The impact of COVID-19 has created challenges like never seen before. It has called upon everyone to contribute in any way they can. For engineers, this includes building new models to predict the effects of the disease. To date, the U.S. Centers for Disease Control and Prevention (CDC) consults 36 models created by institutions and universities across the United States.
Managing COVID-19 requires constantly adapting to the situation in the face of uncertainty. The CDC uses ensembled forecasts to predict the trajectory of COVID-19. The ensemble forecast combines independent forecasts from many universities over a four-week time span. Distilling the trends from multiple sources offers more robust models than any one source could provide.
Johns Hopkins COVID-19 Map
Lauren Gardner, a civil and systems engineer at Johns Hopkins University, developed one of the models included in the CDC ensemble forecast. It was the first real-time map that tracked COVID-19 cases. Many government agencies and public health departments relied on the map early in the pandemic. Gardner explained to CNN, “We were collecting data on a new virus that nobody understood at a time [when] there was not a single web page dedicated to Covid-19 case count.”
When they first created the map, there were only 320 cases reported in Asia. Since that time, the project has dramatically expanded. Now an interdisciplinary team of 25 graduate students, software developers, and researchers maintain the map. The map is updated every hour, which required the team to move from manually collecting data to automating the process.
A team from the Applied Physics Laboratory played a key role in automating the data collection. They designed an algorithm to scrape data on COVID-19 numbers from trusted sites. The gathered data is uploaded to GitHub. A second algorithm then checks the data for anomalous numbers. Any findings that seem potentially wrong are held back for review and are manually approved.
Caltech’s CS156 AI-Powered Model
Despite the success of using ensemble forecasts, a new model created at Caltech is outperforming them. The Caltech model, dubbed CS156, compared 1,500 predictions made by the CDC’s ensemble forecast to its results, and found that the CS156 model was more accurate. The key to its power is the use of artificial intelligence (AI). Few other models currently incorporate AI.
Dominic Yurk, a graduate student who worked on the project, explains, “The standard epidemiology models make basic assumptions about the way a disease will spread, and then let you make tweaks based on infection and recovery rates and so on. AI, on the other hand, takes nothing for granted.”
CS156 grew out of a class project. Yaser Abu-Mostafa was teaching a computer science class on learning systems in the spring semester, which includes building a learning system as a class project. Normally, the class builds a movie recommender system. However, this year, Alexander Zlokapa, a senior Caltech student who runs the Caltech Data Science Organization, suggested the course focus on COVID-19 models.
At first, Abu-Mostafa was concerned about taking on such a somber topic. So, he sent a message to the students asking them what they thought about changing the topic of the project. Abu-Mostafa found that “[t]he response was overwhelming. We all saw an opportunity to make a real difference for the country.”
The student models focus on predicting the COVID-19 mortality rate. This focus was selected because it is a reliable metric. Rates of confirmed cases can be influenced by testing and may not reflect the actual number of those infected. The student models factored in information about population demographics, community density, stay-at-home orders, available clinical information, and other relevant factors.
Throughout the semester, the class created 40 models that made reasonable predictions. The top 10 were even as accurate as existing models. But this was just the beginning. After the semester concluded, many students were interested in continuing their work on creating a better model. They wanted to take the best aspects of the top models and aggregate them into a stronger model.
Abu-Mostafa was able to divert his funding for another project toward continuing the work on the model. He formed a team of 8 graduate students and 18 undergraduate students. A Summer Undergraduate Research Fellowship allowed him to add another seven students. In total, 80 percent of the project’s budget supported student-researcher salaries.
The CS156 model officially launched in August. Abu-Mostafa continued to manage the model once the students returned to class, but he credits the success of their model to the hard work of his students.
University of Wisconsin-Madison Region-Specific Model
Policymakers do not just rely on large-scale models. Understanding how local conditions affect outcomes is also essential. Oguzhan Alagoz, an industrial and systems engineer at the University of Wisconsin-Madison, led a team to develop a model that could show how social distancing measures affect caseloads in New York City, Milwaukee, and Dane County.
Their model was created using cell phone data. The location data tracked how social distancing directives changed the distance people kept between themselves and others when they gathered. This information was used to determine if people complied with social distancing orders. The model showed that social distancing orders did affect behavior, but adherence to the orders varied greatly by location.
Nasia Safdar, a professor of medicine who collaborated on the project said, “Context matters for prevention activities, and region-specific data and the expected effect of mitigation is invaluable to get buy-in and engage communities in mitigation efforts.”
Their model allows them to predict how implementing social distancing orders will impact caseloads in different areas. It can also assess the effectiveness of past plans. For example, they found that had social distancing orders been in effect a week earlier, New York City would have only seen 20 percent of the cases they experienced.
Georgia Institute of Technology Focuses on Supply Chains
Not all models are tackling disease spread. Pinar Keskinocak, a systems engineer at Georgia Institute of Technology, is focusing on how COVID-19 is expected to disrupt supply chains. Pandemics affect more than just our health. They disrupt many systems and cause economic damage, which can disrupt supply chains. The steps taken during a pandemic can compound or lessen these effects.
Keskinocak’s model focuses on how resource allocation strategies, disaster preparedness, and disaster response affect health care facility inventories. In the past, she has created models for cholera, guinea worm disease, malaria, and pandemic flu. Now, she has turned her attention to COVID-19.
“Whenever there is a complex problem, it needs serious analysis or technology and that’s where an engineer comes in,” Keskinocak told ASME. “This is exactly the situation now, very complex, dynamic and uncertain. It’s difficult to understand what’s going on or make decisions just by sitting around a table and discussing.”