Using cultural, historical, and epidemiological data to inform, calibrate, and verify model structures in agent-based simulations

• Published in: Mathematical Biosciences and Engineering
• Main Authors: Lisa Sattenspiel, Jessica Dimka, Carolyn Orbann
• Format: Article in Journal/Newspaper
• Language: English
• Published: AIMS Press 2019
• Subjects: agent-based modeling; model calibration; model verification; sensitivity analysis; replication analysis; epidemic model; Biotechnology; TP248.13-248.65; Mathematics; QA1-939; Newfoundland
• Online Access: https://doi.org/10.3934/mbe.2019152; https://doaj.org/article/c80907ad4d6f4027a4187e861c8ce706

Agent-based simulation models are excellent tools for addressing questions about the spread of infectious diseases in human populations because realistic, complex behaviors as well as random factors can readily be incorporated. Agent-based models are flexible and allow for a wide variety of behaviors, time-related variables, and geographies, making the calibration process an extremely important step in model development. Such calibration procedures, including verification and validation, may be complicated, however, and usually require incorporation of substantial empirical data and theoretical knowledge of the populations and processes under study. This paper describes steps taken to build and calibrate an agent-based model of epidemic spread in an early 20th century fishing village in Newfoundland and Labrador, including a description of some of the detailed ethnographic and historical data available. We illustrate how these data were used to develop the structure of specific parts of the model. The resulting model, however, is designed to reflect a generic small community during the early 20th century and the spread of a directly transmitted disease within such a community, not the specific place that provided the data. Following the description of model development, we present the results of a replication study used to confirm the model behaves as intended. This study is also used to identify the number of simulations necessary for high confidence in average model output. We also present selected results from extensive sensitivity analyses to assess the effect that variation in parameter values has on model outcomes. After careful calibration and verification, the model can be used to address specific practical questions of interest. We provide an illustrative example of this process.