Assessing the Role of Patient Generation Techniques in Virtual Clinical Trial Outcomes

by Jana L. Gevertz and Joanna R. Wares

Virtual clinical trials (VCTs) are a tool for understanding heterogeneous treatment responses. A number of techniques have been proposed to determine the set of model parametrizations ("virtual patients") that get included in a VCT. There is, however, no standard way to set the parameter prior distributions and to choose the criteria for including or excluding a parametrization sampled from the priors in the plausible population. In this work, we rigorously quantify the impact that VCT design choices have in a controlled setting using simulated patient data and a toy mathematical model. Our study provides a foundational understanding of how these choices influence the heterogeneity of virtual populations, and the predictions of a VCT.

Schematic of two methods for the generation of plausible patients for a virtual clinical trial.