Health care workers have found a new use for artificial intelligence in the fight against COVID-19: a chatbot that advises patients whether they need testing.
Providence Health & Services, a nonprofit that operates over 50 hospitals and other medical facilities, has been using an A.I.-powered chatbot named Grace to help handle the huge influx of people seeking medical advice. Those people can use the chatbot from a computer at home to answer a series of questions about whether they’ve traveled out of the country, which symptoms they’re experiencing, and whether they’ve come in contact with others who have COVID-19.
Based on the responses, the chat app can tell people if they should be tested. It can also schedule them for a virtual visit with a clinician, among other things.
Although the chat app works like a basic telephone tree, in which a person’s response triggers a preprogrammed action, it relies on some underlying A.I. technologies to perform some of its more complex tasks.
Maryam Gholami, Providence’s product head for design and software engineering, said her company uses Microsoft’s natural-language processing technology, or NLP, so that it can better understand the nuances of communication. In recent years, A.I. researchers have combined advances in deep learning with NLP techniques to create language systems that can map the relationships between words and even answer trivia questions. Understanding the relationships between words is important because words often have multiple meanings, depending on the context of a given situation.
Because of the advanced NLP technologies, Gholami said, the Grace app doesn’t actually think someone saying “my back is killing me” means that they’re being murdered. Instead, it understands that the person has back pain.
Providence built the Grace app a few years ago to help connect patients with symptoms like coughing to an appropriate clinician, said Providence chief digital officer Aaron Martin. But with the coronavirus bringing in more prospective patients, Providence tweaked the Grace app to handle coronavirus-related inquires.
It “was just not practical” to have already overworked nurse practitioners handling every coronavirus inquiry, Martin said. During March and April, in particular, a huge number of people contacted Providence, and 200,000 of them used the Grace app, Gholami said.
Despite advances in A.I. technologies to make better chatbots, there is still some room for improvement.
For instance, A.I. researchers have difficulty explaining how deep learning systems make decisions, which makes health care professionals nervous. That’s one reason why Providence had originally decided that its Grace app would help people seeking care for ailments like colds or flus rather than serious illnesses.
“Start with something kind of boring, but is low risk and high volume,” Martin said. “Don’t start with cancer.”
Other health care firms are also considering using A.I. to help with the task of triaging. The Mayo Clinic, for instance, recently partnered with health technology startup Diagnostic Robotics on a similar triaging system that is expected to debut at Mayo’s Rochester, Minn., facility next month.
Diagnostic Robotics CEO Yonatan Amir said patients entering the Mayo Clinic’s facility will interact with the company’s software, respond to a series of questions, and then be triaged to the relevant physician or department. Similar to Providence’s triage app, Diagnostic Robotics’ app is designed to “automate low-hanging fruit” and “not something that’s a life-threatening event.”
Patients with diabetes, for instance, will be able to use the startup’s app at on tablet computers at the Mayo Clinic to help them arrange blood or urine tests, Amir said. The app is not intended for people experiencing “severe chest pain that radiates to the left arm for half an hour” or other symptoms that could likely indicate life-or-death medical situations.
“We are not taking any chances,” Amir said.
Stanford University physicians are also testing A.I.-powered triage systems. In April, for instance, Stanford physician Ron Li explained at an online conference how a team is experimenting with a machine-learning system to help doctors know whether some COVID-19 patients need intensive care.
The machine-learning system analyzes the medical data of the patients and then ranks them based on how sick they are. With this information, doctors can better track the severity of illness in certain patients and whether they should be rushed to intensive care.
Li told medical news publication Stat that it will still be up to doctors to “make the call regarding whether or not the patient needs to go to the ICU or get intubated.”
“We’re essentially waiting as we get more and more COVID patients to see how well this works,” Li said.
