Graphical methods have an advantage over numeric methods for model evaluation because they readily shed light on a broad range of complex aspects of the relationship between the model and the data. In this tutorial, we will only focus on the model evaluation step and describe various tools for evaluation of a NLMEM, regardless of whether it is an intermediary model, a key model in the model‐building step, or a best model that can be used for further inferences.Īlthough there are many statistical tools for model evaluation, the primary tool for most biomedical science and engineering modeling applications is graphical analysis. “Model evaluation” is required for both processes to diagnose one or several intermediary or key models in a model‐building step or evaluate a selected model with respect to the modeling objectives.
“Model qualification” is the assessment of the performance of a model in fulfilling the analysis objectives. “Model building” is the process of developing a model on a given dataset to achieve clearly defined analysis objectives. In this tutorial, we refer to this procedure as “model evaluation,” although, in literature, it has been described under several more or less equivalent terms, such as “model diagnostics,” “model adequacy,” “model assessment,” “model checking,” “model appropriateness,” and “model validation.” Model evaluation has to be clearly distinguished from “model building” and “model qualification” processes, which are two steps of model development that require model evaluation but imply different concepts. In any regression modeling, after fitting a model to a dataset, it is essential to assess the goodness‐of‐fit between the model and the dataset and to determine whether the underlying model assumptions seem appropriate. NLMEMs are widely applied because of their ability to quantify several levels of variability, to handle unbalanced data, and to identify individual specific covariates. They describe both population and subject‐specific characteristics, represented as fixed parameters for population characteristics and random parameters for subjects. 1, 2, 3 NLMEM for continuous PK/PD data use nonlinear dynamic models that draw on physiological or pharmacological principles to provide a reasonable approximation of the dynamics of the drug in the body and of their effects. Nonlinear mixed effects models (NLMEMs) have been established as the state of the art methodology in pharmacometrics for analysis of longitudinal pharmacokinetic (PK) and pharmacodynamic (PD) measurements collected in preclinical and clinical studies, especially in drug development.