Comparison of Agent-Based Simulation and System Dynamics Models for Epidemiology
The desire to better understand the transmission of infectious disease in the real world has motivated the representation of epidemic diffusion in the context of quantitative simulation. In recent decades, both individual-based (such as Agent-Based) models and aggregate models (such as System Dynamics) are widely used in epidemiological modeling. This paper compares the difference between system dynamics models and agent-based models in the context of Tuberculosis (TB) transmission, considering smoking as a risk factor. The merits and impact of capturing individual heterogeneity is examined via representing Bacillus Calmette-Gurin vaccination and reactivation in both models. The simulation results of the two models exhibit distinct discrepancies in TB incidence rate and prevalence. Results also suggest that, at the level of practical application, agent-based models offer signifcantly greater accuracy and easier extension, especially when representing a decreasing reactivation rate, waning of immunity and heterogeneous individual attributes. Another experiment sought to evaluate the impact of network structure on TB diffusion. Simulations are conducted under three widely used network topologies, namely random, scale-free and small world. The results reveal large differences between results of agent-based models and system dynamics models, which further give insights into the difference between these two model types in the context of practical decision-making in healthcare.
With growing computational power, modeling techniques have increasingly attracted attention as ways of enriching understanding of the causal pathways of infectious disease. Simulation modeling offers the ability to analyze various possibilities of disease containment and to test "what-if" scenarios. In current simulation studies, two popular approaches to epidemiological modeling are System Dynamics modeling and Agent-Based modeling. System Dynamics models of infectious disease spread commonly implement structural principles drawn from the most traditional mathematical epidemiology models, which are aggregate in character. However, there has been a limited amount of System Dynamics modeling performed at the individual level. With respect to dynamics of disease, the classic System Dynamics model for propagation of infectious disease is the susceptible-infectious-recovered (SIR) model. In such aggregate models, individuals are aggregated into larger groups with same abstracted properties.
Although aggregate modeling can offer powerful insights and has allowed the derivation of the foundational concepts of mathematical epidemiology, there are distinct limitations associated with aggregate modeling when the focus is upon the specifcs of the interactions or social contacts through which the infection is spreading. Spurred by increasing computer resources and the needs for realistic scenario evaluation, agent-based modeling has become increasingly popular. This reflects the fact that it lends extra exibility in terms of representing population as a system of interacting agents with heterogeneous features and abilities. Social network modeling and analysis, as a complement to agent-based modeling, takes into account the importance of contact structure, pathways of infection spread across the associated transmission and social networks.