Evaluation of mixture modeling with count data using NONMEM.

Area of Science:

• Pharmacometrics
• Population Pharmacokinetics
• Pharmacodynamics

Background:

• Mixture modeling in PK/PD analysis can identify unobserved subpopulations.
• Existing statistical tests for mixture existence are lacking.
• Simulation studies are crucial for evaluating such methods.

Purpose of the Study:

• Evaluate mixture modeling's ability to detect subpopulations.
• Assess false positive rates and statistical power.
• Examine parameter estimation accuracy and subject classification.

Main Methods:

• Simulated seizure count data using Poisson distribution and Emax models.
• Employed NONMEM for mixture modeling analysis.
• Varied Emax values and subpopulation proportions for 16 mixture scenarios.

Main Results:

• False positive rate was approximately 0.04-0.078 when no mixture was present.
• Accurate characterization of mixtures improved with greater differences between subpopulations.
• Median parameter estimates were within +/- 28% of true values, with correct subject classification ranging from 0.59 to 0.96.

Conclusions:

• Mixture modeling is a viable tool for identifying subpopulations in PK/PD.
• The method demonstrates acceptable false positive rates and good power.
• Mixture characterization success depends on the magnitude of differences between subpopulations.