Model-Informed Precision Dosing of Protein Kinase Inhibitors: Evaluation of Model-Averaging and Flattened Priors Methods.

BACKGROUND AND

OBJECTIVES

Therapeutic drug monitoring of protein kinase inhibitors (PKIs) usually relies on the measure of a single trough concentration at steady-state (Cmin,ss). When the sampling time differs from the trough, it is theoretically possible to predict Cmin,ss from maximum a posteriori (MAP) Bayesian estimates of PK parameters. However, several questions remain with regards to model-informed precision dosing (MIPD) of PKIs, such as choosing which model to use when several are available in the literature. Alternative techniques, such as flattened priors and model averaging may outperform standard analyses. The aim of this work is to report a comprehensive fit-for-purpose validation of MIPD for sunitinib and pazopanib.

METHODS

Concentration data from 41 renal cancer patients included in the SUP-R trial (NCT02555748) measured 2 and 6 hours after an intake were analyzed (MAP-Bayesian estimation of PK parameters) in order to predict Cmin,ss at the current cycle and at the next cycle. Different models from the literature were tested, as well as the model-averaging and flattened priors features available in the R package 'mapbayr'.

RESULTS

The quality of Cmin,ss predictions depended on the model used. Flattening priors rarely improved or worsened the predictions. Model averaging was robust across the different scenarios tested and should be preferred to using a single model. Overall, a precision of 20% to 25% was achieved, with a minimal bias (< 5%).

CONCLUSION

The benefit of the model-averaging method for the model-informed precision dosing of sunitinib and pazopanib is likely applicable to other protein kinase inhibitors. Thanks to 'mapbayr', this framework was implemented as a standalone shiny application to be used in clinical settings.