Abstract
Despite the well-demonstrated efficacy of infliximab in inflammatory diseases, treatment failure remains frequent. Dose adjustment using Bayesian methods has shown in silico its interest in achieving target plasma concentrations. However, most of the published models have not been fully validated in accordance with the recommendations. This study aimed to submit these models to an external evaluation and verify their predictive capabilities. Eight models were selected for external evaluation, carried out on an independent database (409 concentrations from 157 patients). Each model was evaluated based on the following parameters: goodness-of-fit (comparison of predictions to observations), residual error model (population weighted residuals (PWRES), individual weighted residuals (IWRES), and normalized prediction distribution errors (NPDE)), and predictive performances (prediction-corrected visual predictive checks (pcVPC) and Bayesian simulations). The performances observed during this external evaluation varied greatly from one model to another. The eight evaluated models showed a significant bias in population predictions (from −7.19 to 7.38 mg/L). Individual predictions showed acceptable bias and precision for six of the eight models (mean error of −0.74 to −0.29 mg/L and mean percent error of −16.6 to −0.4%). Analysis of NPDE and pcVPC confirmed these results and revealed a problem with the inclusion of several covariates (weight, concomitant immunomodulatory treatment, presence of anti-drug antibodies). This external evaluation showed satisfactory results for some models, notably models A and B, and highlighted several prospects for improving the pharmacokinetic models of infliximab for clinical-biological application.