Use of normalized prediction distribution errors for assessing population physiologically-based pharmacokinetic model...

Area of Science:

• Pharmacometrics
• Pharmacokinetics
• Computational Biology

Background:

• Existing methods for qualifying population physiologically-based pharmacokinetic (Pop-PBPK) model predictions of continuous outcomes are limited.
• These limitations include failure to account for within-subject correlations and residual error, crucial for accurate model evaluation.
• There is a need for improved validation metrics that address these complexities in Pop-PBPK modeling.

Purpose of the Study:

• To propose and describe a novel method for evaluating Pop-PBPK model predictions.
• The new method specifically accounts for within-subject correlations and residual error.
• To demonstrate the utility of the proposed evaluation approach using simulation studies and real-world clinical data.

Main Methods:

• Derivation of Pop-PBPK-specific normalized prediction distribution errors (NPDE) for model validation.
• Definition of three key measures for evaluating model performance: mean NPDE, goodness-of-fit plots, and residual error magnitude.
• Application of the NPDE-based method to positive and negative control simulation studies and a clindamycin Pop-PBPK model in children.

Main Results:

• The NPDE-based approach successfully identified congruency in the positive-control simulation (mean NPDE = -0.01).
• Incongruency between model and data was correctly identified in the negative-control simulation (mean NPDE = -0.29).
• Evaluation of a clindamycin Pop-PBPK model in children showed successful predictions (mean NPDE = 0), with potential biases noted in pediatric subpopulations.

Conclusions:

• The proposed NPDE-based method offers a robust approach for validating Pop-PBPK models, addressing limitations of current techniques.
• The method effectively distinguishes between congruent and incongruent model predictions, as shown in simulation studies.
• The approach is valuable for real-world applications, aiding in the assessment of model performance and identification of areas for improvement, particularly in specific subpopulations.