Evaluation of bias, precision, robustness and runtime for estimation methods in NONMEM 7

Area of Science:

• Pharmacometrics
• Computational Statistics
• Drug Development

Background:

• NONMEM is the leading software for population pharmacokinetic (PK)-pharmacodynamic (PD) modeling.
• NONMEM 7 (NM7) introduced advanced sampling-based estimation methods alongside traditional ones.

Purpose of the Study:

• To evaluate the performance of NM7 estimation methods regarding bias, precision, robustness, and runtime.
• To compare these methods across various population PK-PD models.

Main Methods:

• Simulated 500 datasets per model for bias and precision analysis.
• Simulated 100 datasets for robustness testing using true and random initial parameters.
• Measured runtime for each estimation method and model.

Main Results:

• Importance sampling demonstrated the lowest bias and highest precision, followed by FOCE/LAPLACE and SAEM.
• FOCE/LAPLACE offered the shortest runtimes across all models.
• Robustness varied by model, with no single method consistently superior.
• Bayesian Markov Chain Monte Carlo (MCMC) performed poorly on all tested metrics.

Conclusions:

• Importance sampling is recommended for optimal bias and precision in NM7 PK-PD analyses.
• FOCE/LAPLACE is the most efficient method in terms of runtime.
• Method selection should consider the specific PK-PD model and desired performance metrics (bias, precision, robustness, runtime).