Podcast Q&A with Mark Hulse: “There’s a Lot We Can Learn” from the Data

During the past few months, a number of highly-respected CIOs have stepped into CDO (chief digital officer) roles, leading some to wonder about the implications for both positions. Do health systems need both titles? What is the reporting structure?

Those are just some of the questions.

And, as is often the case, it depends on the organization and the individual, according to Mark Hulse, who recently stepped into the CDO role at City of Hope. “If you’ve seen one CDO, you’ve seen one CDO.” At City of Hope, a comprehensive cancer center based in California, the primary objective in creating the CDO position was to bring IT, research informatics, and enterprise business intelligence “under one digital vertical.” And in doing so, help drive the organization’s groundbreaking oncology learning platform forward.

Recently, Hulse spoke with Kate Gamble, managing editor at healthsystemCIO.com, about the work his team is doing to drive transformation and leverage analytics to provide decision support and develop predictive models; the measures they’re taking to keep vulnerable patients safe during the pandemic; and how he hopes to keep the virtual health momentum going.

LISTEN HERE USING THE PLAYER BELOW OR SUBSCRIBE THROUGH YOUR FAVORITE PODCASTING SERVICE. 

Podcast: Play in new window | Download (Duration: 16:06 — 14.7MB)

Subscribe: Apple Podcasts | Spotify | Android | Pandora | iHeartRadio | Podchaser | Podcast Index | Email | TuneIn | RSS

Key Takeaways

• “Increasingly, data and analytics are going to be such an important part of what we do and what we offer,” which means leaders need to focus just as much as being a data and analytics company as being a clinical and research institute.
• By leveraging AI and machine learning, organizations are able to “surface patterns within the data that aren’t readily obvious with more traditional analytics techniques.”
• Of the three major components that comprise the oncology learning platform, real-world action is the most critical, said Hulse. “Discovering evidence within the data doesn’t make a whole lot of difference unless you can really influence decision-making at the point of care.”
• City of Hope has developed a number of predictive models, including one that can detect moderate or high-risk sepsis in bone marrow transplant patients by extracting data out of Epic.
• It’s not just about clinical data. When clinicians have all of the pertinent information about a patient, it can foster discussions about key areas such as end-of-life care.

Q&A with Mark Hulse, Chief Digital Officer, City of Hope

Gamble:  Let’s start with a brief overview of City of Hope. Can you give some high-level information about the organization?

Hulse:  Sure. City of Hope is an NCI (National Cancer Institute) designated comprehensive cancer center. We have 51 centers nationally. Our main campus is located in Los Angeles County, just north of downtown Los Angeles. Over the years we’ve expanded to more than 35 sites throughout Southern California. We’re currently building a new cancer center in Orange County that will be opening next year, as well as a new hospital. So we’ve definitely been in expansion mode.

Clinically, we’re a national leader in blood cancers in particular, including bone marrow and stem cell transplants. We’re also leading in emerging treatments such as CAR-T cell therapy, a type of immunotherapy that’s showing a tremendous amount of promise for some difficult-to-treat cancers.

As a major research institute, we typically have well over 500 active clinical trials at any given time. We also have three good manufacturing plant facilities, which allow us to manufacture our own drugs and immunotherapy agents. And so, if one of our researchers has a study involving an early emerging drug that has been discovered at City of Hope, we actually have the capability to manufacture that drug to the top pharmaceutical specifications, and then activate our own early-stage trials. We’ve contributed to a number of major agents that have then been licensed out to pharma companies.

 

Bringing tomorrow’s discoveries to patients today

Gamble:  That sounds really exciting. Is that part of what appealed to you about coming to City of Hope — being part of cutting-edge initiatives?

Hulse:  Yes. Having previously been at Moffitt Cancer Center, I was familiar with the cancer piece. I think what particularly struck me about City of Hope was the level of innovation that is present throughout. All cancer centers — and I’m sure it applies to pediatric and specialty centers — have a culture of dedication to focusing on the patients. You always feel like you’re very much in the fight, if you will, with the patient. But I think what set City of Hope apart is the drive to bring tomorrow’s discoveries to the patients who need them today.

When I spoke with Robert Stone, our president and CEO, as I was considering the role, one thing he said to me was, ‘we’re a comprehensive cancer center. We’re always going to be doing best-in-class clinical care and research. But increasingly, data and analytics are going to be such an important part of what we do and what we offer.’ It’s about becoming a data and analytics company in addition to a clinical and research institute. As a chief digital officer, that was very inspiring to me and it’s something that has really helped lay the vision and the groundwork for a lot of what we’ve been focused on during the past three years.

 

Data & analytics: “There’s a lot we can learn”

Gamble:  Can you talk about some of the initiatives involving data and analytics?

Hulse:  Sure. So, when I thought about this idea of how we can take these discoveries and bring them along very quickly, I thought about the healthcare learning platform, which originally came from the Institute of Medicine. It was a concept that came out around the early days of healthcare organizations adopting EMRs, with the idea being that if you’re just capturing data on patients, it’s just an electronic chart that captures patient information.

But if you look at the data in aggregate and look at specific patient populations, there’s a lot we can learn, whether it’s the quality of care or patterns in the data. As we’ve advanced in terms of data science with machine learning and artificial intelligence, we now have the capability to really surface patterns within the data that aren’t readily obvious with more traditional analytical techniques. So this has really opened up a whole new piece.

 

Oncology learning platform: From real-world data to real-world evidence

We’ve developed a concept at City of Hope of the oncology learning platform. There’s a number of different components, but it’s primarily focused on capturing real-world data out of our EMR and other systems, and being able to link that data. We’ve established a department of applied artificial intelligence in data science; this team works with our researchers and folks in research informatics and other areas to analyze patterns in the data, and to develop predictive models so that we can stay a little bit ahead of what’s going to be happening with patients. In doing so, we move from real-world data to real-world evidence within the data.

It really has three major components: real-world data, real-world evidence, and the last and perhaps most critical one, which is real-world action. When you think about it, discovering evidence within the data doesn’t make a whole lot of difference unless you can really influence decision-making at the point of care, where it’s going to be most impactful. And so, once we have a well-validated predictive model, we run that against the data in real-time and surfacing clinical decision support to our clinicians.

 

Leveraging models to predict sepsis risk

In terms of some of the early work we’ve done, we treat a lot of blood cancer patients, particularly those undergoing bone marrow transplant. These patients are immunosuppressed and very vulnerable to infection — particularly to massive infections like sepsis. And so we developed a predictive model that can detect moderate or high risk of sepsis in these bone marrow transplants. We extract data out of Epic, our EMR, in real-time. That engine is running on this data, and so if a patient is at moderate or high risk, that gets surfaced back into Epic as an alert to the clinicians. We’ve developed a number of these models now.

A number of centers have developed predictive models in sepsis. Even some of Epic’s new components that are doing this type of predictive modeling have a sepsis model. The problem is that you can’t just take a model that’s been developed on one set of patients and apply it to a very different situation. Again, if you’re looking at bone marrow transplant patients, that’s going to be a very different predictive model than it would be in the general population. So that’s a piece we’re focusing on.

 

Studying mutations

Within what we call the digital vertical, which is the group of departments that I lead, the oncology learning platform has been a signature initiative for us. Although we’re rolling it out in many clinical areas as well as some operational and even finance and patient access groups, the main focus has been on supporting our precision medicine efforts. We’ve incorporated genomic data; we offer genomic testing to all our patients, and we can incorporate that data into their clinical data. And again, we’re starting to look at running some predictive models around those pieces.

There’s a small set of very well-known genomic mutations that have been well-studied in patients with lung cancer and breast cancer. If you have one of those mutations, for example, you might not be put on frontline therapy because you’re going to be resistant to that due to your particular mutation. But for every mutation that has been well-studied and well-known and is actionable by clinicians, there are many other genomic mutations that we just don’t understand today. We believe that by accumulating data over time and being able to run advanced analytics, we’ll be able to discover more patterns that will help drive further research in that area.

 

Beyond the “pure clinical aspects”

Gamble:  It’s really exciting to think about the possibilities of using the data for decision support. But, as you alluded to, the challenge is that you can’t just take a model from one set of patients and apply it to others. How can that be addressed?

Hulse:  That’s a lot of the work that our applied AI team is doing. If a model has been developed in a different population or a different center, we can certainly look at that model and run it in test mode against our patients. We can then use statistical techniques to be able to determine whether that model is a proper fit for our patients, whether or not it’s predictive. If not, can we get closer by tuning and tweaking some of the variables — or, what are referred to as features within data science — or are we better off just starting our own model?

We’ve also looked at models that can be applied to a more general population, such as, for example, patients who are at risk for readmission within a specific timeframe. In that case, we’re looking at prognostic models for patients who have very severe disease or maybe end-stage disease. None of the models we use are saying, ‘based on this, here is how you should treat the patient.’ It really just offers additional kind of insight so that clinicians can use their knowledge and expertise to make decisions.

The prognostic model, for example, may indicate that it’s time to talk to the patient about end-of-life care; or, if the patient has advanced directives already established, that it’s time to have that conversation with the patient. Because oftentimes we’re so focused on doing absolutely everything we can, and sometimes we ignore the fact that patients are at a point where they need to think about what their quality of life will be versus just moving on to the next treatment. It’s one of those areas where having that particular information can help foster discussions between clinician and patients, versus just looking at the pure clinical aspects.

 

Patterns in patient satisfaction data

We’re starting to look at patterns now in our patient satisfaction data. Like many organizations, we use HCAHPS and other models for guidance, but sometimes it’s not always obvious why a certain set of patients have a less-than-ideal experience. How can we better predict those, and provide real-time guidance to the people who could make a big difference in the outcome of that patient experience? Those are some of the other areas we’re looking at as well.

 

Remote monitoring in new facilities

Gamble:  You mentioned a new cancer center and hospital being built. What are some of the initiatives you’ll be looking at for those facilities?

Hulse:  In addition to that, we also have construction happening on our main campus in Duarte, including a 170-bed hotel that is scheduled to open in January. Part of the focus there — as well as the new facilities in Orange County — is the patients who don’t necessarily need to be in the hospital, whether it’s because of the illness itself, or the measures being taken to treat the disease. They do, however, still need to have some level of monitoring.

And so, in both cases, we’re looking to provide in-home monitoring devices. We’re working on pilots with a few companies. The plan is for the hotel to eventually be completely outfitted with remote monitoring, and for patients at the Orange County facility to have it in their homes.

As wearable devices continue to advance and become more sophisticated in terms of biometric measures, we’re looking at how we can incorporate some of that data in real-time. That’s what we’re focused on right now.

Part 2