External evaluation and systematic review of population pharmacokinetic models for high-dose methotrexate in cancer patients

Several population pharmacokinetic (PPK) models have been established to optimize the therapeutic regimen and reduce the toxicity of high-dose methotrexate (HDMTX) in patients with cancer. However, their predictive performance when extrapolated to different clinical centers was unknown. In this study, we aimed to externally evaluate the predictive ability of HDMTX PPK models and determine the potential influencing factors. We searched the literature and determined the predictive performance of the selected models using methotrexate concentrations in 721 samples from 60 patients in the First Affiliated Hospital of the Navy Medical University. Prediction-based diagnostics and simulation-based normalized prediction distribution errors (NPDE) were used to evaluate the predictive performance of the models. The influence of prior information was also assessed using Bayesian forecasting, and the potential factors affecting model predictability were investigated. Thirty models extracted from published PPK studies were assessed. Prediction-based diagnostics showed that the number of compartments potentially influenced model transferability, and simulation-based NPDE indicated model misspecification. Bayesian forecasting significantly improved the predictive performance of the models. Various factors, including bioassays, covariates, and population diagnosis, influence model extrapolation. The published models were unsatisfactory for all prediction-based diagnostics, except for the 24 h methotrexate concentration monitoring and simulation-based diagnostics, making them inappropriate for direct extrapolation. Moreover, Bayesian forecasting combined therapeutic drug monitoring could improve the predictive performance of the models.