Throughout this pandemic, America has been playing catch-up: By the time the U.S. reported its first coronavirus-related deaths in March, Covid-19 had been circulating inside communities for weeks. And by the time states began breaking down their numbers by race, the virus had been ravaging Black and low-income neighborhoods at a disproportionate rate.
How prepared the U.S. is to fight emerging infectious diseases like Covid-19 and recurring ones like seasonal influenza depends greatly on the government’s ability to monitor outbreaks so leaders can make timely decisions during—or even before—a public health emergency. President Barack Obama called this kind of disease monitoring “a national security imperative” in 2012 when he created the first national biosurveillance strategy.
But a new study in the journal PLOS Computational Biology suggests current systems to detect outbreaks at an early stage may have critical data gaps that leave out low-income communities who are usually hardest hit by major disease outbreaks. In particular, the researchers looked at the available data about people who show up at health facilities with influenza-like illnesses—which then helps experts estimate the prevalence of the disease—through the national influenza surveillance system. They also looked at so-called next-generation tools like and Biosense 2.0 (which aggregated data from various emergency rooms) and the now-deactivated Google Flu Trends. When the team used data from all three sources to build a model to predict influenza hospitalization rates across neighborhoods of different poverty levels, they discovered a “critical blindspot.”
“We found that the model was highly accurate for individuals who lived in neighborhoods in the upper three-fourths of income distribution, and was inaccurate for individuals who live in the lowest quartile,” says lead author Samuel V. Scarpino, an applied mathematics researcher who heads the Emergent Epidemics Lab at Northeastern University in Boston.
The researchers first divided the ZIP codes in the Dallas-Fort Worth metropolitan region (where they had access to Biosense data) into four groups according to the poverty level. Then they trained their model using a combination of clinical symptoms reports, internet searches and emergency room data recorded between 2007 and 2012, comparing its predictions to the actual number of hospitalizations by week and by year. The researchers also simulated the “real-world use” of such a model, crunching, for example, only the first 15 weeks of data to predict the hospitalization rate of the 16th week.
Their findings cast doubt on U.S. capacity to detect an uptick infections among the most socioeconomically vulnerable populations at an early enough phase to respond.
“It’s important to point out that individuals in this lowest quartile have about twice the rate of hospitalizations, so a much higher burden of disease, than do individuals in other ZIP codes,” Scarpino says. Like with many other health disparities, the disproportionate impact of flu season has
People in these communities live in more crowded neighborhoods and homes, and are more likely to have underlying medical issues
They’re also less likely to have access to affordable health care services, which in turn may have led to the under-sampling of at-risk populations in the government data that fed into Scarpino and his colleagues’ forecasting model. “They don’t have insurance or even if they do, they can’t afford the copayments or they can’t get time off work to go and seek medical care,” Scarpino says. Because the data largely comes from health care sources, Scarpino adds, “the same kind of barriers to access that in part caused the different health burdens also caused us to have gaps in our public health awareness in these most at-risk neighborhoods.”
The next-generation surveillance, according to their study, failed to fill in those gaps left behind by the national system. While crowdsourced data from social media and Google searches may give health officials information that’s closer to real-time than hospitals do, it’s still biased against low-income households who are less likely to
A small part of the study looking at data from the 2009 H1N1 pandemic also found that crowdsourced data like Google Flu Trend was not reliable for predicting anomalous outbreaks. That’s not surprising to Jennifer Nuzzo, an epidemiologist who focuses on biosurveillance at the John Hopkins Center for Health Security. “We know how valuable these types of systems—like influenza-like-illness surveillance and Google Flu Trends—are for flu than for other diseases,” says Nuzzo, who wasn’t involved in the study. “In part it’s because flu is a seasonal disease that shows up generally around the same time each year, so we get to study what people are googling for and how it relates to confirmed cases of influenza.” U.S. health experts just don’t have the same level of evidence for other diseases, which are monitored in other ways. Covid-19, for example, is best monitored through rigorous testing and contact tracing.
Still, the study could have significant implications for U.S. disease management. As Covid-19 continues to sweep the nation—with a second wave of infections expected to hit later this year—and as experts are predicting the presence of more emerging diseases in the future, antiquated data-gathering systems and continual socioeconomic barriers to health care and
Nuzzo agrees: “If you point to where there are strengths and weaknesses in existing surveillance systems, if you know if you have biases or if you’re not capturing a certain type of population, it can help point to where we can do extra work.”
While collaborating with local pharmacies and community clinics that are more likely to serve vulnerable populations may help improve data collection, Scarpino says there isn’t an easy answer for improving the accuracy of disease surveillance, especially amid steep cuts to the CDC budget under the Trump administration. “This is a much bigger problem that we’re going to have to solve around health care and access to care,” he says. “Because the data that we use for public health flows through our health care system, so if our health care system is biased against particular neighborhoods, then almost certainly our data systems will be as well.”
To contact the author of this story:
To contact the editor responsible for this story:
© 2020 Bloomberg L.P. All rights reserved. Used with permission.