CMStatistics 2020: Start Registration
View Submission - CMStatistics
B0772
Title: Agent-based modeling for medical research: Economic impact of generic antiretrovirals in France for HIV patients care Authors:  Nicolas Savy - Toulouse Institute of Mathematics (France) [presenting]
Philippe Saint-Pierre - Mathematical Institut of Toulouse (France)
Abstract: Agent-based modeling consists of a set of models whose aim is to mimic the behavior of individuals in a random environment. In a health context, agent-based modeling may be used to simulate the behavior of patients or the effect of a treatment on patients. To do so, virtual patients are randomly generated, and models are used to predict their medical outcomes under different scenarios of treatment or treatment effects. By comparing these scenarios, it is possible to derive an estimate of the effect of an intervention or treatment on medical outcomes. That strategy is usually called In Silico Clinical Trial (ISCT). We will highlight the milestones of this strategy, which is strongly based on predictive models. Artificial Intelligence has shown wonderful power for prediction, although many strategies are available to predict an outcome from data. We focus our attention on the main methodological pitfall: to perform agent-based modeling, a predictive model is not enough, sharp modeling of the error of prediction is necessary. As an example, we will discuss an agent-based model developed to simulate patient trajectories and treatment use over a five years period. Comparing the cost results obtained for trajectories simulated under different predefined scenarios then allows us to build a Budget Impact Model as well as sensitivity analyses on several parameters of importance.