Big data benefits accelerate at Catholic Health Initiatives

Register now

Catholic Health Initiatives makes no small plans, because it really can’t—the Englewood, Colo.-based health system has more than 85,000 employees and operates 86 hospitals in 18 states, along with 40 long-term and assisted living facilities as well as multiple nursing colleges, home health agencies and academic medical centers.

To steer such a huge organization in a new direction takes a great deal of time, resources and coordination. And it also requires data—big, complex, ceaseless streams of information coming in from an array of different information systems and care settings.

However, Catholic Health initiative has put in place a big data analytics platform that is achieving reductions in mortality, hospital-acquired infections and events that are harmful to patients, according to James Reichert, MD, Catholic Health’s vice president of clinical analytics.

“The end goal is to use analytics to decrease variation in healthcare and optimize clinical outcomes so that, no matter where you go in our system, you can be assured to get the best care possible.” says Reichert. “Our efforts to accomplish this require a lot of data wrangling as we transition into a truly data-driven organization.”

Catholic Health Initiatives’ current big data focus is on population health, but it takes a slightly different tack than many of its peers. Instead of only trying to collect bits and pieces of information about what’s going on with its patient population outside its walls, the health system is also focused on improving its inpatient performance using the value-based purchasing metrics created by the Centers for Medicare and Medicaid Services: decreasing mortality rates and hospital-acquired infections, and improving satisfaction for patients in acute-care settings.

“Most population health efforts focus only on getting data about what’s happening to patients in the outpatient settings, but we’ve found that we needed to focus on both inpatient and outpatient care, and use data that exists within our systems and that we have access to—in most cases, we cannot get reliable access to the data from affiliated physicians and the multitude of EHRs they’re using,” Reichert says. “We feel that by improving our performance in our environment we can make the biggest impact in terms of overall care quality and costs.

“What’s unfortunate is that the healthcare community is treating population health as an outpatient phenomenon, when it reality, it is both and needs to be treated as a continuum. They consider what’s happening out there as separate to what’s happening in the hospitals. But they’re both part of the same care continuum.”

While Catholic Health Initiatives is focused on the “inside” of its clinical operations, it’s conversely looking outside its data infrastructure to fuel its population health efforts. Instead of trying to wrangle all the data streams itself, it has partnered with a handful of vendors to normalize and standardize its information, which is then fed into its data warehouse to create single sources of truth for the primary data it uses for population health analytics.

The organization has a contract with Premier, Charlotte, N.C., to clean Catholic Health’s administrative data; Nashville, Tenn.-based HealthStream normalizes all the data around patient experience and then benchmarks the information against national and regional scores for Catholic Health facilities.

For hospital-acquired infections, the health system interfaces with the National Health Safety Network operated by the Centers for Disease Control and Prevention. Catholic Health uses SAS tools to wrangle, analyze and deliver standard reports across the enterprise, as well as identify problem areas, incorporate new protocols into its IT infrastructure and benchmark performance.

Underpinning those efforts are tools from Carmel, Ind.-based Clinical Architecture that interpret the variations in clinical data and expressions and maps the information to clinical standards such as LOINC lab codes and other data dictionaries.

“One of the biggest challenges has been the incredible variance in medical expressions,” Reichert says. “A single lab concept, such as a hemoglobin level, can be stored in 50 different ways, and how it’s documented varies widely amongst our EHRs in different regions,” Reichert says. “When I talk about data wrangling, that’s a good example of the challenges of getting usable data we can surface up into our analytics program.

“People typically think of data as coded or free text, but we’ve found most of it’s on a continuum—some data is very clean and standard, but most of it is somewhere between the two,” he adds. “You have a lot of work to do to interpret it and then put it in a form that can be used.”

To do so, Catholic Health utilizes a “late binding” approach; that’s strikingly different from the more commonly used “early-binding” strategy in its data warehouse. The health system uses data mining and visualization tools, including SAS Enterprise Guide and Visual Analytics, from Cary. N.C.-based SAS.

In the early-binding approach used by most industries, such as banking and retail, data flowing into the warehouse is quickly related based on business rules and vocabularies. Those data dimensions are relatively simple to identify and link, and in early-binding strategies, all the relationships would be resolved after the data hits the warehouse so it can be provisioned and used to support analytics efforts.

By contrast, the late-binding approach enables organizations to move data from source systems into a warehouse without trying to transform the data upfront by changing its formatting to make it usable for specific purposes and committing it to a relationship. That transformation and binding happens later, when the specific data is needed for an analytics effort.

Minimal data transformation occurs until that data is then linked to a “data bus” comprising a small number of core data elements that are common to almost all analytics use cases in healthcare—these include patient ID, provider ID, date and time, facility ID and others.

The output of the efforts is hospital and physician performance data that are provided in dashboards and PDF reports and utilized by senior-level executives to improve performance. The performance of each hospital is baselined and benchmarked, then discussed at board-level meetings and operational performance reviews. Based on the data, the different hospitals and divisions are asked to perform root-cause analysis and then come up with strategies to improve performance.

When the analytics effort first started at Catholic Health Initiatives, measurable improvement in outcomes was slow and inconsistent, but it has accelerated over time.

The key has been the health system’s strategy to utilize the analytics to identify each facility’s key areas of weakness and developing a long-term focus to address those deficiencies. For example, using transparent, risk-adjusted data, Catholic Health found that five facilities accounted for 60 percent to 80 percent of the clinical opportunity for improvement. Based on those findings, a long-term improvement focus was developed—a three-year plan with yearly resetting of the baseline and goals, but maintaining the same metrics year over year.

Reichert adds that the two key elements that result in success are that executives and clinical leaders trust the data, and those same leaders understand the information at which they’re looking. “The reason we’ve spent so much effort on standardization and normalization is to establish a single source of truth for the different buckets,” he says. “That’s why we baseline and benchmark and risk-adjust—we have such a large geographic area we cover and so many different types of data and technology that we have to ensure that the information is solid.”

Catholic Health Initiatives requires senior-level executives to go through formal online and classroom training for analytics as well as medical concepts. “We found that few executives had any kind of training about how to interpret regression analyses, observed to expected ratios, inclusion/exclusion criteria, among others, as well as medical terms,” Reichert says. “And not every executive is going to know what a CLABSI {central-line associated bloodstream infection} is, for example. Too many organizations in healthcare don’t really consider if executives have the knowledge to really understand the information they’re being asked to make decisions with.”

For reprint and licensing requests for this article, click here.