The objective of this work was to facilitate the\ndevelopment of nonlinear mixed effects models by establishing\na diagnostic method for evaluation of stochastic\nmodel components. The random effects investigated were\nbetween subject, between occasion and residual variability.\nThe method was based on a first-order conditional estimates\nlinear approximation and evaluated on three real\ndatasets with previously developed population pharmacokinetic\nmodels. The results were assessed based on the\nagreement in difference in objective function value\nbetween a basic model and extended models for the standard\nnonlinear and linearized approach respectively. The\nlinearization was found to accurately identify significant\nextensions of the modelââ?¬â?¢s stochastic components with\nnotably decreased runtimes as compared to the standard\nnonlinear analysis. The observed gain in runtimes varied\nbetween four to more than 50-fold and the largest gains\nwere seen for models with originally long runtimes. This\nmethod may be especially useful as a screening tool to\ndetect correlations between random effects since it substantially\nquickens the estimation of large varianceââ?¬â??\ncovariance blocks. To expedite the application of this\ndiagnostic tool, the linearization procedure has been automated\nand implemented in the software package PsN.
Loading....