOV. In the data set presented here the BOV of CL is much larger than those reported in the literature. Those values overestimate the usefulness of estimating CL during one cycle of MTX in order to predict the dose needed to achieve a target exposure in subsequent cycles.

(1)          Evans, W.E., Relling, M.V., Rodman, J.H., Crom, W.R., Boyett, J.M. & Pui, C.H. Conventional compared with individualized chemotherapy for childhood acute lymphoblastic leukemia. N Engl J Med  338, 499-505 (1998).

(2)          Hirankarn, S., Holford, N.H.G., Dombrowsky, E., Patel, D. & Barrett, J.S. Pharmacokinetics of high-dose methotrexate in children with cancer: A mechanism-based evaluation of clearance prediction.  PAGANZ (2012).