it’s Predict

Why It’s So Hard to Predict Where the Pandemic Is Headed Next

Among the professional hazards faced by Carl Bergstrom, a professor of biology at the University of Washington, is that he is often asked where the Covid-19 pandemic is heading. The question comes in many variations—what next week will be like, or the next school year, or the next winter—and has so for as long as…

Among the professional hazards faced by Carl Bergstrom, a professor of biology at the University of Washington, is that he is often asked where the Covid-19 pandemic is heading. The question comes in many variations—what next week will be like, or the next school year, or the next winter—and has so for as long as the virus has been with us. But recently it has gained a certain fervor. Bergstrom works at the intersection of two relevant subjects: how sentient beings like ourselves act on information, and how biological phenomena like viruses spread. So if anyone’s your answer guy, it’s him.

Lately, he’s been replying with a blunt take: “I don’t know.”

It’s a short answer that conceals a good deal of nuance. Since the beginning of the pandemic, the job of disease modelers has not been to tell us precisely where we are going, but to prepare us for many possible futures. This is a fraught business. Offering multiple options in a crisis invites people to run away with one conclusion or another as it suits them, leading to too much sacrifice or too much wishful thinking. (Remember when the Trump administration seized upon the most optimistic forecasts to declare that the pandemic would be over by summer—that is, last summer?) Models can help policymakers decide where to put resources, and they can also help people like you and me find some mooring in an uncertain world. Oracles, however, they are not.

The reason is that at any moment in a disease outbreak, a projection may rise or fall exponentially depending on its initial assumptions. Those assumptions are hard to make. In the beginning, epidemiologists were scrambling to understand the very basics of a new pathogen: how the virus spreads between people, how fast it incubates, the role of super-spreaders and asymptomatic infections in seeding a so-called “invisible pandemic.” Over time, they got a better grip, with the help of a full-court scientific press—more virological and immunological data about how the virus infects, and more epidemiological data about what happens next. Once researchers understood how the virus moved, it was easier to determine how to turn it back with things like masks and social distancing.

But even with answers, that uncertainty never goes away. Consider the present: Delta itself, of course, has also brought its own set of unknowns related to its faster replication and ability to infect. So has vaccination, including the extent to which vaccinated people spread the virus, and how well immunity holds up over time. These all affect how severe the Delta wave will be at any particular time and place. And, as we settle those questions, there’s always the potential for a new variant to throw any long-term calculations off. “We definitely have more information, but I wouldn’t say the number of unknowns has really decreased,” says Emmanuela Gakidou, a professor of health metrics science at the University of Washington. “I wouldn’t say we’ve ever been content that we’ll have a model that will ever be used for more than a week in a row.”

Bergstrom suggests thinking of it this way: In March 2020, how would a disease modeler have predicted the ups and downs that were to come? The pandemic is now said to be in its fourth wave, but the term belies a far more complex topography of stubborn plateaus, gentle bunny hills, and striking peaks. Even in retrospect, the patterns are difficult to explain (and not just because time is now a blur and no longer has meaning). Some changes were due to the virus, and others due to how we responded. During the first wave, public life ground to a halt following national stay-at-home orders. These were replaced by mask mandates and partial, sometimes halting, reopenings.

But it is also a landscape of shifting frustrations and fatigue, wild alternations between pessimism and optimism, such as last fall, when Americans returned to holiday travel amidst what was then the pandemic’s worst surge. And now, despite a summer peak that is as bad as it’s ever been, in many parts of the country society is largely back to business as usual. “People dramatically change their behavior during an ongoing pandemic,” Bergstrom says. “We constantly update our beliefs about how serious this is.”

In some ways, that means more experience with the pandemic can create more uncertainty for modelers, not less. Beliefs and behaviors are now increasingly heterogeneous, varying from state to state and, in some cases, town to town. Delta has arrived at a time when people are growing more polarized in the wake of vaccinations, and confused about what that means for how they should behave. “One month mask mandates are OK, and the next month it’s protests. It’s really hard to predict in advance,” Gakidou says.

“The prevailing theme that continues to make things hard now is the interplay between disease state, how people react, and how people react over time,” says Joshua Weitz, a professor who studies complex biological systems at the Georgia Institute of Technology. It’s a perfectly intuitive idea 18 months into the pandemic that our individual perception of risk, and the behaviors that follow from it, should have a collective impact on the virus’s trajectory. But that wasn’t the universal understanding at the start, Weitz notes, when some believed that the pandemic would pass quickly. In modeling-speak, the term for that (a relic of 19th-century epidemic theory) is Farr’s law: Infections should peak and then wane at relatively equal rates, producing a bell curve. 

This curve wasn’t going to obey. Last spring, Weitz and others could see it was coming back for round two. The first wave hadn’t been completely crushed, and too many people remained susceptible. Cases peaked, then got stuck on the “shoulders” of the curve, declining at a slower rate than many projections suggested, and then plateaued at stubbornly high rates of infection. Behavior, Weitz hypothesized, was not in sync with how models predicted interventions like stay-at-home orders would work. By studying mobility reports drawn from cell phone data, a proxy for how much social contact people are experiencing, he could see that risky behavior decreased as fatalities climbed, but then began to rebound before the corner was turned. “People look around, see the local situation, and they change their behavior,” Weitz says.

One consequence of these reactive behaviors is that it can be hard to analyze how helpful policies like mask and vaccine mandates are. There’s a blurring between cause and effect—and between government actions and what the public is already doing as both react to the rise and fall of transmission rates. For example, he says, if you look at the timing of the mask mandate instituted last year in Georgia, and compare the case rates before and after, you might determine it had little effect. But what if that was because people realized case rates were rising and preemptively donned their masks earlier? What if they just started staying home more? Or what if it was the other way around: The requirement went into effect and few people followed the rules, so the masks never had a chance to do their work? “There is clearly a relationship there,” he says. “I can’t claim we got to the bottom of it.”

For modelers, this uncertainty presents a challenge. To evaluate when the Delta surge may end, one might look to places where it has already surged and crested, like the United Kingdom. But will it die down quickly, or take a slower taper, or perhaps plateau at a steady ra

Read More

Leave a Reply

Your email address will not be published. Required fields are marked *