Q&A: Penn Medicine researchers mine patients' social media data to improve engagement

Yelp reviews aren’t just for finding the best burrito within a one-mile radius. Researchers at the University of Pennsylvania Health System believe social media platforms like Yelp can help identify patient engagement gaps and offer insights into improving the hospital experience for both patients and caregivers.

Since 2013, Raina Merchant, M.D., has led efforts to explore social media data through Penn Medicine’s Social Media Laboratory. Last week, she was named director of the Center for Digital Health, a newly created research arm of the health system undertaking broader efforts to explore social media data.

Social media
UPenn is mining data from social media platforms.

Under Merchant’s direction, researchers at the Center for Digital Health will pull data from Facebook, Yelp, Instagram and other online sites to better understand how patients interact with the health system, and what clinicians can do to improve that experience.

Merchant, who also serves as the associate vice president for the University of Pennsylvania Health System, spoke with FierceHealthcare about her past research and how she expects her future research will influence clinical practice.

FierceHealthcare: Can you start by explaining how the Center for Digital Health came to be based on some of your previous work?

Raina Merchant: This really evolved out of the Social Media and Innovation Lab, which I started in 2013. Really our interest there was in trying to evaluate if there was value in studying social media data. So we spent a lot our efforts trying to study and understand information about health based on what people posted on different social media platforms.

FH: What is your focus moving forward?

RM: The idea is that while I’m still really excited by social media, and I don’t think it’s going away, we’re excited about expanding to include a range of digital tools and platforms that enable connectivity, allow us to think about the patient experience, and allow us to listen and learn from patient better.

We wanted to have a broader mandate that would include all things digital. Many of our projects go beyond just Facebook and Twitter, but looking at other online platforms and online reviews and starting to do some work in artificial intelligence. I think we didn’t want focus on the shiny new objects, but rather we want to be able to forecast what is happening in the future to a certain extent, and start to study and understand what that means for our patients and how we understand more about their health.

Another change in our focus is we will still do a significant amount of scholarly research, but we’re also looking to move beyond that and think about how we directly plug our findings into operations.

FH: How does that affect the clinical side? What can you draw from a Facebook post or a Yelp review that's helpful for a physician?

RM: We had a thought piece in JAMA looking at how Yelp can help us listen to patients better. The traditional hospital surveys really only ask the patient about their experience. They come by mail or email and they ask a set number of questions. We are more interested in the organic things that not only patients care about, but also their caregivers and family members.

A platform like Yelp allows the friend or parent who is sitting there with the patient, and really having a very different experience, to share what they like and didn’t like.

Here at Penn Medicine, we could learn about how to improve the patient experience by understanding what our patients like and don’t like, but also understanding more broadly the places where people are really excited about a particular thing that might be happening or something that was a really negative experience. It’s looking at how we can learn by listening beyond our usual tools.

FH: But there is such a wide variety of information a Yelp review—it’s not uniform. Does that make it difficult to collate the data?

RM: That’s both the challenge and part that makes it interesting—the data isn’t validated or standardized. People are talking about a lot of things about the healthcare experience.

One of our studies showed people talked on average about 16 different things in a review. We work closely with computer scientists that use machine learning and natural language processing to sift through language and help us find key themes and allow us to codify some of these things by using more advanced computational processing.

(Editor's Note: In the study referenced above, researchers used a statistical language processing model called Latent Dirichlet Allocation, which analyzed narrative Yelp reviews to generate 50 general topics based co-occurring groups of words or terms. Merchant noted that instead of using vendors, her group coordinates with computer science faculty and students at University of Pennsylvania.)

FH: Are physicians open to using this information?

RM: We’ve found that people are excited to understand more about the patient experience, particularly if we look at what people are saying about a practice. Many of the reviews on Yelp are less about individual physicians and more about the overall hospital experience.

They are saying, "My overall visits was affected by these particular things." Clinicians have a systemwide focus and they are interested in how to make the system better for all patients.

FH: Maybe it's just me, but is it slightly disturbing how much information is available to you online? Obviously, it’s useful for you, but is that something that patients recognize or care about?

RM: One of the things we pay a lot of attention to is issues around privacy, and issues of consent and transparency. These are issues that are critically important to everything we do.

Because of the newness of this area and integrating these tools into a health research context, we’re really very careful about data. We only look at things in aggregate and we don’t look at individual posts or searches. We’re really looking for signal across lots of different patients.

The idea is that in thinking about precision medicine, we can begin to understand more about individuals and give this information back to them by studying these digital footprints. I think people are interested in sharing to contribute to this bigger cause, but also because buried in their data may be answers to ways we can improve their health and the health of others. I think there is a component of altruism where people are interested in sharing for the purpose of improving health.

FH: How do you envision this work shaping the future of healthcare? In other words, how might hospitals integrate social media data into their approach to care?

RM: The health system of the future is going to have to understand how to listen to patients and be engaged through different channels.

There are lots of industries outside of healthcare where people primarily communicate through digital tools, so I think healthcare systems have to change in some ways or really allow more communication between patients and providers beyond just the in-person hospital visit. We really have to think of different ways of connecting with people where they are and healthcare doesn’t have to be something that is tied to a physical location, but we can use these means of communication to stay in contact with people beyond the face-to-face visit.

We’re thinking more about connectivity and the fact that people are collecting and reporting and sharing all this data, but we need to understand how to study it. The data is in different formats and it comes at different speeds and some we can validate and some we can’t, and we have to think about how we build better algorithms for both processing data and getting it back to patients to refine it and make it better.

Editor's Note: This interview has been edited for length and clarity.