The impact of misspecified covariate models on inclusion and omission bias when using fixed effects and full random effects models

The Journal of Pharmacokinetics and Pharmacodynamics has published a new methodological publication authored by Joakim Nyberg and Niclas E. Jonsson and medical writing by Viviana Moroso. The publication touches on a topic relevant to all pharmacometricians and addresses questions that can arise during model development, particularly when scope reduction is applied.

The analysis described explores the impact that covariate model misspecifications may have on parameter estimates. This work considers cases of model misspecification in which relevant covariate-parameters relationships are incorrectly excluded from a model (potentially leading to omission bias) and cases in which non-relevant covariates are included on model parameters (potentially leading to inclusion bias). Variations of these types of model misspecifications are evaluated for covariate analysis using fixed effects models (FEM) and full random effects models (FREM). The results indicate that model parameters were affected by omission bias when FEM models were misspecified, while they remained unaffected by inclusion bias, especially when FREM was used.

Overall, the authors conclude that caution should be used when applying the common recommendations that propose to limit the covariate-parameters relationships to be tested to those that are scientifically plausible. Furthermore, these results suggest that FREM is a reliable method for full model covariate analysis and support its use as a more informative alternative to traditional FEM approaches.