Half a nation away, a Mayo Clinic mainframe is collecting the location data from those requests, then incorporating it into a complex formula for predicting future case counts in that coughing person’s county.
Welcome to the quest for a nationwide COVID-19 crystal ball, version 2.0.
The new Mayo Clinic tracking website in question has no unique, catchy name. But if its claims to accuracy bear out, the health system’s new COVID-19 forecasting tool has the potential to leap past all other dashboards hoping to connect consumers with the best information about the virus.
That’s because the new website at mayoclinic.org/coronavirus-covid-19 — see, it wasn’t catchy — does more than rewrite the latest reports out of the CDC, or even react to events in the news.
Instead, it taps into the clinic’s locally predictive mathematical modelling software for the virus, then applies those calculations to the latest testing data from every county across the nation.
On top of all that, it packages its predictions all alongside the latest in Mayo-vetted information about the virus. If you haven’t noticed, getting trusted information about COVID-19 has been no socially distanced walk in the park.
The website is getting a soft launch — it currently offers a state-by-state map showing total and new cases for every county in the country, but nothing about the cases expected there next week. When its signature feature is added in the coming days, however, the clinic’s COVID-19 tracking tool will be the first of its kind to predict nationwide case counts on the county level.
Which is after all, how COVID-19 outbreaks ultimately manifest themselves. You don’t talk about an outbreak in Minnesota, you talk about an outbreak in Blue Earth County, or Stearns County, or Dakota County.
How does it do all this? By utilizing a machine-learning algorithm that reacts to test results over time, a system Mayo designed to prepare its own hospitals for the demands they could soon expect.
As software it is “agnostic,” as Chief Value Officer Dr. Henry Ting puts it, about mask orders, school closures, and all the other public health initiatives that take up so much of our attention.
“Our model doesn’t try to make assumptions about the effects of reopening schools, rallies, protests or football games,” Ting says. “You can open schools very differently, and there are a hundred ways to do it. If you are going to use that to project the future, you’re probably going to be wrong, so we don’t use that information.”
“It really looks at what’s happened in the last four months versus what’s happened in the last four days. It doesn’t matter if it’s a school event or a motorcycle rally, our model picks that up. It’s not going to say just because you reopened schools, we’re going to increase cases.”
In addition to how many cases and tests have been identified in a given area by population, the dashboard uses publicly-available, anonymous Google search and social-mobility data from cellphones and Facebook, channeling it all through artificial intelligence capable of daily resetting of the mathematical rules for prediction.
“Most mathematical models put in these parameters and make assumptions that stay constant,” says Ting. “We’re allowing these parameters like testing positivity rate and case doubling time to change on a daily basis to predict the future.”
Armed with data showing how cases have appeared geographically in the past related to the present, it predicts what travelers can expect for, say, Pocatello, Idaho, next week.
“We’ve learned that our model is extremely accurate,” said Ting. “It was able to predict a rise in cases we saw in Florida and Arizona several weeks in advance, allowing our medical practice leadership some warning that there was going to be a surge in hospitalizations, then a peak, plateau and drop.”
Ting said Mayo was able to see from the model that cases would not overwhelm the health system in Rochester, and this allowed the clinic to resume procedures that had been put on hold. During a pandemic filled with an excess of bad information about the virus, with flare-ups that seem to rise out of nowhere, health officials say it can’t come soon enough.
“We certainly are grateful for the data Mayo has been sharing with us and glad they are making this public as well,” said state commissioner of health Jan Malcom during a press call on Thursday, Sept. 17. “We see great value in the real-time, granular level data it offers. … We think the more real-time information we can put out for the public to understand the risks, the better.”
The dashboard will present the severity of a local outbreak in terms of new cases per day per 100,000 in population. Ting said counties with less than 10 cases per 100,000 on a given day “reflects that the community has done all the right things in terms of the safety protocols to protect themselves.”
Counties with 10-20 cases per 100,000, however, are ranked as an intermediate risk, while counties showing case rates above 20, he says, “mark the beginnings of a hot spot.”
“A consumer can use that to number one, to help understand if the behaviors of the society have been effective,” says Ting, “and number two, decide if there’s an area …you may want to avoid in terms of driving or flying to visit.”
All because someone asked Google where they could find a thermometer.