Back to top

Design Thinking in Health Care: How a Virtual Nurse Helps Keep People Out of the Hospital

Illustration of AI nurse app by Misa Yamamoto.

By Cathy Pearl, VP of User Experience, Sensely

Sensely’s app uses an avatar-based nurse to have a daily check-in “conversation” with patients to help them manage their health as they deal with chronic congestive heart failure. Human-centered design is key to building a smart artificial intelligence platform that improves health outcomes for patients on a daily basis.

We’ve all had the experience of our word processor crashing before we’ve hit save, not being able to play a favorite movie because the WiFi is out or gnashing our teeth when using a website because it’s so difficult to navigate. These things can be maddening, but when it comes to healthcare, the stakes are even higher.

At Sensely, we use design thinking to find solutions for real healthcare problems. We have some pretty cool technology: a virtual nurse avatar (our most popular one goes by the name “Molly”), speech recognition, natural language processing and wireless integration with medical devices such as blood pressure cuffs. It’s easy to get distracted by the technology, but the question we must ask ourselves, over and over again, is: what is the problem we are actually solving?

One example of a problem we’re working to solve is to help people with congestive heart failure (CHF). CHF is a condition in which the heart is unable to pump adequately to meet the body's needs, and affects about 5.7 million adults in the United States. According to the Centers for Disease Control and Prevention, heart failure costs the nation an estimated $30.7 billion each year.

“Helping people with CHF” is a pretty vague problem, however; what does it mean to “help” this group of patients? We needed a more specific and measureable goal, and so we focused our sights on the specific issue of reducing the 30-day hospital re-admission rate. The national average for heart failure patients returning to the hospital within 30 days is about 22 percent. (2)

This gave us an objective measure to reach for. We of course had secondary goals as well, such as helping patients stay healthier, but that’s much more difficult to measure. As more patients have been using our program, we came across additional benefits as well, such as the fact that some patients, given these new resources to track their own data, started managing their own diuretic medication. That had never been a specific goal of ours, but came out on its own.

Now that we had a problem defined, and a goal to go with it, we needed to design and refine the solution.

The Sensely app uses an avatar-based “conversation” with the patient to lead them through a daily check-in. Every morning, each CHF patient receives a reminder on their phone to take their check-in. After starting the app, Molly initiates the conversation by asking an open question such as “How are you feeling today?” and then follows up with instructions on taking their blood pressure and weight, and asks a few more questions. The patient’s responses are analyzed and organized. If necessary, an alert is sent to their clinician. Patients can speak or text with the avatar.

Why do we use an avatar? Many of the things we ask our CHF patients to do could be done with a simpler button-pushing app. But the next important piece of design thinking is to keep people in mind. We all know, for example, that we should eat less sugar, exercise more and get a good night’s sleep. Yet how many of us always follow these guidelines?

People often benefit from a “nudge.” “Nudge theory” is that idea that behavioral changes can come about with small modifications, such as when Google offices changed their micro-kitchens to display fruit prominently and put the sugary snacks in closed containers, leading its employees to eat healthier snacks. Another example: if I put my exercise clothes on the floor the night before, I’m just a little bit more likely to exercise when I get up the next day. If my company automatically opts me into my 401k, I’m much more likely to continue on that path. We believe the avatar and voice-enabled conversation add that “nudge” to encourage people to do these types of tasks on a regular basis.

When I speak with patients about using the app, they talk about the feeling of someone “holding their hand.” They apologize to Molly when they miss a check-in, and tell her to have a nice weekend. No one thinks they’re speaking to a person, but the act of having a conversation about this daily task makes it a little more tolerable.

There are two more aspects of design thinking we employ at Sensely that relate to the conversation the patient has with the avatar.

When the user interacts with the avatar, it’s done via a conversational back and forth between patient and avatar. The avatar gives instructions, for example, such as when to put on the blood pressure cuff. She also asks questions, such as if the patient has had any shortness of breath. It’s crucial to make this conversation follow natural human conversation principles; stilted or robotic interactions will lead to frustration and fewer successful check-ins.

Creating these conversations is not simply a matter of saying, “We need to ask these 10 questions… and we’re done!” Think about how you interact with a customer services agent. For example, if you call a company to complain about a problem with your bill, a typical interaction has a few pleasantries before the crux of the issue is discussed. Similarly, a nurse or a doctor doesn’t simply bombard their patient with a series of rapid-fire questions.

To make sure conversations are designed well prior to being developed, we begin with a basic staple of voice user-interface design: sample dialogs. Sample dialogs allow designers (and other stakeholders) to try out what the interaction might sound and feel like, and revise in a low-tech way without using up engineering resources or putting in design that will be too costly to change after it’s been developed.

A sample dialog is a possible path through the conversation—essentially, it’s like a movie or TV script. Here’s an example:

MOLLY

Hello. How are you today?

PATIENT

I’m feeling OK today.

MOLLY

Thanks for sharing. Were you able to take all your medications yesterday?

PATIENT

Yeah, I was.

MOLLY

Great. Okay, now, let’s take your weight. Please tap the scale with your foot, and I’ll wait for it to connect…Okay, we’re all set.

Note that this phase of design doesn’t try to capture every possible scenario. Instead, it looks at the “blue sky” path, in which everything goes well, and a couple of error paths for when things might go off track.

One key thing to note about the above sample dialog is the intro. Would a nurse immediately bark, “Step on the scale!”? No, he or she would presumably ask their patient a question before getting to the vitals measurement stage. As part of our design thinking, we look to real-world examples of how it’s done well. Ideally, we speak with the clinicians themselves to understand not just the content of the questions they ask, but the human aspects surrounding how they ask.

Molly also leads with a question that’s more small talk than medical. This serves a couple of purposes: it eases the patient into the interaction, to get them comfortable. In addition, it teaches them they can talk to the avatar. Some of our users are in their 70s and 80s but when Molly asks them “How are you?” they know what to do. They don’t need an instruction manual.

The other key component to making these conversations succeed is to monitor how they’re doing and find failure points. This is done by looking at patient responses to see where the speech recognition or natural language processing may have failed. In our symptom checker, we ask, “What is your main symptom?” This may seem like a fairly straightforward question, but people respond in a wide variety of ways. For the symptom “abdominal pain,” people might say, “My stomach aches,” “I have tummy pain”, or “My stomach doesn’t feel so good.” If we don’t capture and respond to the normal way humans speak, we’ll never succeed. So it’s important to continually tweak and improve the app, based on real user behavior. Design thinking is not over just because you’ve launched your product!

We also have learned that we’ll continue to test and tweak the product after deployment. For example, one client wanted a large variety of alerts to pop up on their dashboard based on patient behavior. We provided them, but within a short time, the clinicians realized it was information overload, and we scaled it down.

Although we have not yet published our results, initial indications show that our avatar-based, conversational approach is working. Patients who are using our app (and whose clinicians use our app to monitor them) have greatly reduced their 30-day hospital re-admission rate.

At Sensely, we begin by using design thinking to help determine what problem we are trying to solve. Next we use it to craft conversations that the patient can have with our avatar, and we get feedback from real people early in the process to make sure we’re on the right course. Finally, once we’ve launched, the work is by no means over. We monitor performance and work hard to understand where there are issues so that we can fix them and continue to help our patients stay healthier.