The Visual Predictive Check and Real-World Data.

Area of Science:

• Pharmacometrics
• Clinical Pharmacology
• Statistical Modeling

Background:

• Visual predictive checks (VPCs) are standard for pharmacometric model evaluation.
• Real-world data often involves adaptive clinical decision-making.
• This adaptation can lead to apparent model misspecification in VPCs.

Purpose of the Study:

• To investigate how real-world data characteristics, specifically adaptive therapy, impact VPC interpretation.
• To evaluate the performance of VPCs and prediction-corrected VPCs (pcVPCs) under these conditions.
• To propose benchmarks for assessing model diagnostic robustness.

Main Methods:

• Simulated four common real-world therapy adaptation scenarios.
• Generated simulated observations using an unbiased model.
• Assessed model suitability using both VPC and pcVPC.

Main Results:

• Prediction-corrected VPC (pcVPC) correctly identified a well-specified model when only dose amount varied.
• Both VPC and pcVPC erroneously indicated model misspecification when dosing interval, sample timing, or sample frequency varied.
• Adaptive sampling and dosing strategies can confound model evaluation.

Conclusions:

• VPCs can be misleading with real-world data when therapy is adapted based on patient data.
• Dosing interval, sample timing, and sample frequency variations in response to concentrations can falsely suggest model misspecification.
• The proposed scenarios can serve as benchmarks for validating model diagnostic tools.