An automated sampling importance resampling procedure for estimating parameter uncertainty.

Area of Science:

• Pharmacometrics
• Pharmacokinetics and Pharmacodynamics (PK/PD)
• Computational Statistics

Background:

• Accurate quantification of parameter uncertainty is crucial for decision-making in drug development using nonlinear mixed-effects models (NLMEM).
• Existing methods for assessing parameter uncertainty have limitations, with low scrutiny regarding their adequacy.
• Sampling Importance Resampling (SIR) was previously proposed as a fast, assumption-light method, but its settings required careful selection.

Purpose of the Study:

• To develop and validate an automated, iterative Sampling Importance Resampling (SIR) procedure for estimating parameter uncertainty in NLMEM.
• To assess the performance of the automated SIR procedure across a wide range of real-world NLMEM applications.

Main Methods:

• Development of an automated, iterative SIR procedure to overcome limitations of non-iterative implementations.
• Testing the procedure on 25 diverse real data examples of pharmacokinetic and pharmacodynamic NLMEM.
• Comparison of automated SIR with covariance matrix, bootstrap, and stochastic simulations and estimations (SSE) methods.

Main Results:

• The automated SIR procedure achieved appropriate results within an average of 3 iterations.
• SIR demonstrated comparable relative standard errors to the covariance matrix and SSE.
• SIR exhibited similar parameter 95% confidence interval asymmetry to SSE, outperforming bootstrap in speed (approx. 10x faster).

Conclusions:

• The automated SIR procedure is a robust and efficient method for estimating parameter uncertainty in various NLMEM.
• Its user-friendly implementation in the PsN program facilitates practical application in drug development.
• This method enhances the reliability of decision-making by providing accurate uncertainty quantification.