Multivariate Exact Discrepancy: A New Tool for PK/PD Model Evaluation.

Area of Science:

• Pharmacometrics
• Pharmacokinetic modeling
• Statistical analysis

Background:

• Pharmacokinetic (PK) models are crucial for drug development and therapeutic drug monitoring.
• Current evaluation metrics, such as Visual Predictive Check (VPC) and Normalised Prediction Distribution Error (NPDE), have limitations, particularly with dependent concentration data from the same patient.
• There is a need for improved methods to accurately assess PK model performance in complex scenarios.

Purpose of the Study:

• To propose a novel evaluation method for pharmacokinetic models that explicitly accounts for the dependency between concentration measurements within an individual.
• To provide a robust approach for assessing model fit and identifying potential issues in PK model development.
• To facilitate the selection of appropriate models for therapeutic drug monitoring.

Main Methods:

• The proposed method analyzes the distribution of simulated concentration vectors to generate an individual probability score.
• This score reflects the likelihood that an individual's observed concentrations originate from the studied PK model.
• Aggregate analysis of these individual probabilities allows for a comprehensive evaluation of the overall model performance.

Main Results:

• The method's effectiveness was demonstrated through two illustrative examples.
• The approach successfully identified structural model misspecifications.
• Outlier kinetics within datasets were effectively detected, highlighting the method's sensitivity.

Conclusions:

• A straightforward and effective method for evaluating PK models during development has been presented.
• The method shows promise for selecting optimal models for therapeutic drug monitoring.
• Further validation on a larger scale is recommended to confirm the method's broad applicability and robustness.