Bayesian Workflow for Minimal PBPK Models: Case Study of Dapagliflozin

Area of Science:

• Pharmacokinetics
• Pharmacometrics
• Computational Biology

Background:

• Dapagliflozin, an SGLT2 inhibitor, is approved for type 2 diabetes, heart failure, and chronic kidney disease.
• Numerous clinical trials and population pharmacokinetic models exist for dapagliflozin.
• Integrating existing data into a unified model supports informed drug development.

Purpose of the Study:

• To propose a Bayesian workflow for developing minimal physiologically-based pharmacokinetic (mPBPK) models for dapagliflozin.
• To integrate all available pharmacokinetic (PK) data and models for dapagliflozin.
• To support uncertainty quantification and informed drug development decisions.

Main Methods:

• Systematic collection of published dapagliflozin PK data and models.
• Development of a mPBPK model and estimation of posterior parameter distributions using Bayesian inference.
• Evaluation of three Bayesian modeling packages (NIMBLE, MCSim, Torsten) and prior specifications.
• Model validation using crossover clinical trial data and urinary recovery data, followed by sensitivity analysis.

Main Results:

• 18 studies with PK data and 10 with PK models were identified.
• Posterior parameter distributions were comparable across Bayesian packages, with overlapping 95% credible intervals.
• Torsten demonstrated superior sampling efficiency compared to NIMBLE and MCSim.
• Predicted urinary recovery (2.2%-4.4%) aligned with observed values (0.8%-4.0%).
• Glomerular filtration rate and fraction unbound were key drivers of inter-trial variability in urinary recovery.

Conclusions:

• The proposed Bayesian workflow is flexible and transferable for developing mPBPK models for other drugs.
• This approach effectively integrates diverse PK information for robust uncertainty quantification.
• The workflow facilitates informed decision-making in drug development for dapagliflozin and potentially other mechanistically modeled agents.