Variability in the Log Domain and Limitations to Its Approximation by the Normal Distribution

Area of Science:

• Pharmacometrics
• Pharmacokinetics
• Pharmacodynamics

Background:

• Lognormal distributions are frequently employed in pharmacokinetic-pharmacodynamic (PK/PD) modeling.
• Interpreting lognormal variability using traditional normal distribution methods presents challenges.
• Understanding the limitations of normal approximations for lognormal distributions is crucial for accurate PK/PD analysis.

Purpose of the Study:

• To compare the interpretability of lognormal distributions with normal distributions in PK/PD modeling.
• To assess the accuracy of normal approximations for lognormal distributions using formal methods.
• To provide guidance on the appropriate use of lognormal distributions in PK/PD investigations.

Main Methods:

• Utilized formal approximation methods to compare lognormal and normal distribution properties.
• Assessed approximation quality by comparing prediction intervals (PIs) to true values.
• Illustrated findings using 80% prediction intervals.

Main Results:

• Approximated prediction intervals closely matched true values when the lognormal standard deviation (omega) was less than approximately 0.25.
• Approximation precision deteriorated significantly as omega values increased beyond 1.
• At high omega values, lognormal and normal distributions showed no resemblance, highlighting the limitations of normal approximations.

Conclusions:

• Normal distribution approximations are reliable for lognormal distributions in PK/PD models only when omega is small (e.g., < 0.25).
• High omega values render normal approximations inaccurate, necessitating careful consideration of lognormal distribution properties.
• The study discusses additional statistics to aid in the interpretation of nonlinear behaviors inherent in lognormal distributions.