Developing Tools to Evaluate Non-linear Mixed Effect Models: 20 Years on the npde Adventure.

Area of Science:

• Pharmacometrics
• Statistical modeling
• Biostatistics

Background:

• Non-linear mixed effect (NLME) models are widely used in pharmacometrics and other fields for analyzing hierarchical data.
• Developing robust evaluation tools is crucial for assessing the assumptions and adequacy of these complex models.
• Model evaluation guides model building and communicates reliability for specific applications.

Observation:

• Over 20 years, a suite of simulation-based evaluation tools has been developed and refined.
• These tools address the complexities of NLME models, including outcome evolution, inter-outcome links, and parameter variability.
• The focus is on providing informative diagnostics to support model development and assessment.

Findings:

• Normalised prediction discrepancies (npd) and normalised prediction distribution errors (npde) are key developed tools.
• These diagnostics provide graphical and statistical assessments of model performance.
• The tools effectively evaluate various components of NLME models, aiding in their interpretation and validation.

Implications:

• The developed evaluation tools enhance the reliability and interpretability of NLME models.
• These diagnostics facilitate informed decision-making in model building and application.
• This work contributes to best practices in statistical modeling and evaluation for complex hierarchical data.