Using linear regression metamodels for evaluating interventions in an individual-based influenza epidemic model

Abstract

Agent-based simulation modeling is frequently used to model and simulate the spread of transmissible diseases such as influenza, COVID-19, and HIV/AIDS in communities. Besides incorporating disease-specific parameters, these models include a set of parameters to observe the effect of different intervention combinations on the course of an epidemic, bringing the opportunity to use these models as virtual laboratories for decision-making. However, these models are primarily large-scale and complex, increasing the runtime of experimentation. As a solution, metamodeling approaches are frequently employed to represent input-output relationships of simulation models. Instead of running the time-consuming agent-based model, policymakers use the metamodel to obtain predicted outcomes in a comparatively short time. In addition to time-saving advantages, metamodels can provide insights into how disease-specific and intervention parameters affect the outcome of interest. In this regard, this study uses an influenza epidemic model, FluTE, as the experimental platform. Instead of using the original agent-based model, we fit linear regression metamodels to quantify the effect of interventions, such as vaccination, quarantine, and school closure, on the influenza attack rate. After validating the metamodel, we observe that the day on which interventions start, ascertainment delay, the daily number of vaccinations administered, isolation and quarantine compliance probabilities, and the number of school closure days stand as the significant intervention policies.