How do you track the spread of COVID-19 in real time?
That's the question Instagram co-founders Kevin Systrom and Mike Krieger set out to answer when they created RT.live, a website that tracks the rate of infection in all 50 states.
They said it's meant to be an at-a-glance score for evaluating our progress in the fight against coronavirus.
"The goal for us was, what if we tried to boil it down to a single metric that tells you how things are going now," Systrom said. "Forget about forecasting for a second. What's happening right now?"
Systrom and Krieger are just two of the many Silicon Valley entrepreneurs who have been trying to create new tools to help track the spread of COVID-19, which as of Thursday has killed roughly 50,000 people in the United States and left 26 million without jobs.
Apple and Google have partnered to build a system for smartphone-based contract tracing; Facebook last week released county-by-county maps showing where users have voluntarily self-reported COVID-19 symptoms.
RT.live centers on an all-important number: One.
"This number is important because if I infect more than one person, that means the epidemic grows exponentially," Systrom said. "But if I infect less than one person, the virus actually quickly peters out."
"So all we're trying to do is say, look at this number. ... If it's above one, that's bad news. If it's below one, we're doing a good job."
Epidemiologists generally use a unit of measure known as R0 (pronounced "R-naught"), or the basic reproduction rate of a virus. R0 is a measure of how many other people, on average, a single contagious person goes on to infect. But that's a hard number to directly find.
"The problem is measuring it," Systrom said. "Because you can't go around and easily know exactly how many people someone infects. So what you have to do is take a look at the data coming in and infer what that number is."
And so Systrom and Krieger decided to instead go after Rt, or the effective reproduction rate of the virus — the rate in real-time that one person seems to be infecting other people in a given location. The number is broadly calculated using the reported number of new infections and comparing them to past reports.
But the lack of testing capacity, delays in test results and other factors make that data by itself unreliable. So the duo drew on Bayesian statistics, used for generations to investigate the probability of outcomes under changing conditions, and drew from a Bayesian algorithm built to predict the spread of disease in a 2008 paper by a pair of Los Alamos National Laboratory researchers. But while that algorithm only compared today's rate to yesterday's, Systrom drew his result from the past seven days, which he said seems to sharpen the results.
At a broad state level, Rt can be misleading, Systrom said. In California, which Systrom describes as "a tale of two different regions doing two very different things," the Rt suggests Californians aren't doing enough to stop the virus. But that doesn't reflect the reality in specific cities.
"Los Angeles and Riverside are actually growing fairly quickly. The Bay Area is not growing quickly. But when you blend those together, you get an average picture that looks like it's picking up slightly," he said. "So if you go to our site, you'll see that California has an R above one. But that may not feel like what is happening locally."
As of late April, RT.live's data suggests that the rate of infection is trending down in states like New York and Virginia and trending up in states like California and Connecticut. The idea is that over time, as data is added to the model, the Rt will become more and more accurate.
Systrom and Krieger said their next goal is to get more specific data on a county-by-county, metro-by-metro level.
Tracking an infectious disease was not what Systrom and Krieger thought they'd be doing when they stepped down from their roles at Instagram in 2018, six years after the photo-sharing site was acquired by Facebook.
"It's an interesting change from Instagram, where you're working sort of within your own company, keeping everything secret," Krieger said.
Krieger said their operation is now able to make its work public, which means collaboration with statisticians and epidemiologists who have helped improve their system.
And what have their efforts and the available data taught them about COVID-19 and the path toward recovery?
"I think that most people think about this as ... a one way door. Like you open the door, you go through this process and you're on the other side," Systrom said. "I think the truth of the matter is that it's way more complicated. It's way more like a revolving door. Where you go in, and it's possible you go through to the other side, but it's also possible you go around once or twice."
In other words, it's possible we may curb the disease for a time, or in some counties or states, then find ourselves back in a scenario similar to the one we're in today.
"You're going to have some states or some municipalities who get through this and get on the other side and start to open back up, only to become infected from another area where someone travels from, because they haven't gotten through to the other side," Systrom said.
"It's this dynamic nonlinear system that makes this so hard to predict," Systrom said. "So I am hopeful that we've done a good job in terms of lowering our T-values over time. But I think it's going to be difficult going forward. And I think that's why we need something like this."