mvLognCorrEst: an R package for sampling from multivariate lognormal distributions and estimating correlations from...

Area of Science:

• Pharmacometrics and computational drug development.
• Statistical modeling and simulation.
• Quantitative systems pharmacology.

Background:

• Pharmacometrics (PMX) utilizes modeling and simulations (M&S) for drug development decisions.
• Sensitivity Analysis (SA) and Global Sensitivity Analysis (GSA) evaluate model-informed inference quality.
• Accurate simulations require correct parameter correlation handling, which is complex for lognormal distributions.

Purpose of the Study:

• To present mvLognCorrEst, an R package designed to overcome challenges in sampling from multivariate lognormal distributions with correlated parameters.
• To provide methods for estimating partially defined correlation matrices while preserving positive semi-definiteness.
• To support robust simulation-based analyses in pharmacometrics.

Main Methods:

• Developed a sampling strategy by transforming multivariate lognormal distributions to their underlying Normal distributions.
• Addressed issues with high lognormal coefficients of variation (CVs) by approximating non-positive definite Normal covariance matrices.
• Employed graph theory for estimating unspecified correlation terms within a constrained optimization framework.

Main Results:

• The mvLognCorrEst package offers functions to handle correlated lognormal distributions and estimate correlation matrices.
• Demonstrated the package's utility through a case study involving the GSA of a preclinical oncology pharmacometric model.
• Successfully applied the methods to address sampling and correlation estimation challenges.

Conclusions:

• mvLognCorrEst is a valuable R tool for pharmacometricians.
• The package facilitates simulation-based analyses requiring sampling from correlated multivariate lognormal distributions.
• It aids in estimating partially defined correlation matrices, enhancing the reliability of M&S in drug development.