Chatbots gaining ground in healthcare to boost access, hospital workforce

You’re sitting at your desk, when your eye begins to itch. At first, you ignore it—but after a few hours, maybe even a day, you wonder if you should seek care. But where do you go for that care? A telemedicine visit, a primary-care clinic, a retail clinic, an urgent-care center or an emergency department?

Consumerism has been credited with increasing patients’ options, with more health systems setting up same-day services. Yet for patients unfamiliar with the proliferation of options, the growing healthcare landscape can feel expansive, and confusing.

One way some select patients on the West Coast can navigate that system is to ask Grace.

Grace, a chatbot developed by Providence St. Joseph Health, based in Renton, Wash., takes the form of a pop-up chat window on certain health system websites. She asks patients about their symptoms and location, and recommends where they should visit—such as a telemedicine visit, nearby retail clinic or scheduling an at-home visit, depending on the patient’s needs and availability of providers.

As one of her first questions, Grace asks: “What symptom or condition is bothering you?”

“My eye itches,” you respond, in the case of the patient above.

“I understand that you’re asking about eye itch, rashes, itching. Is that right?” she confirms. After a continued quick back-and-forth, you decide to seek care at a nearby retail clinic, with Grace helping to schedule the visit.

Programs like these, which rely on personified computer programs to help patients navigate the complicated healthcare maze, are becoming more common, particularly as health systems look toward other industries to bolster their consumer-focused offerings.

“Consumers are getting used to going to travel websites, or to do their banking, and they’re able to do either a live chat or have their questions answered through a chatbot,” said Maryam Gholami, Providence St. Joseph’s head of digital products.

Chatbots, which are essentially computer programs that can text or instant message a human user in plain English, can be conversational—using artificial-intelligence techniques like natural language processing—or scripted, and can reside on platforms that are web-, text- or app-based.

Health systems throughout the nation are beginning to use them “across the entire customer life cycle,” according Brian Kalis, Accenture’s managing director of digital health, though most are using them as part of efforts to improve patients’ access to care.

“We’re still in the early-adopter stage of implementing chatbots within health systems, however, we’ve started to see that momentum pick up over the past 12 to 24 months,” Kalis said.

And while research is still emerging on the technology, it looks like consumers are interested.

Sixty-one percent of healthcare consumers said they would likely use an AI health assistant to help them estimate healthcare costs, schedule appointments or explain coverage, according to a survey Accenture released last year. And most consumers highlighted the convenience factor of AI health services, saying they would be most likely to use them to get information after hours.

While efforts to bolster access to care, such as using Grace to help guide patients to the appropriate setting, are among the most common chatbot uses today, early adopters are also uncovering efficiencies by applying chatbots in the emergency department, weight-management programs and even as part of human resources—and that’s just for a start.

Providence St. Joseph began offering a pilot version of Grace last year to help patients navigate the often confusing landscape of same-day care options. Historically, patients who felt sick would visit a primary-care physician, or, for more serious or pressing concerns, the ED, said Sara Vaezy, the health system’s chief digital strategy and business development officer.

But that’s changed with growth with the availability of urgent care, retail clinics and digital health—and while “we’ve stood up all these different access points, which is great … we’re now potentially inadvertently confusing (patients), because they’re still getting familiar with all of these different access points,” she added.

The health system began rolling out Grace on its Express Care websites in June of last year, though only some people are able to access the service during testing. Express Care is Providence St. Joseph’s same-day care service, which includes retail clinics and telemedicine visits.

So far, 2% of those who had access to the chatbot have used it to navigate same-day care services, and the health system has seen more than 90% accuracy in its recommendations. The health system, which is still testing Grace, plans to roll it out systemwide by the end of next year.

And Providence is in good company.

Advocate Aurora Health, the combined system created by Aurora Health Care and Advocate Health Care’s merger last year, is also using a chatbot to help guide patients to the appropriate care setting.

The health system’s symptom-checker chatbot, which doesn’t have a name, assesses a patient’s symptoms in detail by using protocols based on Schmitt-Thompson Clinical Content, the guidelines used by many medical call centers to triage patients. For a patient with an itchy eye, for example, the chatbot would ask about the severity and longevity of the issue, as well as characteristics of the patient’s other symptoms and conditions before suggesting where they should seek care.

“While it’s a fully autonomous machine, it’s using a decision tree that’s been well-proven in the clinical triage space,” said Jamey Shiels, system vice president for digital experience at Advocate Aurora.

Advocate Aurora developed the symptom checker in partnership with Microsoft Corp. using a cloud service the tech giant unveiled in 2017 to help healthcare organizations build and deploy industry-specific, HIPAA-compliant chatbots. Today, a beta version of the chatbot that the health system developed through the partnership is available on Aurora Health Care’s website, with plans to launch it on Advocate Health Care’s site in the future, Shiels said.

During a three-month window earlier this year, more than 2,000 people used the symptom checker, 35% of whom ultimately clicked on a recommendation from the program, according to Advocate Aurora data. The most common recommendation by the chatbot was urgent care, followed by a same-day or next-day appointment with a primary-care physician.

While health systems declined to share cost data on their chatbot projects, Buoy Health—a company that offers a symptom-checker chatbot and connects users with nearby providers—estimated that its service reduces healthcare costs by about $174 per use. That’s based on asking what level of care a user initially believes they’re seeking, and comparing that to what they say after receiving a recommendation, said Dr. Andrew Le, the startup’s co-founder and CEO.

Banner Health, a system based in Phoenix, had a problem in its EDs. It was pretty common—patients reported feeling that their wait was too long or they were unclear about when they would see a physician. Overall, it seemed that “the customer felt a little bit lost in our emergency care experience,” said Jeff Johnson, Banner’s vice president of innovation and digital business.

“Our patient and consumer insights data were showing that while (patients) were very satisfied with the care, the overall experience was not exemplary,” he added.

So this past spring, the health system rolled out a chatbot in EDs at three of its hospitals, following six months of pilot testing. Now patients who check in at the ED receive a text message inviting them to engage with a web-based chatbot. The chatbot explains the triage process and provides updates on the patient’s visit, such as alerting them when their care team places a lab order.

“It’s taking the patient through that entire experience of the emergency room and never letting them feel left in the cold, or abandoned or forgotten about,” Johnson said.

Unlike projects such as Providence St. Joseph’s Grace or Advocate Aurora’s symptom checker, Banner’s ED chatbot is scripted. That means it can respond to specific prompts—answering questions like whether the patient can leave the waiting room or where they can buy a snack—but it’s not able to synthesize or respond to messages with open-ended questions from a patient, such as, “Should I stick around for an itchy eye?”

But since the chatbot has gone live in Banner’s EDs, it’s already helped to improve patient satisfaction, according to Johnson. He said the health system’s net promoter scores, which aim to gauge patient satisfaction, show a more than 30% increase for emergency care experience after deploying the chatbot in an ED.

Banner also observed a correlation between use of the chatbot and lower numbers of patients who left the ED without treatment in its pilot, Johnson said. However, he emphasized the health system doesn’t yet have enough data to form a conclusion on that link.

Based on the success Banner has seen in improved patient satisfaction so far, he said the system plans to deploy the chatbot in EDs systemwide by year-end.

The chatbot’s success has also encouraged Banner to expand its chatbot strategy.

The health system is now planning to design chatbots for at least 10 more use cases, including ones that guide patients through appointment scheduling and pre- and post-surgical care. For those projects, Banner is working with LifeLink, the same vendor it used for the ED chatbot.

“The bot may not work for everything, though we’re certainly finding that it works in a lot of cases,” Johnson said.

An observation that patients who are seen by caregivers more frequently have better outcomes led one Florida doctor to look at adopting chatbots. Dr. Lloyd Werk, chief of general pediatrics at Nemours Children’s Health System in Florida, directs the system’s Healthy Choices Clinic in Orlando—a multidisciplinary weight-management program for pediatric patients who suffer from obesity. 

The U.S. Preventive Services Task Force in 2017 released recommendations noting pediatric patients with obesity demonstrated greater weight loss when enrolled in behavioral interventions with at least 52 contact hours. That would mean roughly weekly engagement for a yearlong program like the one at the Healthy Choices Clinic.

But most families can’t visit the clinic that frequently due to travel time and school schedules, among other challenges, according to Werk. So the clinic began looking into new strategies to supplement monthly clinic visits, such as routine telemedicine visits and, more recently, a chatbot.

“Patients that fall off the map are less apt to achieve their goals,” Werk said. “We’re looking at the chatbot to help bring us to the last mile.”

At the Healthy Choices Clinic, that chatbot is Tess.

Tess was developed by a startup in California, X2AI, which created it as a conversational, mental health chatbot. X2AI collaborated with Werk’s team over the course of 18 to 24 months to customize it for the Healthy Choices Clinic’s weight-management needs.

The weight-management version of Tess encourages patients to keep up with goals they’ve discussed during visits at the clinic. If a patient says they want to be a better athlete, for example, Tess might help them track goals like walking after school or cutting down on fast food. And if a patient has trouble reaching a goal, Tess asks what barriers they’re facing, such as feeling too busy, to help them problem-solve, Werk said.

Tess hasn’t been fully deployed at the clinic. Werk’s team recently wrapped up a feasibility study on the program, published in Translational Behavioral Medicine in May, and the chatbot is now on hold at the clinic while they seek funding for additional studies. 

That speaks to one of the reasons chatbots are less common in clinical care today, as developers need to engage in time- and resource-intensive feasibility studies and clinical trials, not to mention approval from institutional review boards.

Still, early research is suggesting some successes.

In the study of 23 adolescent patients at the Healthy Choices Clinic, patients had nearly 300 conversations with Tess over the course of 10 to 12 weeks. Patients indicated 96% of their interactions with Tess were helpful, and researchers estimated Tess provided $8,933 in savings, given that reading and responding to the 4,123 text messages included in those conversations would have taken up to 137 staff hours.

“We need future research to see what parts of the chatbot are most effective,” Werk said. “Is it the frequency of interactions? Is it having interactions after hours? … I think that we’ll figure that out over the next few years.”

Patients aren’t the only ones excited by these automated consumer experiences. An unexpected, yet successful, area that Houston Methodist has deployed chatbots in is human resources, where a chatbot named Mia helps the health system’s talent acquisition team with hiring experienced nurses.

“We’re expanding so rapidly as people move to Houston, move to Texas,” said Thomas Vernon, Houston Methodist’s vice president of talent management. “Our hospitals are full, we’re building towers in just about every one of our locations. Our hiring needs are continuing to be a steep upward curve.”

Houston Methodist launched Mia as a chatbot on its career website in December, where she pops up to guide candidates through the initial application steps for select job openings. That means asking candidates who view specific job posts about their location, years of experience and—for those who meet the qualifications—ultimately their availability for a phone interview.

Houston Methodist first rolled out Mia, developed by an AI company called Paradox, for specific nursing positions, but has since expanded it to additional nursing and medical assistant roles.

“The average RN that comes through our career site goes there seven times before they click on anything,” Vernon said, noting that Mia helps to engage possible candidates who might still be on the fence about starting an application.

“You’re shopping, but you don’t want to commit. … Well, when people do that today, Mia might pop up,” Vernon said.

Houston Methodist has hired nearly 70 candidates who began their application process with Mia and, on average, the system conducts 16 phone interviews set up through the chatbot each day.

Part of the “beauty” of using chatbots, according to Vernon, is that they’re available 24/7, which allows the health system to set up interviews at the point a candidate is applying, rather than having to wait for a member of the health system’s team to respond during business hours. Fifty-two percent of the conversations with Mia have taken place outside of business hours.

“We’re attracting and hiring night-shift nurses—for example—from California, who might be (applying) at the end of their 12-hour shift, when our offices are closed,” Vernon explained.

This past spring, Houston Methodist rolled out its second chatbot project for HR—a program named Zoe that answers employee questions about benefits, such as offering clarifications on the health system’s policy on paid time off. So far, Zoe has conducted a few hundred chats with employees, while Houston Methodist is fine-tuning the project.

“We’re looking at chatbots and (robotic process automation) from many fronts,” Vernon said of Houston Methodist’s future chatbot plans. “I’m dreaming big.”