Model-Based Conditional Weighted Residuals Analysis for Structural Model Assessment.

Area of Science:

• Pharmacometrics and Pharmacokinetics
• Statistical Modeling
• Biostatistics

Background:

• Nonlinear mixed effects (NLME) models are crucial for analyzing longitudinal data in drug development and disease research.
• Accurate assessment of NLME model assumptions is vital for reliable conclusions.
• Existing methods may not sufficiently detect structural model misspecifications.

Purpose of the Study:

• To propose and validate a novel method for assessing structural model misspecification in NLME models.
• To evaluate prediction bias using conditional weighted residuals (CWRES).
• To quantify the magnitude of bias in model predictions.

Main Methods:

• Developed a method based on assessing bias in conditional weighted residuals (CWRES).
• Applied the method to two integrated models of glucose homeostasis: the integrated glucose-insulin (IGI) model and the integrated minimal model (IMM).
• Modeled CWRES to identify systematic trends and quantify misspecification magnitude (ΔOFV_Bias).

Main Results:

• The new CWRES bias assessment method correctly identified bias in the integrated minimal model's (IMM) glucose sub-model when misspecification occurred.
• The method accurately calculated the absolute and proportional magnitude of prediction bias.
• CWRES bias trends correlated well with the true patterns of model misspecification.

Conclusions:

• The proposed CWRES bias assessment is an effective and automated diagnostic tool for NLME model development and evaluation.
• This method enhances the reliability of conclusions drawn from longitudinal data analysis.
• The tool is implemented in Perl-speaks-NONMEM, facilitating its practical application.