Weekly Update On The Science Of COVID-19: Forecasting The Future

May 27, 2020

Scientists use different models to try to forecast the spread of the coronavirus disease. The models can inform public health decisions like when to reopen businesses, but—just like weather forecasting—they’re not perfect at predicting the future. KNAU’s Melissa Sevigny spoke with computer modeler Joe Mihaljevic of Northern Arizona University for our weekly update on the science of COVID-19.

Credit CDC

Melissa Sevigny: Tell me how those models work and what kind of information goes into them?

Joe Mihaljevic: There are several different strategies for modeling the spread of infectious disease and there’s not great consensus on what the best strategy is and that’s why you get these multiple different model versions that are floating around that give you different predictions. Some of these models are using mathematical equations, so a little bit of calculus and differential equations, to try to understand how different classes of people will change over time: so people who are susceptible, verses they’re exposed but not yet infectious, or then they’re infectious but they’re asymptotic, etc. So we have different classes of individuals, and we’re trying to understand how the numbers of individuals in each one of those classes change over time.

So one of limitations is that the models are only as good as the data that goes into them.

Right, and so some of these models can be quite sensitive to some of the parameters. For example, how long does it take from when an individual’s exposed to when they start shedding the virus and they’re infectious?... We might need to understand parameters like, if you become infectious and symptomatic, how long might it take between you getting those symptoms and visiting the hospital? Those parameters are important if we want to understand how COVID-19 might affect hospitalization rates.

Are there are other shortfalls or limitations that you wanted to talk about?

I think one of the biggest limitations is that, when you think about the weather, let’s say, there’s a lot of models that try to predict what the weather will be in the future, how bad a hurricane might be. There’s a whole government organization, the National Weather Service, that develops these models and comes to a consensus about which model will do the best job or which combination of models will do the best job of predicting this weather phenomenon and the uncertainty in the weather phenomenon. There’s not that type of consensus building, necessarily, with disease models … I think that’s changing pretty rapidly, there are calls to come to consensus and to really rigorously test the different models to understand which ones are doing better job than others, so we can move forward with those best models rather than having this splattering of many different models.

So given all this complexity that we’re talking about, how do you use a model or models to help us make decisions?  

I think there are two different ways to look at modeling in terms of informing public health policy. One of those is that models can make short term forecasts, meaning we can try to predict what’s going to happen in the next two weeks…. This could help policymakers understand, OK, in the short term how many resources do we need to devote to this disease, or are we going to be at a shortfall for the hospital beds or ventilators in our region? … Models then can also be used for more longer-term predictions, but with a lot of caveats. We know we can’t predict the future, we don’t know what’s going to happen three months from now, even. But we can use our models to try to understand scenarios of what might happen or what might be the worst case that could happen…. Across the country, all of the lockdowns and interventions and wearing masks definitely slowed down the disease, and so now we’re reaching a point where some of those inventions are being lifted. How will that affect the spread of the disease? We don’t quite know yet.

Joe Mihaljevic, thank you for speaking with me.

Thanks for your time, Melissa.