AI vs. The (Next) Pandemic
WHICH VIEW IS right? Maybe all of them. It’s true that AI healthcare tools did little to prevent the initial, rapid spread of COVID-19. But it’s also true that AI has been used to effectively respond to the crisis ever since. AI tools spotted the current outbreak and predicted its spread weeks before the World Health Organization (WHO) identified the novel coronavirus in Wuhan, China. AI-powered X-ray models repeatedly found success diagnosing patients with COVID-19. And more recently, AI companies were among the first to identify drugs that have, so far at least, been effective in treating the symptoms of this disease.
So whether or not AI ultimately leads directly to a breakthrough, this technology is one of society’s key tools for responding to our current crisis. It will be critical in helping the near-term fight against COVID-19—and in predicting and preparing for the next contagion down the road.
“It took what is hopefully a once-in-a-lifetime disaster to make us think about how we could have done things better, and the pandemic has proven that you need data and you
need it fast,” says Peter Pitts, president and co-founder of the Center for Medicine in the Public Interest and a former FDA associate commissioner during the George W. Bush administration. “That’s going to require public officials, industry technologists, public health experts from academia and elsewhere to realize that one of the failures in this crisis was the ability to predict. And what’s one of the best tools out there to help with the ability of prediction? Artificial intelligence. So, now, how can we get together to think about how to use AI better?”
Talking to industry analysts, technologists, health professionals, and AI-focused academics, the answer to Pitts’ question seems to boil down to four AI applications necessary for conquering the next COVID- 19-like pandemic.
SOUNDING THE EARLY ALARM
In retrospect, one of the earliest alarm bells about the emerging, novel coronavirus in China stemmed from AI—specifically, a Toronto AI company called BlueDot. The company’s Global Early Warning System spotted something unusual on December 31, 2019; two days later, similar alarms rang in the San Francisco headquarters of an AI firm called Metabiota. Both companies use these AI tools to scour the Internet for news reports and government releases that may indicate a potential public health crisis. Using natural language processing applications, both companies are constantly on the lookout for words like “coronavirus” or other terms that could indicate high-risk diseases.
Those simple AI applications allowed the firms to identify some kind of outbreak underway in Wuhan, China an entire week before the WHO reported “a cluster of pneumonia cases” in Wuhan and the preliminary determination, by Chinese scientists, of a novel coronavirus.
Of course, BlueDot and Metabiota’s tools didn’t stop analyzing data surrounding the disease upon identification. Both companies also use software that can track flights, including passenger counts, worldwide. BlueDot even tracks cell phone movement from one country to another. So as the WHO was warn- ing of an unusual cluster of pneumonia cases in Wuhan on January 9, BlueDot already had analytics available showing thousands of potentially infected people had boarded flights out of Wuhan as well as data on where these passengers traveled and moved upon arrival. Such information helped the company predict that COVID-19 might soon follow in countries including South Korea—a place that did ultimately have an outbreak soon after Wuhan locked down.
Not a lot of people had heard of BlueDot or Metabiota at the beginning of 2020, but this AI application is being taken very seriously now that the WHO and others warn that COVID-19 will not be the last viral outbreak of its kind. If AI can ultimately identify then pinpoint the potential spread of a new infectious disease, in theory cities can start to prepare for the public health consequences before a virus arrives.
That’s partially the hope behind a new AI tool developed by Stratifyd, a Charlotte AI firm that started out focused on bioterrorism tracking for the government. Called the Augmented Intelligence Platform, Stratifyd’s AI tools scan social media postings for widespread talk of illnesses then match that chatter with any official statements (i.e., those from public health organizations and governments) to spot problems disparate doctors may have communicated to each other.
“Our work,” says Stratifyd CEO Derek Wang, “is trying to understand how we can leverage news media and social media in different languages all around the world.”
The company crafted a baseline dataset that relies on identifying certain indicators of behavior based on keywords and what’s be- ing discussed in multiple places online where there may be clusters of sick people. “Maybe you have a group of people in one center city that talk about shortness of breath, or fatigue,” Wang says. “You see that on day one. On day two, you see they’ve started feeling a lot of pain or you see them say, ‘I’m sweating like a pig.’ Those key terms and phrases automatically bubble out and tell you something is going on.”
If this AI system can prove successful at this pattern recognition, it’d give Stratifyd notice of groups of sick people potentially before—or at least in conjunction with—the time health- care workers in one region realize they have common patient problems. “We take this from a data-driven, AI-driven method,” Wang says. “That way we can match what’s being said to what’s being found in laboratory pathogen results and see how global transmission of certain pathogens may be happening—hopefully ahead of time.”
FINDING NEW PATHS TO TREATMENT AND RECOVERY
Optum is a pharmacy benefits manager and health services company based in Eden Prairie, Minnesota. The work may sound dense and, well, it is. The company uses AI to automate some of the “very, very unsexy, but also really complex” parts of the healthcare business, says Sanji Fernando, a senior vice president at Optum who heads up the company’s AI and analytics teams. Since COVID-19 began spreading around the world, Fernando has also been putting Optum machines to work on finding solutions to the problem currently vexing most of the world.
COVID-19, unsurprisingly, has proven to be highly complex. “We don’t know what the answer is to the pandemic,” Fernan- do says. “It is what you’d call in AI an ‘unsupervised problem.’ You can model the spread of the disease. But those models rely on prior assumptions, and in a changing landscape, your assumptions can change at a very rapid pace.”
The longer COVID-19 carries on, however, those assumptions give way to concrete data. And with the data coming in, Fernando says Optum has already found good news. The company believes it can use the information already collected about patients worldwide to build automated, AI-driven treatment models pertaining to future patients. The idea: By using AI to collect worldwide data on any measures that helped people recover from COVID-19, healthcare providers will then be able to create blueprints for successful treatment regimens. That kind of AI deployment could mean having effective, lifesaving treatment options that precede pharmaceutical treatments.
“I don’t think we’ll have to wait 18 months or two years for a vaccine,” Fernando says. “Given the way the dataset is grow- ing, we might be able to train models now for use in the fall.”
Optum is far from the only organization leveraging AI in the hopes of identifying treatments that could save lives and minimize the health impact of COVID-19. Several researchers have already used machine learning to diagnose COVID-19’s presence within seconds of digitally reviewing lung CT scans. Others have analyzed the ribonucleic acid sequence of COVID-19 in hopes of finding drug combinations that can treat the disease
ahead of a vaccine. And in mid-March, South Korea-based Deargen had an AI model specifically suggest the FDA-approved antiviral drug remdesivir could work in treatment—a month before Dr. Anthony Fauci shouted out a study on the drug as “quite good news.”
MAKING LOCKDOWNS AS EFFICIENT AS POSSIBLE
Most stories on AI’s role in pandemic planning focus on the disease itself (identifying or treat- ing it, for instance). But clearly a pandemic’s impact goes beyond health. If COVID-19 lingers for years not months, AI could also be key in helping to keep workers on the job even as outbreaks persist around the country.
Tom Davenport, a professor of IT and management and AI expert at Babson College in Massachusetts, envisions machine-learning tools that could create risk profiles for individuals who might be exposed to the disease. Those profiles would match things like a person’s age, current and past medical conditions, and health conditions of their household partners to all known risk factors for serious complications or death from COVID-19.
AI in essence could then produce millions of customized risk profiles for people, which could better inform companies weighing whether to reopen a physical space. And if the state of testing in the U.S. remains incomplete, a data-driven approach gauging risk might be the best resource society can hope for in a timely manner.
“We could then have some people distance themselves because they’re highly likely to get a disease and other people keep working or keep dining in restaurants, and so on, because they are much less likely to become ill,” Davenport says. “You could combine that with the use of existing wearable devices that can alert you when you’ve come in contact with someone else who has the disease. There’ll be a huge amount of data in all that, so you absolutely need AI.”
There’s one big catch to using AI to leverage all that health data: Most health data in the U.S. and Europe is private. And for now, polls show that Americans like their health records being closely guarded. Still, Stratifyd CEO Wang believes that in crisis moments, many will choose to opt into apps or anything else that could help them stay healthy and informed about the spread of a potentially deadly virus.
“For our company,” he says, “we don’t need your name, we don’t need your demographic information… we don’t need your anything except what you voluntarily give. And for me, personally, right now, if you tell me my iPhone is going to tell me whether I have been in contact with another potential patient, I don’t think I have any choice but to opt in.”
AI FOR BETTER DATA, SO DATA CAN MAKE BETTER POLICY (AND AI)
The numbers were grim, but they weren’t the same. On May 1st, CNBC reported that the WHO’s data on COVID-19-related deaths in the U.S. had hit a record one-day high of 2,909 people. That same day, The Washington Post, which bases its count in part on data from Johns Hopkins University, reported 1,723 COVID-19 deaths.
That two estimates could be so far apart shows another pandemic challenge—there’s a data problem that impacts public health when illnesses like SARS, Ebola, and COVID-19 spread. If AI tools are going to learn from vast amounts of public health data and start generating solutions to big problems, centralized and accessible datasets—lots of them—are important.
“We don’t have a current data collection agency that the whole country can use and analyze the data from,” Davenport says. “If you try and figure out how many tests there are in New York on a particular day, you’ll get various different numbers. It would be nice if we had some sort of common countrywide approach to gathering data. The countries that have handled this crisis best—South Korea and Singapore and so on—do have that.”
Some of this data discrepancy simply stems from the age of our core reporting systems. Partly out of understandable concerns for privacy, healthcare firms have for a long time kept medical records in ways that are now hard for AI tools to access. “Data is definitely a major limitation for AI,” says Lian Jye Su, an analyst at ABI Research. “In the past, healthcare institutions utilized handwritten records. This is not useful when it comes to the training of machine learning-based AI.”
Bridging such data gaps is likely to mostly be a matter of pushing policymakers and regulators at the federal levels to fund IT upgrades in critical industries and to consider changing rules that limit what can and can’t be accessed. And the aftermath of a pandemic—when every- one will be looking for new solutions to head off some kind of recurrence—may be an excellent time for AI companies to lobby for this change in Washington.
“AI has to be a game changer because we’re not becoming less connected globally, we’re becoming more connected,” says Peter Pitts, the former FDA associate commissioner during the Bush Administration. “Maybe there should be an AI caucus in Congress. Maybe a presidential task force on artificial intelligence and pandemics. Maybe a presidential AI czar. AI is what we need to be talking about now.”