Team converts a flu early warning system to track COVID-19

cc-blog-5.7-banner-1

Every now and then, a problem surfaces that some person or group is uniquely suited to solve. They are prepared. They have the expertise. They have the equipment. That’s what it was like for the Optum Flu Forecast team when COVID-19 hit.

Almost immediately there was demand for a COVID-19 version of the Flu Forecast, a data-driven early warning system that debuted last year. The challenge for Paul Nielsen, Optum vice president of strategic programs, and his team: to create a new COVID-19 forecasting model with the original Flu Forecast as its foundation … and do it fast.

“I told the team, ‘We need to start building our COVID surveillance now,’” recalls Nielsen.

The result is the Optum COVID dashboard. Using private and public data, the artificial intelligence (AI) tool uses daily data on diagnoses, hospital admissions and a number of other metrics to predict when and where COVID-19 may strike next.1

Real-time access to health data helps providers and health plan executives visualize vital trends and quality measures.

Right now, the model, created at the end of March, is designed to predict COVID-19 infections five days in advance. That accuracy and lead time will grow as the system becomes smarter.

“We’re planning to have a 14-day forecast available in the next two weeks,” says Nielsen. “We access real-time data and learn from the real-time data as it’s coming in. Is it spreading? Where?”

Machine learning algorithms process millions of data points to provide geographic-specific predictions of cumulative confirmed cases of COVID-19. The shaded area represents the level of uncertainty in the prediction, which is influenced by the quantity and quality of data available.

Models rely on machine learning

Instead of manually adjusting variables, the Optum COVID dashboard — and its predecessor, Flu Forecast — both use a type of AI called machine learning. That means the model considers all available data to form its predictions, not just a select few variables that are adjusted by human experts.

Nielsen compares machine learning to how humans change our minds based on what we’ve read. “You might’ve had one opinion the day before, but then you read more about that subject and that opinion changes, because now you have more information,” says Nielsen. “Machine learning does that in an automated way. It’s an AI that’s learning, literally.”

So, what does the model “read” to learn and make accurate predictions? All sorts of data from many different sources, including information collected by the influenza division at the Centers for Disease Control and Prevention as part of its ILINet, the Influenza-Like Illness Network.

Nielsen points to the power of medical and pharmacy claims, with their demographic, geographic and diagnostic information: “The massive amount of data we can access and pull from gives us the opportunity to have an extended forecast.”

It’s not a question of which method of modeling — AI-driven or traditional — is better. They both serve important, complementary roles.

It’s very personal”

One of the first tasks facing the Optum COVID dashboard team was to de-identify data. That meant stripping it of any identifying features, so everybody’s private information stays private. It’s a time-consuming process, but the extra help compressed the timeline.

“Everybody was working seven days a week,” says Nielsen, who recalls daily conference calls that would last for hours, even on weekends. “We did something that should have taken anywhere from three to six months in roughly three, four weeks.”

Thanks to the team’s hard work, this new tool, with its dashboard and multiple data streams, can be used to make smarter, more informed decisions that ultimately help patients, providers and members.

When the team was working around the clock to finish the dashboard, Nielsen says they had a simple motivation: humanity. “We’re all consumers of health care. We all can be affected by COVID. We all have families,” says Nielsen. “It’s very personal.”

Click here to learn more about applications of AI in health care.

Read more stories.

____________________________________

1UnitedHealth Group Q1 2020 Earnings Report, accessed 5/6/2020. https://www.unitedhealthgroup.com/content/dam/UHG/PDF/investors/2020/UNH-Q1-2020-Release.pdf

One thought on “Team converts a flu early warning system to track COVID-19

Leave a Reply