Discipline: Biological Sciences
Subcategory: Microbiology/Immunology/Virology
Session: 3
Tanisha Thomas - North Carolina Agricultural and Technical State University
Co-Author(s): Briquelle Martin, Appalachian State University, Boone NC; Andrew Ackerman, Clemson University, Clemson, SC
Contact Network models are becoming the predominant alternative to equation-based models for analyzing epidemiological data. Whereas the equation based methods (mostly derived from ordinary and partial differential equations) use a mass action model that assumes equal likelihood of contraction among all individuals in a population, network structure allows for more dynamic (and subsequently more accurate) modeling of the fluid nature of human social interaction. Up to date, most of the work in contact network epidemiology has been done under the framework of the Susceptible-Infected-Recovered (SIR) model. This provides a simple and broadly applicable groundwork, yet in the case of many diseases is oversimplified and can lead to error in model prediction accuracy.
We have attempted to use dynamical modeling in conjunction with the Susceptible-Exposed-Infected-Recovered (SEIR) model. This extra phase takes into account the potential for a disease to be present in a host (for a considerable period of time) without presenting symptoms or becoming contagious. Further, we hope to improve the quality of the network structure itself by optimizing the edge-weight calculations through analysis of weights dependent on both a single variable and multiple variables. In each case, we calculate the weights through an application of bond percolation theory. Firstly, with respect to a single variable, we analyzed a set of airline data provided from the Federal Aviation Administration which details the number of flights (and its respective passengers) departing and arriving at any given airport within a specific timeframe. Thus, we will consider the airport (or more accurately the city) to be the node subject to infection and air travel the mode of transmission.
Next, with respect to multiple variables, we will analyze a data set from a 2012 Kenyan Household Contact study. In order to use bond percolation theory, we need a coherent probability distribution. Thus, we will treat each variable as an independent event and multiply their respective probabilities to create a singular resultant weight distribution. Finally, our system is dynamic in the sense that when a node is infected, human tendency dictates that this node’s contact will decrease by some percentage proportional to the awareness of the disease. Our calculations will reflect this decrease over time and thus fluctuate in a more realistic manner.
In comparison to previous SIR, static models with equal weights or weights dependent upon one variable, our model demonstrates a much higher predictive capacity. On both a large scale nation-wide epidemic (flight data) and specific contained population (Kenyan Household data) we demonstrate a high level of increased accuracy.
Our study provides insights that expand beyond previous work in the field and increase ability to predict, track, and prevent epidemics.
Funder Acknowledgement(s): National Science Foundation, grant HRD-1719488; National Security Agency, grant H98230-18-1-0097
Faculty Advisor: Kossi Edoh, kdedoh@ncat.edu
Role: I was responsible for obtaining data sets and cleaning them/ preparing them so they would run properly through the program. Also, I wrote portions of the paper.