Validating Physiologically-Based Pharmacokinetic Models Using the Continuous Ranked Probability Score: Beyond Being...

Area of Science:

• Pharmacokinetics and Pharmacodynamics
• Computational Biology and Bioinformatics
• Drug Development and Pharmacology

Background:

• Physiologically-based pharmacokinetic (PBPK) models are vital in model-informed drug development (MIDD).
• Existing model evaluation guidelines lack specific quantification methods for validation.
• There is a need for precise metrics to assess PBPK model predictive accuracy.

Purpose of the Study:

• To propose a rigorous validation approach for PBPK models using the Continuous Ranked Probability Score (CRPS).
• To demonstrate the applicability of CRPS for both individual and population-level predictions in PBPK modeling.
• To introduce a skill-score approach for single PBPK model validation.

Main Methods:

• Application of the Continuous Ranked Probability Score (CRPS) for PBPK model validation.
• Comparison of two PBPK models using the proposed CRPS-based validation technique.
• Development of an accessible online tool to implement the CRPS validation method.

Main Results:

• The CRPS effectively quantifies how well PBPK models recreate observed data distributions.
• CRPS allows for the comparison of predictive performance between competing PBPK models.
• The skill-score approach using CRPS facilitates single model validation.

Conclusions:

• The proposed CRPS-based validation method offers a more thorough and quantitative assessment of PBPK models.
• This metric is versatile, applicable to individual and population predictions, and comparable models.
• The CRPS metric and associated tool can enhance the reliability of PBPK models in drug development and other modeling applications.