Development of visual predictive checks accounting for multimodal parameter distributions in mixture models.

Area of Science:

• Pharmacometrics
• Statistical Modeling
• Population Pharmacokinetics

Background:

• Nonlinear mixed effects models (NLME) typically assume unimodal interindividual variability, which is insufficient for populations with multimodal parameter distributions.
• Mixture models are essential for identifying and characterizing subpopulations within complex datasets.
• Standard visual predictive checks (VPCs) are not designed to evaluate mixture models effectively.

Purpose of the Study:

• To develop and validate mixture model-specific visual predictive checks (VPCs) for populations with multimodal parameter distributions.
• To compare different strategies for data splitting in mixture model VPCs.
• To enhance the diagnostic power for evaluating complex population pharmacokinetic models.

Main Methods:

• Developed mixture model-specific VPCs using two data-splitting strategies: MIXEST (most likely subpopulation) and IPmix (individual probability of subpopulation membership).
• Evaluated strategies using simulated data, identifying limitations of MIXEST-based allocation.
• Applied the refined IPmix-based VPC strategy to an irinotecan pharmacokinetic model.

Main Results:

• The IPmix-based data splitting strategy avoided the bias observed with MIXEST, ensuring accurate subpopulation representation.
• Mixture model-specific VPCs successfully identified model misspecifications not apparent with standard VPCs.
• Application to the irinotecan model highlighted a significant difference in SN-38 clearance between UGT1A1 genotypes.

Conclusions:

• Mixture model-specific VPCs, particularly those using IPmix assignment, offer enhanced diagnostic capabilities for complex population pharmacokinetic models.
• This approach improves the evaluation of mixture models, leading to more robust pharmacokinetic and pharmacodynamic analyses.
• The developed tool provides a powerful method for identifying and characterizing subpopulations and their unique parameter distributions.