Estimating inestimable standard errors in population pharmacokinetic studies: the bootstrap with Winsorization

Area of Science:

• Pharmacometrics
• Computational Biology
• Statistical Modeling

Background:

• Population pharmacokinetic (PopPK) analysis is crucial for drug development.
• Small sample sizes in early phase I studies often lead to inestimable standard errors.
• The NONMEM software is widely used for PopPK modeling.

Purpose of the Study:

• To develop and validate a method for obtaining reliable standard errors in PopPK analysis from small sample size phase I studies.
• To assess the performance of the nonparametric bootstrap approach with and without winsorization for estimating standard errors in NONMEM.
• To provide a framework for robust PopPK parameter estimation in challenging datasets.

Main Methods:

• Simulated concentration-time data for 19 subjects using a two-compartment model.
• Analyzed simulated data with NONMEM to obtain true standard errors.
• Applied the nonparametric bootstrap approach with winsorization to generate replicate datasets.
• Computed and compared winsorized and non-winsorized parameter estimates and standard errors against true values.

Main Results:

• Winsorized standard error estimates were more accurate than non-winsorized estimates, especially with skewed parameter distributions.
• The bootstrap approach combined with winsorization and at least 150 replicates yielded reliable standard errors for NONMEM PopPK parameters.
• The validated approach was successfully applied to a real-world dataset.

Conclusions:

• The bootstrap approach combined with winsorization offers a robust framework for estimating inestimable standard errors in NONMEM PopPK modeling with small sample sizes.
• This method enhances the reliability of pharmacokinetic parameter estimates in early drug development.
• The findings are applicable to both simulated and real-world PopPK datasets.