Longitudinal aggregate data model-based meta-analysis with NONMEM: approaches to handling within treatment arm...

Area of Science:

• Pharmacometrics
• Statistical modeling
• Meta-analysis

Background:

• Longitudinal data in clinical studies exhibit complex correlation structures.
• Accurate meta-analysis requires appropriate statistical models to handle these correlations.

Purpose of the Study:

• To evaluate multi-level random effects models in NONMEM for longitudinal meta-analysis.
• To compare different strategies for implementing random effects and residual correlation.

Main Methods:

• Simulated linear and non-linear models with varying random effects structures.
• Compared estimation models including study and/or treatment arm-level random effects.
• Assessed model performance with and without residual correlation.

Main Results:

• Fixed random effects parameters were accurately estimated across all strategies.
• Models incorporating within-study and treatment arm correlation best characterized variability in some scenarios.
• Models with only study-level correlation outperformed others in specific situations.
• Models with only treatment arm-level random effects were not superior.

Conclusions:

• Multi-level random effects models are essential for complex correlation structures in longitudinal meta-analysis.
• The choice of random effects structure impacts variability characterization.
• NONMEM can implement these advanced models for robust analysis.