Evaluation of uncertainty parameters estimated by different population PK software and methods

Area of Science:

• Pharmacometrics
• Population Modeling
• Statistical Estimation

Background:

• Parameter uncertainty estimation is vital for population model building and evaluation.
• Standard error (SE) estimation, a key measure of uncertainty, can be unreliable.
• Evaluating different estimation methods is necessary for robust population modeling.

Purpose of the Study:

• To compare the performance of various non-linear mixed-effect estimation methods.
• To assess the accuracy of standard error (SE) estimations and parameter estimations across methods.
• To evaluate convergence properties and computation time for different estimation techniques.

Main Methods:

• Evaluated maximum likelihood methods (FO, FOCE, nlme, SAEM) and Bayesian (WinBUGS).
• Used simulated datasets from a one-compartment PK model with 9 designs.
• Applied bootstrap techniques to FO, FOCE, and nlme methods.

Main Results:

• Methods showed concordant SE estimations for fixed effects; SAEM and WinBUGS showed under/over-estimation for random effects.
• FO yielded biased SE and parameter estimations, especially with sparse data.
• SAEM and WinBUGS demonstrated systematic convergence, while FOCE failed in 50% of cases.
• Bootstrap with FOCE was computationally intensive; bootstrap with nlme caused crashes.

Conclusions:

• FO method resulted in biased parameter and random effect SE estimations.
• FOCE provided unbiased results but faced convergence challenges.
• Bootstrap improved SEs for FOCE only when random effect confidence intervals were required.
• WinBUGS offered consistent results but with long computation times.
• SAEM provided unbiased parameter estimates with minor SE under-estimation.