The Use of Full Random Effects Models in Model-Informed Precision Dosing

Background: Model-informed precision dosing (MIPD) uses mathematical models for a priori and a posteriori dose predictions. One of its challenges is how to handle missing covariate data. The purpose in this study was to compare the performance of Full Random Effects Models (FREM) against a reference fixed effect model (FEM) in handling varying missing covariate scenarios. Methods: Missing covariate data is typically addressed via mean imputation in FEM. Conversely, FREM treats covariates as observed data, allowing the derivation of FEM for any available covariate subset without parameter reestimation. To evaluate this, a reference FEM (REF) and a FREM were estimated using simulated data from an established population pharmacokinetic model across various residual unexplained variability and clearance interindividual variability levels. A priori and a posteriori dosing was evaluated for multiple covariate scenarios. Predictive performance was assessed using mean absolute relative prediction error and mismatch percentage between predicted and true doses. Results: For a priori predictions based on the covariates originally available in REF, performance was mostly comparable between FREM and REF. However, FREM demonstrated a performance advantage for a priori predictions when used to derive FEM that incorporated covariates not originally included in REF. A posteriori performance, which also depends on the available data, across all scenarios was similar between FREM and REF. Conclusions: FREM-derived FEM provide a performance edge for a priori dose predictions by using additional covariates that were not originally included in REF, whereas both methods perform similarly for a posteriori predictions. Overall, FREM seems to offer a more flexible approach for MIPD when covariate availability varies.