A team from Northern Arizona University recently received a grant from the National Institutes for Health (NIH) to develop a platform that will help local public health experts find and evaluate models to predict the spread of the disease.
During outbreaks, researchers and public health experts will develop models to predict the spread of the pathogen. The processes involve mathematical equations based on patterns followed not only by the disease and its particular variant, but also by the behavior of people in different domains.
As seen with COVID-19, changes can happen quickly and often, so models are updated frequently and can be quite varied.
The NAU project, known as EpiMoRPH (Epidemiological Modeling Resources for Public Health), is developing a platform that can compare multiple models to determine their accuracy and the situations or areas in which they work best.
People also read…
“If you’re a public health worker in a particular city or state, you can go into our system and figure out what existing models might work best for your area, and then our system could run that model for your area and make predictions. said Joe Mihaljevic, an assistant professor in NAU’s School of Informatics, Computing and Cyber Systems (SICCS) and the project’s principal investigator.
EpiMoRPH is a collaboration between NAU researchers including Eck Doerry, Crystal Hepp, and Samantha Sabo, as well as TGen North, Arizona State University, University of Nebraska Lincoln, and Northwestern University
As Mihaljevic described it, epidemiological models are mathematical equations intended to show the transmission process of a disease at a specific place and time. To do this, modelers assess the past behavior of the disease.
“It’s really about trying to understand in the past, so far, how many hospitalizations have we seen over time, and then trying to fit that pattern in the model to the data,” Mihaljevic said. “If it can fit this pattern in the past, then we have more confidence that anything we see in the future could be true. It tries to fit the past to the data…then make a long-term forecast of two at three weeks.
The models use information — about how a disease spreads in different ZIP codes, for example, or in rural versus urban areas — to make their predictions.
With a disease like COVID, this may include behavioral differences between different variants or sub-variants or changing public health recommendations.
“When we put vaccines in place, we didn’t know the whole dynamics of vaccines, like how long would they last before someone could get re-infected, and so our models had to make assumptions about those things until that we have data and that we can update the models and improve them,” Mihaljevic said.
Especially for new or rapidly changing diseases, modelers enter equations based on their current understanding of these factors. Some of these models will make more accurate predictions than others, which the EpiMoRPH platform will be designed to determine.
The accuracy of a model is based on how closely its predictions match what is actually happening in that time frame. Comparisons between a model’s predictions and reality can also be used to make adjustments to increase its accuracy in the future.
Mihaljevic said the models need to be constantly updated.
“Especially when we don’t know as much about a particular emerging pathogen, we may not know how to build the best models; we just make assumptions that seem reasonable. Only over time do we realize which models work best, but there is no automated infrastructure to understand which models might work best in which places, because different assumptions may be better in different places “, did he declare.
Early in the COVID pandemic, for example, Arizona used a statewide model to predict hospitalizations. However, the differences between cities in the state meant that they were difficult to use on a more local level.
Once the platform is complete, the idea is to make it a place where local public health experts could find models that would work best in their region and help them develop predictions specific to that region. It relies on model submissions and updates hopefully from across the country.
“So in our system, people in Flagstaff could say, ‘I want to handle this [statewide] model, ‘but if I predict it using my location data that we have here in Flagstaff, then we might get a more accurate and spatially refined prediction for that person,’ he said.
He said the system would have a dashboard-like update feature that modelers can use to see their model’s performance and possibly guide their updates.
The project received a $3.5 million grant from the NIH in April. It is currently funded for up to five years with the potential for renewal, during which time the team hopes to develop a prototype of the platform, test and consult with public health partners throughout the process.
“I hope our system can integrate a lot of things that we have learned and improve our ability to quickly build good representative mathematical models that explain the spread of the disease,” Mihaljevic said. “But more than that, part of the project is actually working closely with public health actors… to make sure that as we develop the technology, we’re building something useful. which will be used not only by modellers. and academia, but also public health.