Data-driven disease modeling could improve regional response
Above: A multidisciplinary team of ASU researchers received funding from the National Science Foundation to use data from various neighboring geographical locations to help model the spread of COVID-19. Image courtesy of Shutterstock
Look up the number of confirmed cases of the novel coronavirus on the Arizona Department of Health Services website, and you will see the extent of the pandemic in each county. But how did the disease spread throughout Arizona? And how can we know what areas will be more severely affected in the future?
A multidisciplinary team of Arizona State University researchers with expertise in networked systems, time-series modeling, statistical modeling, machine learning and geospatial analysis received Rapid Response Research (RAPID) funding from the National Science Foundation to develop a data-driven model for predicting the spread of COVID-19 over time across different locations. The model is intended to help communities proactively design intervention measures to combat diseases.
Current epidemiological models are descriptive and look at data to understand the past characteristics of the disease spread, such as the basic reproduction number R0 (pronounced “R-naught”), which indicates how infectious a disease is. Gautam Dasarathy, the principal investigator on the project and an assistant professor of electrical engineering in the Ira A. Fulton Schools of Engineering, is guiding his team to add a new perspective to current models with this year-long project.
“We’re proposing something that is forward-looking, where our algorithms look at data and tell you what might happen in the future,” Dasarathy says. “The goal is then to use these predictions to design interventions, such as allocating extra hospital beds or personal protective equipment so we can be better prepared for what is to come.”
The team consists of faculty members from five ASU schools and three colleges, including the Fulton Schools, The College of Liberal Arts and Sciences and the Herberger Institute for Design and the Arts.
Time-series predictions across geographic locations
The team is working to model the spread of COVID-19 over time and space, using machine learning and data from various Arizona counties.
By looking at the development of the disease using what is known as a time-series analysis, the team is hoping to make predictions about how the disease will impact local communities in the future.
“Everything is a time-series,” says Dasarathy. “For example, the number of hospitalizations [due to the coronavirus] in the United States is a series of numbers that evolves with time. Time-series analysis techniques would allow us to understand how such series of numbers behave and how they might evolve in the future.”
He says, “If you predict that 10 days from now there is going to be a heightened demand for personal protective equipment in a county, that is a time series prediction. Therefore, you can preemptively free up resources so that you can provide support in 10 days.”
However, the researchers understand that no virus is a stagnant force. In addition to looking at the evolution of the disease over time to make predictions about the future, the team is developing techniques to track the spread of the disease across different locations.
“You keep seeing time-series data that are very location-specific,” says Dasarathy. “For example, Arizona publishes its hospitalization data and California publishes its own hospitalization data. We’re trying to develop an AI-based system that will be able to jointly look at this data to make predictions using geographical information.”
The project will look at counties or states as nodes within a network of locations that interact with one another.
“We’re developing methods to do time-series predictions on the network,” says Dasarathy. “We know how places are related by how similar the populations are or how close these places are to each other, so the model will incorporate such geographical information and other statistical similarities to make much better predictions and interventions in a pandemic situation.”
Pavan Turaga, an associate professor in the School of Electrical, Computer and Energy Engineering, one of the six Fulton Schools, and the interim director for the School of Arts, Media and Engineering, says this model can be used in future applications.
“If successful, we will be looking at advancing the palette of robust predictive models, bringing more contemporary toolkits to the hands of clinicians and epidemiologists, and engaging in a longer-term collaboration,” he says.
Connections with the community
The project has a short timeline and aims to have immediate societal impact. The information gleaned from the model will be shared with the goal to help local decision-makers, leaders in the health sector, and communities as a whole make more informed decisions about how to stay safe during the pandemic and ensure that hospitals are able to provide for patients.
Patricia Solis, the executive director of the Knowledge Exchange for Resilience at ASU and associate research professor of geography in the School of Geographical Sciences and Urban Planning, is working to bridge the gap between the predictions of the model and community outlook.
“There’s still a lot to be learned about this particular virus,” says Solis. “What we’re trying to do at the Knowledge Exchange for Resilience is make sure that this process of building knowledge is connected to the community. While the research team is looking empirically at the data, we are trying to inform, connect and translate what community actors are thinking and asking.”
The Knowledge Exchange for Resilience, an initiative funded by the Virginia G. Piper Trust, aims to link community needs with research at ASU, and is key in connecting the work the team is doing to model the spread of COVID-19 with societal action.
“People are tracking a lot of different things, such as infection rates, testing rates and deaths, as well as where the virus is moving and where it is coming from,” says Solis. “Just to get a handle on what that means for us in Arizona, there’s a lot of information to deal with.”
Looking at the predictions of the model, the next steps are to interpret them and decide what actions to take to ensure the safety of the public.
“You don’t want to wait for a crisis before you start having these conversations,” she says. “So social and organizational cohesion is very important for resilience.”
While the project is currently focused on impacts within Arizona, the team hopes that the tools developed will apply more broadly — not only to better understand how COVID-19 is progressing throughout the United States but also equip us to deal with future pandemic events.
Adding to current epidemiological models for the virus, the team hopes to use their data-driven model to provide local communities with a more comprehensive understanding of the impact and spread of COVID-19.
“We are not experts in epidemic modeling,” says Dasarathy, “but what we are good at is designing algorithms that allow data to do the talking. This project is an expression of confidence by the National Science Foundation that us data-sleuths can bring something useful to the table in dealing with such catastrophic events.”
The research team consists of Gautam Dasarathy, assistant professor of electrical engineering, Huan Liu, professor of computer science and engineering, Doug Cochran, research professor of mathematical and statistical sciences, Patricia Solis, the executive director of the Knowledge Exchange for Resilience at ASU and associate research professor of geography in the School of Geographical Sciences and Urban Planning and Pavan Turaga, an associate professor in the School of Electrical, Computer and Energy Engineering and the interim director for the School of Arts, Media and Engineering. The project is a multidisciplinary collaboration across several schools and colleges that the team hopes will help to improve regional response to COVID-19.