SAN FRANCISCO — As the new coronavirus spreads, health tech startups with medical chatbots are scrambling to update their algorithms to screen feverish and coughing Americans and advise whether they should be evaluated for infection with the virus.
So far, these artificial intelligence-powered chatbots are turning up lots of people with the flu. That’s unsurprising at this time of year. It speaks to the small number of coronavirus cases in the U.S. — and how hard it may ultimately be for AI systems to differentiate among the myriad pathogens that cause the same flu-like symptoms that a mild case of the new virus appears to cause.
The apps don’t appear to have been involved in turning up any patients who actually have the 2019-nCoV coronavirus — there have been 11 confirmed cases in the U.S. to date — nor is it clear whether any patients flagged by the chatbots have even proceeded to undergo lab testing. One company, 98point6, said it has contacted the Centers for Disease Control and Prevention about two suspected cases it turned up, but neither of those patients were ultimately isolated or referred to get testing for the new virus.
Still, the companies say, they’re offering people reassurance and helping them get treatment — while keeping them out of the waiting rooms of overwhelmed clinics and urgent care centers. And if the coronavirus spreads more widely in the U.S., these companies could be poised to play a more prominent role in triaging suspected cases.
The AI systems — which are part of a burgeoning industry meant to offer patients remote care and automate rote aspects of a traditional clinical visit — generally interview patients via text message, asking them questions about their symptoms. When patients report flu-like symptoms such as cough, shortness of breath, and fever, the chatbots are now asking pointed follow-up questions related to the coronavirus, which originated in China and has infected more than 24,000 people there. Those questions include: Have you traveled to China recently? Or have you been around someone who has?
The companies say the updates to their algorithms are based on guidance that the CDC has issued to health care professionals about the criteria that should prompt a patient to be evaluated for infection with the virus. That guidance is online, and patients who are worried about possibly being infected can easily access it. So why, then, are these chatbots necessary?
“We think we can really help direct people to the right care at the right time and give them quality information as it relates to this problem,” said Dr. Brad Younggren, chief medical officer at 98point6. He added: “We can keep them from going into an environment to be tested, when maybe testing isn’t necessary.”
The app-based coronavirus screenings being promoted by health tech companies supplement the in-person screenings being conducted in airports, clinics, and emergency rooms across the country.
Only in 260 cases to date have U.S. patients been escalated to get lab testing through the CDC — until Tuesday, the only way to get an infection confirmed in the U.S.
When patients have met criteria for getting tested, local physicians have taken a sample — involving nasal and throat specimens, as well as the patient’s blood serum — and shipped it to the CDC’s headquarters in Atlanta. On Tuesday, the Food and Drug Administration cleared the CDC’s test for wider use, and the CDC plans on quickly rolling it out to state health departments and international partners.
Seattle-based 98point6 offers virtual primary care visits via its apps to patients in the U.S. Patients begin chatting with an AI before getting handed off to a physician who continues the conversation via text message. Patients can pay out of pocket for the visit — annual subscriptions start at $20, plus $1 for each encounter — or their employer or health plan may cover the cost.
98point6 went live with its coronavirus screening on Jan. 24 — just a few days after the first U.S. coronavirus case was confirmed, in a Washington man who had recently traveled to Wuhan, China, where the outbreak originated. The startup has updated its algorithm several times since then, Younggren said. An update last week added fever to the list of patient-reported symptoms that triggers questioning about travel history to China.
In the two instances when a 98point6 physician has contacted the CDC about suspected coronavirus cases, the agency advised that those patients did not need to be tested or isolated. In each case, the physician then telephoned the patient to let them know they weren’t at risk.
Bright.md, a Portland, Ore., startup that sells care automation software to health systems, went live with its own coronavirus screening on Jan. 29. Its product uses AI to conduct remote interviews with patients.
“We felt like that was an important aspect of helping our delivery system partners manage both risk and exposure — but also public concern,” said Dr. Ray Costantini, Bright.md’s CEO. So far, he said, that’s mostly meant “making sure that all of those patients who do have flu symptoms get treated appropriately and effectively and quickly to minimize exposure for the more common condition that really is a much bigger risk to population health and people’s well-being.”
When Bright.md’s AI interviews raise a red flag for a possible coronavirus case, the software automatically arranges for the patient to have a video encounter with a physician at the health system. The 15 health systems in the U.S. and Canada that use Bright.md’s software sometimes let their patients use the system for no additional cost, or charge them a fee of up to $35 per visit.
In the U.S., fear about the coronavirus may be spreading faster than the virus itself.
So when Buoy Health, a Boston-based startup that offers an AI-powered symptom checker, added a coronavirus tool to its platform on Tuesday, it did so with the goal of correcting misinformation, said Andrew Le, Buoy’s CEO and co-founder.
After users enter symptoms into the latest version of the Buoy app, it asks if there’s a specific diagnosis they’re worried about. As soon as a user starts typing “coron…,” for example, the word “coronavirus” appears. Tapping the word prompts the app to ask about relevant symptoms and risk factors, all of which Buoy took from the CDC guidelines.
Depending on their answers, users will get directed to one of two paths: high or low risk for coronavirus. High-risk consumers are advised to head to an emergency room — and are told to call beforehand to give the facility the chance to prepare.
“We see patients every three seconds,” said Le. “We have the potential here to help identify outbreaks before people get to the hospital.”
As part of its more long-term efforts to address coronavirus, Buoy is also partnering with a HealthMap — a disease detection project run by researchers from Boston Children’s Hospital — to help map any potential clusters of the new illness. If Buoy and HealthMap were to document a grouping of coronavirus-like symptoms in one location, for example, they could alert public health officials and hospitals.
“We’re trying to match what’s happening in the world with what’s happening on the ground,” Le said.