Gov. Brian Kemp’s aggressive scheme to lift Georgia out of COVID-19 lockdown may cost many thousands of lives, according to models prepared by epidemiologists and computer scientists at Harvard and the Massachusetts Institute of Technology in partnership with The Daily Beast.
The findings come as governors across the United States aim to restore economic activity following months of pandemic-related infections and over 50,000 deaths—a number widely understood to be an undercount. Meanwhile, over 26 million Americans have filed for unemployment benefits in recent weeks, a number that is itself a likely undercount of the economic toll.
Georgia’s Kemp has perhaps been the boldest of any governor about moving on, issuing a pair of executive orders allowing fitness centers, tattoo and massage parlors, bowling alleys, and hair salons to reopen last Friday with some mitigation measures. Other businesses, like restaurants and theaters, began opening Monday. The state’s shelter-in-place decree, meanwhile, was slated to expire on Thursday.
Those policies are placing Georgians at spectacular risk, the new models found.
As of Friday, by official counts in Georgia, at least 871 people statewide had lost their lives to COVID-19. If Georgia had maintained its pre-Friday lockdown policy, the Harvard/MIT team’s simulation—which used data from the Johns Hopkins Coronavirus Resource Center and accounts for local demographics and health conditions based on Census and survey data—estimated the state would have logged a total of between 1,004 and 2,922 coronavirus fatalities by June 15. That fatality range, like all such ranges detailed in this article, includes deaths that had already been documented (in this case, 871).
By contrast, under Kemp’s current plan to reopen, if approved businesses returned to just 50 percent of their pre-pandemic activity (or “contact”) levels, that range could reach 1,604 to 4,236 deaths. At 100 percent of pre-shutdown activity, the projected final body count could soar to a range between 4,279 and 9,748.
Even if employee-on-employee contact returned to just one-quarter of what it was before the disease hit, and interactions among the general public—beginning April 30—reached 20 percent of the old norm, the researchers projected that deaths in the state could hit 3,563.
“What we find, no matter what we assume, is that reopening on Monday was just too early,” said Jackson Killian, Ph.D. student at Harvard’s John A. Paulson School of Engineering and Applied Sciences, who worked on the models. “If you let people go out and have contact again now, you end up causing deaths that could have been avoided.”
Based on the nature and speed of COVID-19’s spread through Georgia, Killian and his team estimated the virus may have arrived in the state as early as Feb. 1, or at least weeks before the first diagnosed cases—a possibility Kemp himself has acknowledged. To be clear, the models cannot prove or verify that the first infection happened on that date, but used it as an assumed start date based on the available information and the spread to date.
The governor’s office did not respond to repeated requests for comment for this story.
For their part, the team behind the models framed their approach not as an argument for absolutes, but a testament to dire stakes.
“The stay-at-home orders cannot go on indefinitely,” said Maimuna Majumder, faculty member at the Computational Health Informatics Program and Harvard Medical School who led the creation of the models in partnership with Milind Tambe, a professor of computer science and director of Harvard’s Center for Research on Computation and Society. Instead, she emphasized the need for a “new normal [that] still allows people to go back to work” and that acknowledges “each of us can make a difference by physically distancing ourselves at, for example, grocery stores.”
Turgay Ayer, an associate professor at Georgia Institute of Technology’s School of Industrial and Systems Engineering, recently released a state-by-state COVID-19 simulator with colleagues at Harvard and researchers at Massachusetts General Hospital that he said found results “in line” with the estimations from the Harvard- MIT group.
Ayer’s simulator showed that—under minimal restrictions, with no other interventions—there could be up to 20,000 deaths by Aug. 30 in Georgia, but he noted that was a worst-case scenario he didn’t expect to see. That’s because he believes politicians like Kemp will reimplement some restrictions once a resurgence of infections appears.
“Once we start to see a second spike in infections in late July and early August, the policymakers will put some of these social distancing measurements back in place,” said Ayer. But the numbers do show one thing very clearly, he said: “If you lift the restriction too soon, a second wave will come, and the damage will be substantial both medically and economically. We don’t want to throw away the sacrifices we have made for weeks now.”
The Harvard and MIT modelers working with The Daily Beast also looked at two neighboring states that, like Georgia, were hesitant to implement shutdowns in the first place, and are now mulling their own reopening plans: Florida and Mississippi. The results were similarly alarming.
To be sure, Florida Gov. Ron DeSantis and Mississippi Gov. Tate Reeves have been more cautious than Kemp. DeSantis has, so far, mostly kept his state’s social distancing measures in place, while allowing localities to reopen beaches. He has also convened a Re-Open Florida Task Force to present a program for resuscitating commerce in the Sunshine State. The shelter-in-place order in Florida, like that in its neighbor to the north, was scheduled to sunset at month’s end.
As of Friday, 987 Florida residents had been identified by the state as having died from COVID-19. Should DeSantis back off plans to reopen businesses and renew his stay-at-home decree through June 15, the Harvard/MIT/Daily Beast model projected his state would witness a total number of deaths as small as 988 or as large as 3,014 due to the virus.
But if DeSantis had implemented Kemp’s aggressive reopening policies in recent days, the loss of life might have escalated to a range of 1,273 to 4,106 fatalities in the lowest-contact scenario, or even as high as 15,523 deaths if businesses returned to their pre-COVID-19 levels. DeSantis’ office did not reply to repeated inquiries from The Daily Beast.
Reeves, meanwhile, appears to be plotting a course between Kemp’s attempted renaissance and a more prolonged shutdown. The Mississippi governor inked a decree on April 24—by which point 201 of his constituents had been identified as having died of COVID-19-related causes—that will keep the state’s gyms, salons, and theaters mostly closed and continue to limit eateries to take-out and delivery. But it will enable other retail stores to reopen at 50 percent capacity and for elective surgeries to resume. This fiat superseded an earlier shelter-in-place order with a looser “safer at home” policy, which is scheduled to remain in effect through May 11. The group from Harvard and MIT did not have the opportunity to model that new agenda in their simulation.
Still, the team determined that had Reeves left his old order in place he could have contained the death toll to a range between 213 and 640 by June 15. Were he instead to have followed Kemp’s lead, the range of deaths might have spiked to between 1,865 to 3,463, assuming Mississippi businesses and patrons returned to their pre-pandemic habits.
“There is no higher priority for Governor Reeves than ensuring the health and well-being of all Mississippians,” said a spokeswoman for his office, Renae Eze, noting the virus’ present impact on the state has been substantially less severe than the worst projections. “Thanks to the strategy executed by the governor and our state health officials, our testing is robust, our numbers are low, and our curve is flattening.”
Regardless of how credible claims of flattening curves may be when testing remains so scant, the analysis performed by the Harvard and MIT team showed these same governors could have saved many of their constituents had they ordered social distancing sooner.
Had Kemp instated his shelter-in-place order on March 23 (when New York City instituted its policy) instead of the date he actually did—April 3—the analysis found his state could have seen as few as 148 COVID-19 deaths by April 24 and possibly no more than 427, far lower than the actual documented count of 871.
Likewise, the simulation projected that had DeSantis locked down Florida on March 23 instead of April 3, the tally of fatalities in his state on Friday could have fallen to somewhere between 103 and 376, rather than the actual total of 987. If Reeves had acted on the earlier date, only 36 to 111 Mississippians might have died because of the virus as of April 24, instead of 201.
Of course, the Harvard and MIT models—like all such models—has critics.
Dr. Jeffrey Klausner, an adjunct professor of epidemiology at the University of California Los Angeles who previously worked for the Centers for Disease Control and Prevention, acknowledged the analysis and similar simulations “can help policy-makers frame a response.” But he argued such projections “overinterpret the benefit of stay-at-home orders” and underestimate the impact of other factors that go into determining the infection’s reproduction number.
“It’s very difficult to input the right assumptions to get a useful outcome,” added Dr. Irwin Redlener, director of the National Center for Disaster Preparedness at Columbia University and an expert on U.S. readiness for pandemics. “The infectivity of the virus, people following these rules, containment—you don’t really know what you’re dealing with.”
Still, Redlener said, it’s too soon to reopen states without enough tests and contact-tracing to keep track of a resurgence of infections. “It’s not responsible of governors to rush into a return to business as usual, even if it’s relatively slow,“ he said. “This is a serious risk. We’re playing with fire.”
Tambe, who co-created the models with Majumder and their team, acknowledged they may not map precisely onto reality. Still, he questioned whether the other factors model detractors cited—more diligent hand-washing and mask-wearing—would improve broadly enough in the weeks ahead to have an impact comparable to government orders. And he asserted that the purpose of the simulations was less to provide flawless predictions than to inform elected leaders and health officials as they consider methods to revive sedated economies.
“We’re not saying this is the answer,” he said, acknowledging that a permanent lockdown was impracticable. “It’s one in the arsenal of tools that policymakers may employ.”
When presented with doubts about the benefits of projecting pandemic death, Ayer—the Georgia Tech modeler—responded by quoting British statistician George E. P. Box, who famously said: “All models are wrong, but some are useful.” In an absence of sufficient data to look back on—a real problem for a pandemic experts are still learning about every day—no model will be perfect, Ayer said. But a careful and meticulous one is a much better alternative for policy-makers to “having no models and relying on gut feeling.”
The idea, Ayer added, is to look at the dozens of models currently available and see where the similarities lie, what the trends are, and what is likely to happen over time, as opposed to focusing on specific numbers.
“A lot of experts have said that lifting restrictions too soon would lead to a second wave, and that’s what a lot of the research has shown,” said Ayer. “All of the modelers are using the best available evidence out there, but our understanding of the disease is evolving over time.”
Or as Majumder put it, “A model is only as good as the assumptions we put into it, and when we have a novel pandemic, our knowledge is changing every second.”
The models provided for this story were created by Jackson A. Killian, a Ph.D. student at Harvard’s John A. Paulson School of Engineering and Applied Sciences; Marie Charpignon, a Ph.D. student at MIT's Institute for Data, Systems, and Society; Bryan Wilder, a Ph.D. student at Harvard’s John A. Paulson School of Engineering and Applied Sciences; Andrew Perrault, a Postdoctoral researcher at Harvard’s Center for Research on Computation and Society; Milind Tambe, Gordon McKay Professor of Computer Science and Director of Harvard’s Center for Research on Computation and Society; and Maimuna S. Majumder, faculty at the Computational Health Informatics Program (CHIP) based out of Boston Children's Hospital and Harvard Medical School.
