Epidemiology team guides national policymakers
The Gillings School’s Justin Lessler explains how the COVID-19 Scenario Modeling Hub helps inform decisions about vaccines, mask mandates and more.
Since the start of the COVID-19 pandemic, epidemiologist Justin Lessler has been a go-to expert for how the SARS-CoV-2 situation might play out and what that could mean for humanity. He is well known among policymakers, like those at the Centers for Disease Control and Prevention and the White House, as well as the country’s top media outlets — The Atlantic, CNN, NBC News, The New York Times and many others.
One reason they keep seeking guidance from Lessler, professor in the Gillings School of Global Public Health’s epidemiology department, is the COVID-19 Scenario Modeling Hub, which he helped create in late 2020 and continues to oversee. Lessler came to Carolina in 2021 after more than a dozen years at the Johns Hopkins Bloomberg School of Public Health.
The Well scheduled a Zoom interview to learn more about the COVID-19 Scenario Modeling Hub and how it influences decisions about vaccines, mask mandates and more.
Justin Lessler
Why did you launch the COVID-19 Scenario Modeling Hub?
Early in the pandemic, a group of epidemiologists around the country created the COVID-19 Forecast Hub, which combines results from multiple models to predict what might happen four weeks out.
We saw a need for generating longer-term COVID-19 projections combining insights from different models and making them available to decision-makers, public health experts and the general public. Typically, our modeling rounds project six to nine months into the future.
How does the hub work?
It’s important to note up front that while I’ve been one of the most vocal people involved, this is a team effort. We have between five and nine teams doing national models and even more at the state level. Every round of modeling involves multiple institutions.
We set up our planning scenarios in a two-by-two table. We end up with four scenarios on two axes. For example, a variant may or may not be severe, and governments may or may not keep mask mandates in place. That gives us four scenarios.
Once we agree on scenarios, we begin a new round of modeling, in which the different teams around the country plug in data — actual mechanistic modeling of the disease process — to project the probability of what might happen in different scenarios. Each round takes about six weeks to complete. We completed our first round in January of 2021, and we recently completed Round 16.
Why is your ensemble approach important?
We know from weather forecasts and a whole host of other predictive efforts that these sorts of ensemble methods outperform any individual model all the time. Individual models have latent assumptions in them.
Back in April of 2020, there were all these individual scenario models. There was the Imperial College model showing what would happen if nothing were done. There was the Institute for Health Metrics and Evaluation model that the White House loved, but that model was assuming that lockdowns continued through the summer. There were individual models showing lockdowns lifting and various things happening afterwards.
What happened is that these models all showed different things and people would say, “Nobody knows what they’re talking about.” Or they’d say, “Oh, I like that one. That one is the one I’m going to focus on because it confirms my agenda or preconceived notions.” I like to call this the choose-your-own-adventure phase of pandemic modeling.
They were all modeling different policy scenarios. Nobody was modeling the same thing. In my mind, the first and foremost thing about our hub isn’t the ensemble. It’s getting a bunch of people together and asking the exact same question.
Does your COVID-19 Scenario Modeling Hub provide a forecast?
Not exactly. We’re not like the COVID-19 Forecast Hub, where they update every week. They pose the same question, just with newer data. The COVID-19 Forecast Hub is in the business of saying what will happen. The COVID-19 Scenario Modeling Hub is in the business of saying what is likely to happen if certain conditions hold. In other words, we are providing what I like to call “planning scenarios.”
It’s hard enough to forecast weather more than a couple weeks out, and for weather we get something like a billion data points an hour. For infectious diseases, on the other hand, we’re looking at poorly measured data — one measure of cases, one measure of hospitalizations, one measure of deaths — once a day, if we’re lucky.
Another difference: If a hurricane is coming toward Miami and you forecast that and people evacuate Miami, the hurricane doesn’t care that you evacuated Miami. It’s going to hit Miami or not. But if it’s an infectious disease, and you predict that Miami is in for a big wave, and the mayor of Miami says, “Okay, everybody, wear your masks and stay in your homes,” and people do that, you’re not going to have your big wave. There are these feedback loops and big changes in how people behave.
So how might the COVID-19 Scenario Modeling Hub influence a policy decision? Do you have an example?
Round 15 played a role in the decision by the CDC’s American Council of Immunization Practices to recommend the rollout of the reformulated bivalent booster in September rather wait for more data on the omicron subvariant BA.5. We showed that you could save thousands of lives and prevent many thousands of hospitalizations by having the booster campaign happen in September rather than November.
They also cited Round 14, where we compared a narrower vaccination campaign targeted only at the 50-plus population. So the vaccine campaign happened in September for people 12 and up.
What four scenarios did you model in Round 15?
One level was a wide uptake of boosters starting in September for people 18 and older. The second level of that was the same type of campaign but not starting till November. And then the other axis was BA.5 sticking around, with no new variants. And the other level of that axis was a new variant emerging, which looks surprisingly consistent with the actual variants we saw.
You recently completed Round 16. What do your findings tell us about this coming winter?
We’re in a situation where we have lower vaccination rates than we hoped. We do have new immune escape variants. So based on those projections, it looks like peaks in hospitalizations in December or January are likely. And while it’s not the most likely, the possibility exists that those could get rather high and sort of rival previous omicron waves.
How have the scenarios changed since you began two years ago?
We started by focusing on the initial vaccine rollout and non-pharmaceutical interventions and behavior change. As we got into the summer of 2021, the new variants — delta and then omicron — became a big part of what we were looking at in addition to continued vaccine and booster uptake.
And then means of immunity started becoming a bigger focus in that it became clear that we weren’t just seeing new variants, we were also seeing some waning protection from the vaccine and natural infection. So that became a big focus. And then after omicron, and as we entered the summer of 2022, we focused on potential new variants and when we would get the bivalent vaccine booster and how well it worked. That’s where we are now.
Are you working on another round?
We’re not rushing for a Round 17. We’re not going to update unless there’s a really good reason to feel like the data have changed the picture. Or, more likely, because there are new questions coming from the CDC or the White House or the many other stakeholders that we talk to. We develop the scenarios with those questions in mind. We in the coordination team will say, “OK, what two-by-two problem are we going to work on this time?” Hopefully there’s no super compelling reason. When there’s a reason to rush, it’s not because things are going well.
