The Science of Data and Technology in Patient Care: An interview with Sara Murray, MD

Sara Murray

UCSF is full of passionate people who care about our patients and community. I recently got to sit down with one of those people, Sara Murray, MD, who is a faculty member in the Department of Medicine and Medical Director of Informatics for UCSF Medical Center, to talk about how she’s used data science, technology, and investigative curiosity and skills to make improvements to patient care. We also discussed the importance of mentoring and her new project, an app that could help physicians more efficiently receive patient data while on service. Our interview is lightly edited for clarity.

How would you describe what you do in 30 seconds?

I am a Hospitalist and Medical Director of Informatics for the Health System. I lead our Advanced Analytics and Innovation (A2I) team–a team of five very talented individuals who do work in data analytics, data visualization, and data science to support quality improvement and value improvement initiatives across the Health Center. Our mission is to use EHR data in new ways, applying novel techniques to gain valuable insights in the problems facing our health system. I also do a large amount of EHR [Electronic Health Records] implementation work, working with our build teams and our clinical stakeholders to improve the EHR and the way it integrates with our practice. In both of those roles, I support a wide array of innovative work across UCSF.

How would you describe the field of artificial intelligence for someone who isn't familiar?

Artificial intelligence is essentially the integration of computers into work that has historically required human intelligence. One subset of artificial intelligence that I'm very interested in is machine learning and predictive analytics–using computers to predict the future and come up with smart suggestions for us. The most familiar example of this is Netflix trying to predict which movies you will enjoy based on your past views. In healthcare this might be predicting patients who are at risk for transfer to the intensive care unit or rehospitalization. But I would say for someone who wasn't familiar with it, artificial intelligence is really the application of computer science to problems with the goal of the computer learning or problem-solving like we humans do, and sometimes even better than we are able to.

How is that integrated into your work?

My team has been involved in leading the use of predictive analytics across UCSF. One of the things we are working on is using Epic's cognitive computing platform, which is a platform for deployment of artificial intelligence or predictive analytics within the EHR. It is unique because it allows seamless integration of predictions within the space that clinicians are doing their work. Clinic coordinators can then use that information to reach out to the most at-risk patients who might benefit from extra reminders or support. While our goal is to provide maximal support to all of our patients, having predictive analytics such as this help us direct resources where they are most needed. We are now working on several other models including models that help us identify which patients in the hospital are at risk for deterioration or unexpected complications. 

Beyond predictive analytics, there are also unique opportunities to use EHR data in totally novel ways. In the EHR we have meta-data–essentially, data about data–historically this has been used just for record keeping, but my team is doing work to uniquely leverage this data to better understand what’s happening to patients in the hospital. You may have heard of the project that I lead doing spatial and temporal mapping of C. difficile [a bacteria that causes diarrheal illness in hospitals] where we essentially used data in the EHR that no one really knew existed or could be used for this purpose, to build a model that helped us identify areas of potential disease transmission. Two years after identifying the CT scanner in the ED as a hot spot, follow-up analysis showed that we dramatically reduced C. difficile transmission in that scanner. We are now using EHR data in unique ways to understand patient movement through the hospital and potential barriers to discharge.

How did you become interested in EHR analytics and AI?

I initially was a very typical health services researcher. I was doing work with large administrative data sets looking at big problems in large populations but I was frustrated with the granularity of administrative data in that it was really limited to diagnostic codes and basic demographic information. I really wanted to get at the rich clinical data—and so, when we got Epic, I learned SQL programming and how to access Clarity, which is the back door database that houses the entirety of the EHR. The first project I did involved extracting EHR data to build a data-mart that we could use for use for something called EHR phenotyping, as I became very interested in how we might identify populations of individuals with a specific disease accurately using EHR data. As I increasingly used EHR data and data science techniques in my research, my work naturally transitioned to questions focused on how we apply these techniques in real time to impact our patient population which naturally aligned with my current health system role.

You talked a little bit about EHRs and being able to extract underlying data, are there any other types of technology that you rely on in your role as a researcher?

I'm interested in technologies that integrate with the EHR to optimize the delivery of healthcare for patients and providers. One of the things that I'm working on in collaboration with Michelle Mourad and Raman Khanna, is launching an Applet Core for the Division of Hospital Medicine that will ultimately serve others in the Department of Medicine as well. Epic is our EHR and while we have control over some things, there are limitations to what it offers. The idea of the applet core is that when faculty have ideas for small apps that could make clinical work easier, or make research better, we could rapidly develop an app, integrate it with the EHR, and evaluate how well the tool is working. I suspect as we built more tools, we will learn a lot in the process, and become more efficient.

Could you give an example of an app that someone might want to develop?

We have a vision for an app that we call Stream. As a hospitalist, I have my team of 14 patients and I spend all day clicking in and out of charts to see if things have happened or what the result was, and I don't have a good system for having that data come to me in real time. So, our vision for this app would be a tool where you could you open your phone instead of having to click in and out of charts—and there would exist a feed of all the critical things that are happening to your patients. For example, Patient A had a critical lab value come back, Patient B just got back from the CT scanner, and Patient C had a consult note written–all in one space in real time. I suspect a tool like this will allow providers to multitask in a much more meaningful way.

What decisions do you anticipate that you would be making with that information?

I think there's a couple goals. First of all, we have major issues with throughput in the hospital and if you look along the trajectory of someone in the hospital there are infinite points where things could be slowed down–you can imagine that we're getting the data in real time and we're acting on it in real time there may be efficiency benefits to that. If we're getting the critical value and acting on it before the nurse pages us we might close the loop on things more quickly. There's also the provider satisfaction metrics we can evaluate, asking questions like, do providers feel like they are getting information in a more meaningful way? Are they spending less time at the computer and more time at the bedside? I think an app like this could meaningfully impact clinician workflow.

Can you describe the role that mentoring has played in your career?

I think mentoring is critical, probably as critical as your personal drive to accomplish things. I have been fortunate to have had incredible mentors. I actually started out in Rheumatology before Hospital Medicine and I had incredible mentors in Rheumatology who have remained mentors even as I moved into Hospital Medicine and Clinical Informatics. Mentorship is important in multiple capacities. First you need life mentors—people who will say OK, let's take a step back, let's look at your career. Are you accomplishing what you want to accomplish? Are you in the right space? Then I think the right kind of project mentors as being super important as well. There are people who you can you can talk to about your project ideas, get immediate feedback, people who will know help you write papers and challenge you to do better. I would attribute a lot of my success to really great mentorship, honestly. And that's one of the benefits of being at UCSF. There are a lot of fantastic mentors. Now I have moved more into a role where I mentor others and I'm trying to pay that forward and I think there's a culture of that here.

What advice do you have for other women in technology roles?

Historically women are under-represented in technology so it's really important that women who are interested in this space pursue it. As we know, diversity is incredibly important in driving innovation and creativity, so we need women at the table when it comes to health technology just like we need women at the table in all sectors in healthcare. I would say–find mentors who will provide you with opportunities. At UCSF we’re very lucky that we have good representation of women in technology and many of us are eager to mentor people who are interested in informatics. We [the Division of Hospital Medicine] have a Clinical Informatics fellowship now, and that is also a wonderful opportunity for individuals interested in pursuing this field.

Thank you, Sara, for giving us a glimpse into your world of data and patient care. For more information about Sara Murray, you can check out her UCSF Profile here.