Cloudy With a Chance of Covid

For Caitlin Rivers, the coronavirus pandemic is a story of missed opportunities. When reports first emerged that a new virus was causing severe pneumonia in Wuhan, China, the U.S. government could have published predictions of how the virus might spread to help citizens prepare. It could have evaluated the efficacy of social distancing, mask wearing, and travel restrictions. And it could have used models to help hospitals anticipate surges of hospitalized patients and the PPE, ventilators, and other supplies they would need to treat those infected with SARS-CoV-2. 

Early, rigorously reviewed models of the coronavirus’s spread could have produced findings that could have informed policy that might have saved lives. Instead, the pandemic turned normal life in the U.S. upside down in March. Researchers around the country did publish forecasts and predictions about how the disease might spread, how many people might need hospital beds and ventilators, and how many would die. Some of those were shared with policymakers, which helped to inform the response. But many of the forecasts contradicted each other or didn’t pan out. 

The solution, say Rivers, PhD, MPH, a senior scholar at the Johns Hopkins Center for Health Security, and biodefense expert Dylan George, PhD, is a national infectious disease forecasting center. It would have gathered forecasters under one virtual roof and allowed scientists to produce the forecasts and analyses that would help public health officials respond to new threats. Although the center Rivers and George envision would focus on infectious diseases affecting the U.S., it would rely on international and domestic surveillance data of everything from specific diseases like influenza and Zika to tracking the emergence of unusual new syndromes. This will be used for mathematical models to build their forecasts. 

“We want to create something like the National Weather Service, but for infectious disease,” Rivers says. “A forecasting center will help us make the decisions about responding to outbreaks so that they don't become epidemics or pandemics.” 

Forecasts, explains Justin Lessler, an associate professor in Epidemiology, are a specific type of prediction. A forecast makes a specific calculation about where and when a certain event might happen, based on present and past data. A prediction tends to be more general. A weather forecast, for example, will tell the public when and where a specific hurricane is likely to make landfall. A prediction will estimate whether this year’s hurricane season will be more or less active than usual.

Google offered a real-life look at the challenges of disease forecasting when it launched its Dengue Trends in 2004 and Flu Trends in 2008. The projects used aggregated internet search data on terms associated with the diseases, such as “symptoms of flu” or “treatment for high fever and joint pain,” to estimate disease transmission rates in different countries. Google then extrapolated those current data to forecast disease spread in the next few weeks. It should have been Big Data’s shining moment. Instead, largely due to flawed algorithms Google used to translate search data into forecasts, both programs failed.

In the years since, universities around the U.S. and around the world—including at the Bloomberg School—have built their own disease forecasting teams. Some, such as those at the University of Washington and Imperial College London, published influential COVID-19 predictions that spurred U.S. and U.K. and governments to enforce shutdowns because they showed the likely number of illnesses, hospitalizations, and deaths if life persisted as usual. 

“We want to create something like the National Weather Service, but for infectious disease. A forecasting center will help us make the decisions about responding to outbreaks so that they don't become epidemics or pandemics.” 

These predictions, Rivers says, could have influenced policy more quickly had the U.S. forecasters been centered in government and not academia where they were a step removed from the policymakers who can turn their calculations into action. Studies from early this fall showed that thousands of lives would have been saved if researchers had been able to convince government officials to implement physical distancing guidelines even just two to three weeks earlier. A national forecasting center can’t force a government to listen to science (or even common sense), but Rivers says it would bring existing small groups of forecasters together under one umbrella and put them in close contact with the officials who can translate the forecasts into action.

Infectious disease surveillance will enable scientists not only to identify diseases new to science (as COVID-19 was just a year ago), but also to identify where and when known diseases, such as influenza and West Nile virus, may cause outbreaks in the future. The forecasting center would then advise local, state, and federal officials about where and when to implement measures like physical distancing, mosquito control, or vaccination programs and show how these interventions could affect disease transmission. When more details on a COVID-19 vaccine become available, forecasters will be able to determine the proportion of the population that needs to be immunized and how this will alter morbidity and mortality rates.

“There isn’t going to be any single silver bullet, but we would be able to evaluate the effectiveness of various interventions that we know are on the table,” says Elizabeth Lee, PhD, a research associate in the Infectious Disease Dynamics Group. “Together, we can use these come up with a coordinated strategy to control disease transmission.”

The government has invested billions of dollars into the infrastructure needed to gather the reams of data needed to make accurate weather forecasts. But given that the U.S. has been unable or unwilling to collect even some of the most rudimentary public health data on the coronavirus, it makes Kate Grabowski, PhD ’14, ScM ’07, an assistant professor in Pathology at the School of Medicine and in Epidemiology at the Bloomberg School, is skeptical about whether a disease forecasting center will be able to amass enough information to make good predictions.

“A forecasting center is only as good as the data that goes into it,” she said.

Rivers hopes a national disease forecasting center will create a positive feedback loop, with data and forecasts feeding off each other to improve public health—something that could have made a difference in containing COVID-19. “We would have had another tool in our toolbox, had we had these capabilities in place,” says Rivers.

The best time to have created a disease forecasting center was several years ago, before COVID-19 struck, she says, “but the next best time to build a forecasting center is now.”