Over the last several years, Healthcare Informatics has had the privilege of covering the birth and toddler stages of the data analytics movement in healthcare. Our editors have spoken with the chief information officers and chief medical information officers of some of the pioneering organizations as they set up data warehouses and data marts, the robust business intelligence capabilities, and data governance and quality initiatives. The exciting thing now is that even though many organizations continue to struggle with setting up programs, those leading organizations are moving from talking about analytics to actually applying it to multiple use cases.
Keith Figlioli, senior vice president of healthcare informatics at Premier Inc., recently spoke with Healthcare Informatics during the meeting of his organization’s Data Alliance Collaborative, which includes representatives from 12 of the largest health systems across the country.
In the past year things have matured fairly rapidly, said Figlioli, who heads up Premier’s enterprise technology and data initiatives. “The discussions we have been having over the past three years have all been about big data and merging claims and clinical, for both operational and population health reasons. The discussions we have had over the last 48 hours have been so fundamentally different from even a year ago,” he said. The provider organizations are moving away from just setting up the infrastructure. In large systems, they invest six months to a year to set up the infrastructure to begin to do analytics. It takes a considerable amount of time to get enterprise data management and governance in place, he said, but these organizations are now beyond that. “You are seeing them proliferate the use cases throughout the whole organization,” Figlioli said. “It is exciting.”
One example Healthcare Informatics has highlighted is the 42-hospital Carolinas HealthCare System, which has devoted considerable resources to its efforts to leverage analytics capabilities to support population health initiatives. The organization has done some impressive work with predictive analytics on cutting readmissions related to chronic obstructive pulmonary disease.
But if the leading health systems are poised to make great progress, other organizations still have hurdles to get over. Most have developed some strategy they are trying to execute that involves an enterprise data warehouse, and combining claims, clinical and supply chain data, says Judy Hanover, research director at research firm IDC Health Insights. “They are trying to assimilate that information. Most have a number of different repositories and different degrees of success combining that information.”
Data Quality and Governance
Data quality is a huge issue, Hanover adds. Free-text notes in the EHR are unstructured, and the approaches to unstructured data are much further behind approaches that look at structured data, she says. “The value in unstructured data is clearly there for organizations that choose to tackle it, whether through text analytics or natural language processing.”
“It is astounding how many organizations are creating enterprise-wide data warehouses and dumping in massive amounts of data without knowing about the stewardship of each data source,” says Larry Yuhasz, senior director of innovation and population health at Truven Health Analytics. There are huge challenges around attribution, matching data to the right patient and the right physician, he says.
Figlioli agrees that data quality is a key problem. “We just interviewed 50 to 75 people, ranging from executives down to analysts, and the No. 1 issue they mentioned by far is data quality. It was two standard deviations away from all the other issues. Even somebody with a single instance of Epic or Cerner, and a single instance of an ERP, once they pull all that together, they start realizing how bad that clinical data is,” he says. Even more complicated is pulling data from affiliates into your system. “They are having all sorts of challenges intermingling with other pieces of data. It is a pretty complicated set of issues once you start getting into it,” Figlioli says. By far the No. 1 thing we talk about with members is data governance. We are seeing professionals come from other industries, such as consumer product goods and insurance, migrate to healthcare because they understand data governance.”
Analytics Talent Shortage
In fact, the dearth of data analysts is a massive problem, Figlioli adds. “It is the single biggest gap across the industry right now. Even if you get the infrastructure and data governance in place, then what?” he asks. “The data is served up. Do you have front-line analysts who know how to do this?” He jokes that if you changed your title to data scientist on LinkedIn, you would have 50 job offers by the end of the day.
One reason these jobs are tough to fill is that the executives need not only analytics training, but also clinical and operational expertise and skills in being able to wed together data from different functions, Hanover says. “They have to understand the origins and elements that are contained in that data and how to be sensitive to that and put it together in a way that yields meaningful results,” she adds. “That is a rare skill set, and most successful programs have a really skilled person at the helm.”
Hanover says the industry is starting to see new job titles such as chief analytics officer. “These new titles that are emerging reflect the role that analytics plays in bringing their business model forward.” The talent shortage is one reason there is a lot of interest in analytics as a service. Vendors can employ data scientists and leverage economies of scale to deliver data management to a number of organizations as a service, she says.
As the healthcare market transitions from fee-for-service to value-based payment, it will have a big impact on what analytics efforts measure and how they measure it, notes Truven’s Yuhasz. There is a growing focus on paying for episodes of care and avoiding readmissions, he says, but an episode could be two days or it could be six months for some chronic conditions.
“That temporal dynamic is very challenging for those designing analytics,” Yuhasz says. The data you use over time changes in terms of where it delivers value. Reading vitals in an inpatient setting every 15 minutes is crucial, but in post-discharge case management that is no longer of much value. “Creating the filters over time, by encounter and by user, is another interesting challenge,” he says.
Yuhasz also says both payers and providers have to change their mindset around sharing data. Health systems are going to have to get claims data. It is not particularly valuable at the point of care, but for managing performance it is incredibly valuable, he says. Health plans and commercial accountable care organizations (ACOs) need the clinical data. “This is where convergence is happening in experiments, where they are collaborating.”
In the fee-for-service world, the data itself was often considered a strategic asset. But in an at-risk world, even those who spent millions on an enterprise-wide EHR system do not have all the data on their patients. “And all of a sudden, the blind spots they have create tremendous risk, when they work with affiliate physicians or long-term-care providers,” he says.
Yuhasz says he sees customers just beginning to experiment with predictive analytics. The challenge is that in order to validate the hypothesis, you have to create a learning cycle, he says. “You have to say here are the assumptions of risk for this patient and pick an intervention. But how do you know the predictive model works unless you go back and assess? We see lots of examples of customers in the first blush of making assumptions, but not yet completing the loop.”
Figlioli says more organizations will start to feed the results of analytics to the point of care over the next few years. “That will pick up pace as the tolerance for data-based decision-making makes its way through healthcare,” Figlioli says, “but we are probably still in the first inning there.”
The Role of HIEs in Analytics
Shaun Grannis, M.D., associate director of the Center for Biomedical Informatics at the Regenstrief Institute in Indianapolis, Ind., spoke about the increasingly important role health information exchanges (HIEs) can play in analytics efforts. Grannis has been studying the intersection of HIEs and analytics with the Indiana HIE. His research has shown that patients do not receive all their care in one health enterprise. For instance, 40 percent of visits in Indiana to emergency departments are by patients who come from different healthcare systems.
“So If I want to understand healthcare utilization patterns, I need to integrate data from multiple sources,” Grannis says. “If I am an ACO and want to have a complete picture of the patients I am responsible for, I need integrated data from people who are sometimes my competitors. We have agreed that integrating data to improve outcomes makes sense. So our HIE in Indiana actually provides ACOs data for analytics purposes.”
Shaun Grannis, M.D.
In fact, he says, the ACO model has been a nice boost for HIEs, and opened up a new waterfront of opportunities of services to provide. “The ACO model is a pattern we will see repeat itself over the next several years,” Grannis says. “It requires data in an integrated, standardized fashion. We can provide predictive models to identify high healthcare system utilizers.”
Data quality is always a central challenge in the HIE space, he notes “We always want more complete data,” Grannis says. “When you look at aggregate healthcare data, there is heterogeneity there. We are doing work to accommodate the noise, pluck the signal out, and set realistic expectations. But we see nothing but continued improvement in that data quality.”
As Grannis put it, “There is substantial value in this aggregate data today and it will improve over time. The game now is finding those bright spots in current data where it shows real value. It will get better, and the value of analytics will continue to grow.”