Across its four hospitals, 15 outpatient clinics, and a homehealth service, Springfield, IL-based Memorial Health System leverages EHRs from four different vendors. Its hospitals are all within about 30 miles of the capital city of Illinois, so patients often visit multiple locations, and sometimes their names are listed differently in each facility.
“I may be Tom Janssen here in Springfield, but I may be listed as Thomas Janssen up in our Lincoln (IL) facility,” says Thomas Janssen, Memorial’s manager of business intelligence and enterprise data warehousing. “One of our outpatient clinics may even have a record of me from my younger days registered as Tommy Janssen.”
By Neil Versel, Contributing Writer
By creating a master data warehouse and longitudinal patient record, Memorial Health System has the pieces in place to improve patient care and cut costs using analytics.
Across its four hospitals, 15 outpatient clinics, and a homehealth service, Springfield, IL-based Memorial Health System leverages EHRs from four different vendors. Its hospitals are all within about 30 miles of the capital city of Illinois, so patients often visit multiple locations, and sometimes their names are listed differently in each facility.
“I may be Tom Janssen here in Springfield, but I may be listed as Thomas Janssen up in our Lincoln (IL) facility,” says Thomas Janssen, Memorial’s manager of business intelligence and enterprise data warehousing. “One of our outpatient clinics may even have a record of me from my younger days registered as Tommy Janssen.”
Enter analytics. In 2012, Memorial made the decision to centralize its data in a common repository to “master” its medical and financial records, then apply business intelligence (BI) to this cleaned-up information.
“It was our desire to master that data into a ‘golden record’ so that we would have more clarity,” Janssen said. “That’s really the driving force behind bringing this data into a centralized warehouse and being able to report on it as a golden record, or a single version of truth, as we call it internally here.”
This “single version of truth” is what Memorial has begun to rely on for reporting purposes and for business intelligence. “We’ve bulk-loaded the records out of the back-end source systems, and then we have daily transactional records as they’re being gathered in those source systems that now feed into the warehouse,” Janssen said. That serves as the “foundational layer” for business intelligence, he explained.
Prior to embarking on this data warehousing and business intelligence initiative, Memorial used Access databases and Excel spreadsheets as source data. Frequently, Memorial executives were making business decisions based on the information (often third- or fourth-generation data) contained in these documents. This was problematic.
“Every time a human touches data, there’s a possibility that numbers get transposed. It introduces error into the equation, so one of our purposes for a centralized enterprise data warehouse is that we wanted to start reining in some of those Access databases and Excel spreadsheets and point our business units to a single warehouse, so everybody can pull the data out of one central repository,” Janssen said. “The first step in any successful business intelligence initiative is to ensure everyone is referencing the same data.”
Scrubbing Health Data Is An Evolutionary Process
In December 2012, Memorial brought in BI vendor Information Builders to develop the data warehouse, master the data, clean up the records, and purge duplicates. For reporting, the health system uses a single-purpose data computing appliance called Greenplum, formerly a product of EMC but now offered by an EMC-VMware joint venture called Pivotal Software.
“It looks like a large refrigerator,” Janssen said of Greenplum. “It’s a purpose-built appliance that has the servers and data necessary to store our warehouse and conduct BI processes on top of that. You don’t have to bring multiple vendors in and try to piece the components together.”
Memorial is about two years into the process and has mostly completed the bulk-loading of data into the informatics engine. “We are reviewing that data now, so they have significantly compressed that timeline for us,” Janssen said.
As of December, Janssen considered the data in beta status, after going through pre-alpha and alpha phases. Memorial continues to analyze the data and assess the various test-case scenarios that it’s running through the informatics engine to make sure the records coming out of back-end systems are merged properly and mastered correctly.
The technology is not yet available to everybody. “It’s in the hands of those folks whom we’ve worked most closely with in determining what the merging rules and mastering rules would be. They’ve been the ones who have been with us in providing the guidance all along,” Janssen said.
“We have weekly meetings to review any of the findings, but we wanted to make sure that all of that was rock-solid before we put it in the hands of the executives.” At press time, Janssen, who reports directly to Memorial CIO and CMIO Dr. David Graham, expected this effort to be complete by early February.
The process has not been completely smooth, of course. “We often find anomalies in the data,” says Janssen. “Sometimes the systems don’t actually edit-check the data, so there is some free-form text present that can be difficult to discern and work through. We’ve had instances where fields we thought would be computable data are actually in PDF form because the document itself was scanned into the system,” Janssen said.
Janssen admits the process is evolutionary. Over the long term, Memorial will be able to analyze PDFs, dictated recordings, and images using its BI platform. However, the provider is focusing its efforts on targeted data sets to start.
For example, in its beta testing, Memorial has been tracking data elements such as patient demographics and clinical encounters. “We’ve broken this out into a number of domains,” Janssen says. “There’s the patient domain — everything about a patient that we would need to know.” The provider domain includes not only a list of doctors each patient saw, but also other clinicians that may have been working with the direct care team.
Clinical records form the basis of a longitudinal medical record, a history of the patient over time. “It’s creating a history of the patient, the physicians that they’ve seen, the nurses and workforce that attended to them with each of the encounters, and the financial aspects of it — from what it cost the patient to our internal costs,” Janssen said.
Accurate Health Data Comparisons Drive Action And Improvement
Memorial, including its flagship 500-bed Memorial Medical Center and three affiliates, two of which are critical-access hospitals, wants to track costs by facility. “If we wanted to look at how one of our critical-access hospitals is doing and compare it to another, we could do that. If we were to specify floor, wing, and bed, we could drill down into that level of detail, or we could compare one nursing unit to another,” Janssen explained.
Comparisons can be made based on several metrics, including service delivery, throughput, and patient satisfaction. “We are able to drill down into that data and see what it represents,” Janssen said.
Even before bringing the technology into full production, Memorial is looking to move into the next phase of analytics, specifically the realm of predictive analytics, according to Janssen. Like health systems all over the country, Memorial seeks to prevent readmissions and reduce “frequent fliers” to the emergency department by identifying patterns.
“If we see that a patient has come in [during] January, April, and July, we would want to be able to look at that data and make predictions based on the data that’s there so we could make, with some certain degree of accuracy, a calculation that the patient, for whatever reason, may present again in October,” he explained.
“That’s an important step to get us to what we’re really after, which is what’s called prescriptive analytics,” Janssen said. “What we’re ultimately after is the ability to provide better patient care at a lower cost to the patient and the health system.”
Prescriptive analytics will be able to help Memorial understand why the hypothetical patient may have visited the ED in January, April, and July. “We could really drill down into the data then and find out that they’re presenting with heart attack symptoms. That’s good for us to know as to why they may be coming back in October,” Janssen said. “It’s not just a cool little toy for IS [information systems]. It’s a game-changer.”