Pharmacokinetic parameter estimations by minimum relative entropy method

Area of Science:

• Pharmacokinetics
• Statistical modeling
• Nonlinear regression analysis

Background:

• Estimating pharmacokinetic parameters is crucial for drug development and clinical application.
• Traditional least squares methods (OLS, WLS, IRLS) can yield misleading results due to the "choice of weights" problem.
• Extended least squares (ELS) offers a newer approach to nonlinear regression analysis.

Purpose of the Study:

• Introduce the minimum relative entropy (MRE) method for pharmacokinetic parameter estimation.
• Compare the performance of MRE against various least squares methods, including OLS and ELS.
• Evaluate the robustness of MRE in addressing the "choice of weights" problem in pharmacokinetic modeling.

Main Methods:

• Developed and applied the minimum relative entropy (MRE) method, extending relative entropy to positive functions.
• Conducted an intensive simulation study using four pharmacokinetic models (mono-/biexponential, Bateman, Michaelis-Menten).
• Employed several variance models for observation errors to assess method performance under different error distributions.

Main Results:

• Ordinary least squares (OLS) performed best when error variance was constant, but poorly otherwise.
• MRE outperformed ELS and OLS when observation variance was proportional to its mean.
• ELS was superior when the standard deviation of observation was proportional to its mean, with MRE performance being comparable to ELS.

Conclusions:

• MRE is a reliable method for pharmacokinetic parameter estimation, comparable in performance to ELS.
• MRE offers a potential solution to the "choice of weights" problem by not assuming error distributions.
• The choice of method (MRE, ELS, OLS) depends on the specific characteristics of the observation error variance in pharmacokinetic studies.