All models use both between subject variability (BSV) and between occasion variability (BOV) for Bayesian estimation of the individual parameters. An occasion is defined by the dosing interval associated with any dose which has response observations (e.g. medicine concentrations) measured before the next dose.
Two methods are available for choosing how to predict the dose using these sources of variability. The first method uses both BSV and BOV to predict the target dose. The second method uses just BSV. This is equivalent to averaging the BOV random effects. The second method may give a more stable dose prediction but any real changes that might have occurred from occasion to occasion are not shown. A simulation study has shown the potential benefits of using the averaging method (Abrantes, Jönsson et al. 2019).
It is your choice to decide which method to use depending on what you know about the patient status and how it might have changed from occasion to occasion. Being able to see the dose prediction changes with dose occasion may help to identify non-random influences such as drug interactions which are not part of the model.
Some models, e.g. for warfarin, do not have BOV affecting the parameters used to predict the target dose so the dose predictions are the same with both methods.
When BOV is used for parameter estimation then some models are provided that use just BSV for predicting the Bayesian dose. NextDose models whose name does not include “_AVG” use the first method (both BSV and BOV to predict doses). Models with “_AVG” in the name use the second method (only BSV to predict doses).
Copyright All rights reserved | Developed by Sam Holford & Nick Holford 2012-2019
Abrantes, J. A., S. Jönsson, M. O. Karlsson and E. I. Nielsen (2019). "Handling interoccasion variability in model-based dose individualization using therapeutic drug monitoring data." British Journal of Clinical Pharmacology Accepted doi:10.1111/bcp.13901.