Providence creates chatbot to improve COVID-19 care and manage hospital resources

On January 20, 2020, Providence Regional Medical Center in Everett—located outside Seattle, Washington—treated the first known case of COVID-19 in the United States. As the area became the epicenter of the country’s first wave of COVID-19 infections, hospital staff and administrators knew they had to find a way to manage a potentially overwhelming increase in patients.

Chief Product Officer Maryam Gholami of Providence’s Digital Innovation Group and her team had already been evaluating the Azure Health Bot service for a few months at the time of the first known COVID-19 case. “We had done enough experimentation and had seen enough momentum with consumers using our smart health assistant that we knew that we needed to scale the solution, with best-in-class AI capabilities such as natural language processing [NLP],” Gholami says. “An assessment bot with different pathways in it could converse with consumers and navigate them to the right care option while answering their questions and helping with their anxiety. This would also help us prioritize who should see a provider right away and scale our customer service support at a very critical time, when we were extremely resource-constrained.”

Taking advantage of its strategic collaboration with Microsoft and the Chief Digital Officer’s Data Science Team, Providence implemented and configured Microsoft Azure Health Bot, an AI-based virtual assistant solution with voice and chat capabilities, to support COVID-19 assessment and FAQs based on CDC guidelines.

Providence’s Executive Vice President and Chief Information Officer,  B.J. Moore, and Executive Vice President, Chief Digital Officer, and Managing Partner of Providence Ventures, Aaron Martin, both have deep technical industry experience and are focused on bringing technology in healthcare up to par with other industries. “It’s about creating a digital interface to the healthcare system,” Martin says. “We had built a whole bunch of different technologies that we used to pivot toward addressing COVID-19, helping patients get care with processes in place to ensure safety.”

Working with partners at Microsoft, including Jean Gabarra, Vice President of Health and Life Sciences AI, and Hadas Bitran, Head of Microsoft Health and Life Sciences Israel, the Providence Digital Innovation Team, with Gholami leading the product development, quickly configured and augmented their existing Azure Health Bot to screen for potential COVID-19 symptoms. In less than a week, the new configurations were operating at full capacity. By handling the initial level of screening, Azure Health Bot immediately relieved pressure on the call centers at Providence’s 51 hospitals and more than 1,085 clinics in the western United States.

Within months, Providence's Azure Health Bot with COVID-19 templates was handling tens of thousands of patients locally, and the service was also rolled out to hundreds—and then thousands—of other providers. Providence's configuration became the basis for the CDC’s coronavirus assessment tool and has since been adopted by health authorities and providers worldwide.

Azure Health Bot helps save lives

Early on in the health crisis, Providence’s President, Chief Clinical Officer Amy Compton-Phillips, M.D., called Martin and raised the alarm on the outpouring of healthy—but concerned—citizens who could be headed their way and potentially endangering themselves. “We’re about to have a huge surge of the ‘worried well,’” Martin recalls her saying. The danger? With the lack of available personal protective equipment (PPE) at the time, there was a fear that uninfected people would go to clinics concerned that they might be sick and end up becoming infected during that visit. At the same time, Providence’s contact centers were swamped with thousands of incoming calls, leading to overwhelmed staff and patient hold times as long as five hours. This presented a risk for patients who needed swift medical advice and a burden for the nurses answering the phones.

Gholami’s team got right to work implementing and configuring the Azure Health Bot to create custom workflows for COVID-19. “I'm not sure how much sleep we got, but we knew that it was important for the community,” she recalls. Not only did she already have “100% trust” in Microsoft’s fulfillment of security and compliance requirements, but she also knew they could rely on the efficiency of Azure and the support of Azure engineers. “It felt like co-developing something together. They were integrated as part of our team,” she says of the support her team received from their Azure cloud architects.

The AI-based Azure Health Bot is a mobile- or browser-based interface that guides the user through a series of questions intended to assess the likelihood that they have COVID-19 and, if so, to assess the seriousness of the symptoms. Someone with mild symptoms and no likely exposure might be advised to rest at home and monitor their condition, while someone whose “symptoms and circumstance” more closely matched COVID-19 could be connected to another Providence-developed platform for a live video visit with a physician. If the user had severe symptoms, such as shortness of breath and high fever, they could be referred immediately for emergency medical assistance. The service also answers frequently asked questions based on the latest guidelines from the CDC and Providence clinical teams. On average, users engaged with the Azure Health Bot for two to three minutes per conversation.

An increase in capacity

As the virus spread through the Pacific Northwest, some 250,000 inquiries, mostly from the region, came through the platform in the first few months. It helped keep the worried well from overwhelming emergency rooms and ICUs. “It allowed us to dynamically increase capacity of the healthcare system,” Moore says. “Before that, nobody was thinking that you could actually increase capacity through technology.”

One important factor in its usefulness is the flexibility of the tool. Because Azure Health Bot is simple to update, the questions could be changed easily as new risk factors were understood. For example, an initial question about travel to China was expanded to include several other countries. Then, as community spread took hold and international travel became less relevant, the question was dropped altogether.

The interface is also conversational, making it not only accessible but also comforting for those who do not have evident risk. “You feel like you’re having a conversation with a human being,” Martin says.

That’s crucial. “Healthcare is very personal, so we have to bring intelligent elements such as conversational interface, including chat and voice along with personalization to that experience,” says Gholami. That’s the reason Providence was interested in a chatbot as a key component of its user interface. “Azure Cognitive Services, with its family of AI-based services and APIs, provides us with foundational capabilities to innovate upon and bring our vision of adding a smart assistant to every step of the consumer and caregiver journey,” says Gholami. Now only using written English, Providence's Azure Health Bot may soon add languages like Spanish and Mandarin, as well the ability to speak while doing sentiment analysis of conversations, all through the AI-based capabilities of Azure.

Providence also contributed a template to the Azure Health Bot gallery that other organizations can use to rapidly design and deploy Azure Health Bots to fight influenza. Providence is the first health system to publish a scenario template in the Microsoft Azure Health Bot Service catalog gallery, and they are offering it to other health systems to use for free. The scenario template for flu combines information from multiple sources with FAQs and directs users to a vaccine location finder.

A mission to serve

In the first stages of the outbreak, when there was no widespread testing available and very limited PPE, the Azure Health Bot configured by Providence played an important role beyond helping the hospitals and patients. “Early on, the best assessment of how many people in our communities may have COVID was that interaction with that bot,” says Moore.

Providence’s mission is to serve all, especially the poor and vulnerable. This mindset affected the way the hospital chose to handle this new technology. A tool like this could have been used as a strategic differentiator to separate Providence from other hospitals in the area and in the country. Instead, it made the tool freely available. “We didn’t hesitate for a millisecond to say, ‘Share this with every health system, share this with every country,’” Martin says.

“The efficiency with which Microsoft could share the technology was impressive,” Martin adds, “because it’s one thing for us to say, ‘Take these protocols and run with them,’ but the ability for Microsoft’s reach to make that happen very, very quickly was really impressive.”