Simulation of correlated continuous and categorical variables using a single multivariate distribution

Area of Science:

• Biostatistics
• Computational Biology
• Clinical Trial Design

Background:

• Clinical trial simulations require physiologically realistic virtual patient populations.
• Covariate distribution modeling is crucial for generating representative virtual patient data.
• Existing methods like the Discrete method face limitations with multiple categorical covariates, leading to sparse subgroups.

Purpose of the Study:

• To introduce and evaluate a novel statistical methodology, the Continuous method, for covariate distribution modeling.
• To compare the performance of the Continuous method against the Discrete method in generating virtual patient covariate distributions.
• To assess the efficiency and accuracy improvements offered by the Continuous method.

Main Methods:

• Developed and applied a Continuous method for sampling complete covariate vectors from a single multivariate function.
• Compared the Continuous and Discrete methods using simulated and real data with varying covariate characteristics.
• Evaluated methods based on their ability to match target population distributions, summary statistics, and subgroup proportions.

Main Results:

• Both Continuous and Discrete methods accurately generated summary statistics and population proportions.
• The Continuous method performed comparably to the Discrete method across various scenarios.
• The Continuous method demonstrated advantages in efficiency by analyzing the full population, enhancing precision of covariance estimates.

Conclusions:

• The Continuous method offers a robust and efficient alternative for covariate distribution modeling in clinical trial simulations.
• This method effectively addresses the limitations of the Discrete method when dealing with multiple categorical covariates.
• The Continuous method enhances the accuracy of virtual patient population descriptions, crucial for reliable clinical trial simulations.