CMStatistics 2022: Start Registration
View Submission - CMStatistics
B0898
Title: RVine copula as virtual baseline generator in agent-based modeling in clinical research Authors:  Nicolas Savy - Toulouse Institute of Mathematics (France) [presenting]
Philippe Saint-Pierre - Mathematical Institut of Toulouse (France)
Abstract: Simulation is today considered the third pillar of science, a peer alongside theory and experimentation. Indeed, simulation is a relevant means to analyze complex systems. In medical research, a wide range of questions can be investigated through simulation involving various simulation tools. In a series of articles, the authors have reflected on the use of agent-based models for clinical research. The common idea is to use the enormous amount of information available about the patients of interest, the disease of interest, the drug of interest and the trial design in order to build a stochastic model mimicking the course of clinical research. The strategy consists of identifying design weaknesses, measuring the performance of a trial within a predefined framework while reducing the number of logistical barriers. Two classes of models have been identified: first, models to generate baseline values for a patient cohort and second, models to define the evolution of the outcome(s) of interest. The first class of models are called virtual baseline generators and the second class execution models. Our focus here is on the first model class. Virtual Baseline Generators essentially consist of Monte-Carlo generators of covariate vectors. We will discuss the relevance of the RVine copula as a VBG. These models are attractive because they are able to capture the correlation structure of a training dataset and generate data based on that structure.