Duke Health Leaders Build the Predictive Analytics Foundations to Improve Patient Care

March 14, 2018
At Durham, N.C.-based Duke University Health System, health IT leaders are moving forward on many fronts to leverage data analytics to improve patient care, such as early sepsis detection.

Across the country, leading health systems are building the data architectures that are necessary to leverage business intelligence tools and analytics for effective population health and proactive disease prevention. At the Durham, N.C.-based Duke University Health System, health IT leaders are moving forward on many fronts to leverage data analytics to improve patient care, such as early sepsis detection.

“It’s a very exciting time in healthcare, as more and more people understand and appreciate the power of analytics. We are certainly going beyond the traditional flavors of business intelligence with pre-defined reports, which of course will have its place, and migrating to using predictive modeling and machine learning to figure out how to tailor care for both individual patients and populations of patients,” says Eric Poon, M.D., chief health information officer at Duke University Health System. Poon also is a professor of medicine at Duke University School of Medicine and he practices general internal medicine at Durham Medical Center, part of Duke Primary Care. Poon also earned a Master’s in Public Health from Harvard University.

The Duke University Health System encompasses three hospitals—Duke University Hospital, Durham Regional Hospital and Duke Raleigh Hospital—as well as physician practices, home hospice care and various support services across North Carolina.

In his CHIO role, Poon aims to align technology solutions with organizational objectives, the clinical and analytic information systems that impact patient care. For Poon, a big focus right now is forging the necessary collaborations and partnerships across Duke Health in order to bring the power of data science to bear in the context of healthcare.

“Traditionally, the folks who we need to bring together include the clinical and operational folks who have insights into the challenges that they themselves face, or they see their patients face. And then you need to marry that up with the data scientist who can apply a variety of methods to help solve the problem, and then both parties need to come together and access the right data in order to do any meaningful data science. That collaboration is essential, and we are making a lot of inroads in helping all those three parties understand each other’s capabilities and provide the glue to keep those projects going,” Poon says.

Eric Poon, M.D.

And, he adds, “I think these areas, which are exciting and new, require people with different skillsets to come together and those skillsets typically sit in different pockets of the organization. How to support the collaboration and building the trust, that, I think, has to be a continued focus for those in leadership positions.”

Poon will join other healthcare IT leaders to discuss this topic, as well as other topics related to healthcare IT in North Carolina and the region of the Southeast U.S., October 19 and 20, during the Health IT Summit in Raleigh, sponsored by Healthcare Informatics, and held at the Sheraton Raleigh Hotel Downtown.

In the past six months, Duke Health’s IT and analytics leaders have made significant progress in their work to build the foundations for predictive modeling, Poon notes. “Making sure that we provide high-quality data sets to inform the model building is a key step. In addition, making sure that there is the infrastructure to house these data sets in a secure way and putting it behind a secure storage environment, making sure we have the appropriate tools and computational power to support the activities of the data scientists, these are issues that we are very actively working on and I think we’re making a lot of progress,” he says.

The next mile of the journey, Poon notes, is to get the insights from those predictive models back into the hands of clinicians at the point of service. “After all, a data science project that results in a manuscript is not, by and of itself, going to help any individual patients. So, we are having a lot of conversations about how to effectively get those insights back into the hands of the right clinicians at the right time and at the right point in the workflow. And, then, think about how to do it robustly and think about how it articulates with state of art in terms of the standard of care, and it’s something that I’m very excited to be working on,” Poon says.

Poon and other IT leaders at Duke Health are currently working to leverage predictive analytics models to help clinicians identify the early signs of patient deterioration and to identify patients developing early signs of sepsis. “Both projects are at a point where they have built really promising predictive models and they are actively working with the data scientists to figure out how do you get those predictive analytics back into the right workflow,” Poon says. “I think it’s very important to make sure that we understand how these predictive scores can and should be used in the processes of care, and we want to make sure that we adhere to standards of care, and where the standards of care do not give further guidance, this is where these predictive scores can be very helpful.”

Technology leaders at Duke Health have encountered many practical challenges along this journey, including the need to help clinicians and senior executives understand both the power and the limitations of predictive modeling and machine learning algorithms, he says. “Understanding under what circumstances would it be okay to turn predictive analytics into prescriptive analytics, that’s a step that I think many people are trying to make, to understand the rules of engagement. I do think that making sure that we can support this activity across multiple domains, because there are many opportunities to use these types of techniques, and so there will be many opportunities to use different data sets, and so, how do you make all those data sets available, even as things as simple as that, will take continued work,” he notes.

Moving Forward with Value-Based Care Initiatives

Poon and his team have also focused on providing the IT infrastructure and tools to support the health system’s work to move forward on value-based care and payment models. For the last three years, Duke University Health System has been participating in the Centers for Medicare & Medicaid Services (CMS) Medicare Shared Saving Program (MSSP) in the Track 1 Accountable Care Organization (ACO) model, and, beginning in 2018, the health system will move into Track 3 of the program, which includes both upside and downside risk. “Over the last couple of years, we have worked very successfully both with our practices here at Duke and our local ACO partners to make sure that we meet all the quality metrics for the MSSP ACO and we have been very successful about coming up with appropriate workflows so we can leverage the electronic health records (EHRs) to minimize the need for manual data review and verification. We have been reaching out to some of our outlying practices that are on our EHRs with the appropriate information so they can appropriately close care gaps,” Poon says.

He continues, “I think the activity over the last year has been thinking about how do we harness the value of claims data to understand where there might be opportunities for us to better manage the utilization in a rational way for patients; where are there gaps in care, where there are variations that cannot be easily explained by the conditions of the patients so that we can proactively have the right conversations with clinics and groups of clinicians to propose clinical pathways.” And, he adds, “I think we’ve had a lot of success in the inpatient arena, using DRGs (diagnostic related groups) as a unit of analysis, to look for variations with patients that are discharged on similar diagnosis. I think we are actively looking at how to apply that discipline in the management of outpatient problems, like depression and congestive heart failure.”

IT leaders also have worked with the health system’s EHR vendor, the Verona, Wis.-based Epic, to better utilize claims data and developed a cost utilization dashboard, Poon says. “That is a claims database where we are using claims data from our MSSP program so that we can understand where our PMPMs (per member per month) are trending and where there might be pockets of opportunities,” he notes.

Looking ahead, Poon sees continued work to provide the advanced IT tools and capabilities that health system leaders will need in the ongoing transition to value-based care. “We’ve actually also done a lot of work in making sure that our diagnostic codes appropriately reflect on the complexity of the patients that we serve. That is an important area that we will continue to focus on; working on reducing readmissions and reducing unnecessary visits to the emergency room,” Poon says.