The Reference-Corrected Visual Predictive Check: A More Intuitive Diagnostic for Non-Linear Mixed Effects Models
About the publication
The prediction-corrected visual predictive check (pcVPC) is a commonly used model diagnostic, especially in the context of heterogeneous study designs or adaptive dosing. While useful, the transformation applied in prediction correction can result in plots that are unintuitive and difficult to interpret — even for experienced modelers.
This publication presents the reference-corrected visual predictive check as an alternative approach. Developed by Martin Bergstrand, Niclas Jonsson, and Moustafa M. A. Ibrahim (Previously Pharmetheus, currently Alexion, AstraZeneca Rare Disease), rcVPC addresses key limitations of pcVPCs by leveraging a user-defined reference set of independent variables to normalize observed and simulated data. This normalization supports clearer y-axis interpretation and enhances the communicability of results.
A distinguishing feature of rcVPC is the ability to manipulate the time variable in the reference dataset, enabling visualization of steady state exposure–response relationships in the presence of delayed effects — a challenge for standard VPC and pcVPC methods.
In this publication, the methodology is illustrated through examples inspired by real case-studies with comparisons to traditional VPCs and pcVPCs. The results highlight how rcVPC can improve the interpretability of model diagnostics and support more effective communication during model development.
Acknowledgements
The authors thank Rikard Nordgren, the Dept. of Pharmacy at Uppsala University, for implementing rcVPC in PsN, and to Annika Eklund for medical writing support in accordance with Good Publication Practice guidelines (http://www.ismpp.org/gpp3)
