Can healthcare providers and plans leverage machine learning and data science to make inroads in new areas of activity? That certainly is what leaders for an integrated health system and a Blues plan are helping to make possible at their organizations. On May 1 at the World Health Care Congress, being held at the Marriott Wardman Park Hotel in Washington, D.C., Vijay Venkatesan, chief data officer at the Seattle-based Providence Health & Services, and Sherri Zink, senior vice president and chief data and engagement officer at the Chattanooga-based BlueCross BlueShield of Tennessee, shared their perspectives on some of the data-driven work they’re doing at their health system and health plan, respectively.
On Monday afternoon at the World Health Care Congress, Venkatesan and Zink shared insights from their ongoing work, in a session entitled “Analyze the Convergence of Machine Learning and Data Science to Enhance Real-Time Patient Engagement and Experience,” the two leaders offered sequential presentations on their data-driven and data-facilitated initiatives. Their session was one of the sessions that composed the Data Analytics and Technology Summit, one of 14 concurrent summits within the World Health Care Congress event.
Venkatesan, who reports directly to his health system’s CIO, told the audience Monday afternoon that “Our IS vision is to transform health through simple, reliable, and innovative approaches to technology solutions.” And in that quest, he said, “The hardest thing is finding the data. How do we find the data, and who is the oracle of the information? How do we bring together the producers and consumers of data, in an interoperable platform, to begin to have a discussion of strategy and tactics? We want to bring together the consumers of data with the producers of data, creating partnerships between data producers and consumers, and to enable consistent platforms and tool sets to enable the good use of data, and govern the issue of data.”
Speaking of the journey around the strategic leveraging of data at Providence, the third-largest not-for-profit integrated health system in the U.S. Venkatesan said that “We want to achieve the end goal of outcomes as a service. We’re really trying to help you drive to improved outcomes through data,” he said, speaking of the users of data in his organization, from frontline clinicians to clinician leaders, to administrative executives and managers. “We have a broad mandate,” he added, noting that the data analytics that he and his colleagues in the data department are developing for their organization are supporting, among other things, “population health, clinical care and personalized health, shared services, and regions and ventures,” as well as “the employed provider network, ACO, government programs, care management, payor contracting, health plan, and regional executives,” among other areas that they are supporting.
The challenge? “Data exists in various pockets in your organization; we’re a 50-hospital system with 23,000 physicians and a lot of clinics,” Venkatesan noted. “You can guarantee that there are a lot of different analytics initiatives going on. Still, even though we have all these great pockets, it’s hard to find the information across the system. The human network bottleneck problem is a real problem. It can take you 15 emails just to figure out whom to approach about data, and another 20 emails to get action. So how do we approach this human bottleneck problem that makes this conversation a bit easier?” What’s more, he noted, “It’s becoming more and more important to get to achieve speed to action. Most healthcare organizations, like ours, are under a lot of financial burden, and you have to do things in an agile and nimble way.”
The solution has been a new mechanism, which has been named myHIway, which emerged out of a “two-part strategy,” Venkatesan said. “We built a simple platform using text mining. We built a Google-like platform called myHIway, to help people find things. We did basic topic modeling using natural language processing. It goes across all the data stores and organizes things in a way that’s easy to see. So just having a data mart isn’t enough any longer,” he stated.
“You have to create a platform where others can contribute data,” Venkatesan told his audience. “So we said, we’ll build a data lake that can host data. We created this data lake for incoming flow: raw data in any format, any size goes into it. And we started building data apps for it, and we built targeted use cases for denials management and other purposes. Our goal,” he said, was to say, yes, we’ll create an electronic data warehouse, but we’ll also invest in hosted platforms, with very targeted use cases.”
Creating myHIway has “allowed us to streamline our strategy,” Venkatesan said. “And if you’re a consumer of our services, all you have to do is to go to myHIway. And the idea is that as you build more and more targeted, use-case apps,” demand for data support will become better rationalized over time, he added.
Connecting with plan members in Tennessee
At BlueCross BlueShield of Tennessee, reaching out and connecting with plan members has been a major objective of senior executives, said Sherri Zink, senior vice president and chief data and engagement officer for the plan, which covers 3.3 million members across Tennessee, including 11,000 employer groups, as well as large Medicare and Advantage and Medicaid populations.
Zink told the Data Analytics & Technology Summit attendees that the key to moving forward as a health plan in the arena of plan member engagement, is to shift one’s data analytics strategic from retrospective to predictive to prescriptive, in stages. “Who do we want to outreach to? Which clusters of members are driving the biggest medical costs?” she asked. “We often think we need to focus on the high utilizers. But oftentimes, it turns out that you need to focus on individual who may be seeking care only once in six months, or who has a sick child. Machine learning can help us to figure out what’s going on” at both ends of the spectrum, in order to identify plan members both who might be over-utilizing and who might in effect be under-utilizing.
One key area that has been developed at BCBS Tennessee, Zink noted, is the Member Scorecard. “Every year, usually twice a year, we launch the Scorecard to our members,” she said. The scorecard incorporates a number of data points, including around gaps in care, such as missed wellness visits, and the need for hemoglobin a1c or HPV tests to be scheduled, as well as including health education topics. What’s more, it’s delivered via “multi-channel delivery—everyone wants it delivered differently,” she said.
“In terms of interacting with consumers, we also pull in health assessments, activity data, lifestyle data, and interaction data from interactions with their wellness coaches and physicians,” Zink continued. “We also pull in survey comments, focus group feedback, and other forms of data, such as biometric, EMR, and medical monitor data,” she said.
Zink went into detail around some of the components of the data and information that BCBS-Tennessee shares with its plan members, in order to engage them more deeply in their health and healthcare. “The consumer journey is important,” she stressed, after explaining some of the details of BCBS-Tennessee’s consumer engagement strategy and its data facilitation. “One of the things we have to realize is that the predictive models we use can’t be ‘one-off.’ The consumer will shut you down and will no longer pay attention to you if you send them too much information that’s not meaningful to them. So we put together actionable information, and determine where we can optimize predictive models and figure out what’s going on with a particular individual. Our predictive models tee up ‘cues’ to consumers’ behavior. And they rely on an integrated predictive platform.”
Progress is being made, Zink noted. “We’ve moved into that next phase of machine learning and algorithms, in which we’re able to note the ‘cause and effect’—the relationship between actions and outcomes.” For example, she noted, “We launched seven campaigns at the same time,” reaching out to plan members using different communication strategies. Experimenting to find out what might work best, she said, “Each group we reached out to contained a control group and a trial group. And within 60 days, we found out which of the seven campaigns were most successful. And the analysis told us where this outreach was working, and where it wasn’t. And one of the things we found out was that the hour of the day when plan members were reached out to made a huge difference. We found that calling in the morning and afternoon worked better.” And leveraging data analytics facilitated deeper analysis of patterns, she noted.
“How do we sustain engagement over time? One of the things about these control and trial groups and machine learning, is that it helps us to launch better member engagement campaigns,” and to benefit by learning more as those campaigns are rolled out. “Predictive modeling has helped us to figure out how to move things successfully to members. Now, we’re zeroing in on what the most impactful population management initiatives are, and we’re improving our outreach strategies” in that context.
As they move forward in these different spheres, Zink and Venkatesan agreed, it is very important to continuously improve data analysis processes, in order to gain more and more from the important initiatives moving their healthcare organizations forward.
