6 Big Data Analytics Use Cases for Healthcare IT
Making use of the petabytes of patient data that healthcare organizations possess requires extracting it from legacy systems, normalizing it and then building applications that can make sense of it. That’s a tall order, but the facilities that pull it off can learn a lot.
BOSTON—The increasing digitization of healthcare data means that organizations often add terabytes’ worth of patient records to data centers annually.
At the moment, much of that unstructured data sits unused, having been retained largely (if not solely) for regulatory purposes. However, as speakers at the inaugural Medical Informatics World conference suggest, a little bit of data analytics know-how can go a long way.
It isn’t easy, namely because the demand for healthcare IT skills far outpaces the supply of workers able to fill job openings, but a better grasp of that data means knowing more about individual patients as well as large groups of them and knowing how to use that information to provide better, more efficient and less expensive care.
Here are six real-world examples of how healthcare can use big data analytics.
1. Ditch the Cookbook, Move to Evidence-Based Medicine
Cookbook medicine refers to the practice of applying the same battery of tests to all patients who come into the emergency department with similar symptoms. This is efficient, but it’s rarely effective. As Dr. Leana Wan, an ED physician and co-author of When Doctors Don’t Listen, puts it, “Having our patient be ‘ruled out’ for a heart attack while he has gallstone pain doesn’t help anyone.”
Dr. John Halamka, CIO at Boston’s Beth Israel Deaconess Medical Center, says access to patient data—even from competing institutions—helps caregivers take an evidence-based approach to medicine. To that end, Beth Israel is rolling out a smartphone app that uses a Web-based- drag-and-drop UI to give caregivers self-service access to 200 million data points about 2 million patients.
Admittedly, the health information exchange process necessary for getting that patient data isn’t easy, Halamka says. Even when data’s in hand, analytics can be complicated; what one electronic health record (EHR) system calls “high blood pressure” a second may call “elevated blood pressure” and a third “hypertension.” To combat this, Beth Israel is encoding physician notes using the SNOMED CT standard. In addition to the benefit of standardization, using SNOMED CT makes data more searchable, which aids the research query process.
2. Give Everyone a Chance to Participate
The practice of medicine cannot succeed without research, but the research process itself is flawed, says Leonard D’Avolio, associate center director of biomedical informatics for MAVERIC within the U.S. Department of Veterans Affairs. Randomized controlled trials can last many years and cost millions of dollars, he says, while observational studies can suffer from inherent bias.
The VA’s remedy has been the Million Veteran Program, a voluntary research program that’s using blood samples and other health information from U.S. military veterans to study how genes affect one’s health. So far, more than 150,000 veterans have enrolled, D’Avolio says.
All data is available to the VA’s 3,300 researchers and its hospital academic affiliates. The idea, he says, is to embed the clinical trial within VistA, the VA EHR system, with the data then used to augment clinical decision support.
3. Build Apps That Make EHR ‘Smart’
A data warehouse is great, says John D’Amore, founder of clinical analytics software vendor Clinfometrics, but it’s the healthcare equivalent of a battleship that’s big and powerful but comes with a hefty price tag and isn’t suitable for many types of battles. It’s better to use lightweight drones—in this case, applications—which are easy to build in order to accomplish a specific task.
To accomplish this, you’ll need records that adhere to the Continuity of Care Document (CCD) standard. A certified EHR must be able to generate a CCD file, and this is often done in the form of a patient care summary. In addition, D’Amore says, you’ll need to use SNOMED CT as well as LOINC to standardize your terminology.
Echoing Halamka, co-presenter Dean Sittig, professor in the School of Biomedical Informatics at the University of Texas Health Science Center at Houston, acknowledges that this isn’t easy. Stage 1 of meaningful use, the government incentive program that encourages EHR use, only makes the testing of care summary exchange optional, and at the moment fewer than 25 percent of hospitals are doing so.
The inability or EHR, health and wellness apps to communicate among themselves is a “significant limitation,” Sittig says. This is something providers will learn the hard way when stage 2 of meaningful use begins in 2014, D’Amore adds.
That said, the data that’s available in CCD files can be put to use in several ways, D’Amore says, ranging from predictive analytics that can reduce hospital readmissions to data mining rules that look at patient charts from previous visits to fill gaps in current charts. The latter scenario has been proven to nearly double the number of problems that get documented in the patient record, he adds.
4. ‘Domesticate’ Data for Better Public Health Reporting, Research
Stage 2 of meaningful use requires organizations to submit syndromic surveillance data, immunization registries and other information to public health agencies. This, says Brian Dixon, assistant professor of health informatics at Indiana University and research scientist with the Regenstrief Institute, offers a great opportunity to “normalize” raw patient data by mapping it to LOINC and SNOMED CT, as well as by performing real-time natural language processing and using tools such as the Notifiable Condition Detector to determine which conditions are worth reporting.
Dixon compares this process to the Neolithic Revolution that refers to the shift from hunter-gatherer to agrarian society approximately 12,000 years ago. Healthcare organizations no longer need to hunt for and gather data; now, he says, the challenge is to domesticate and tame the data for an informaticist’s provision and control.
The benefits of this process—in addition to meeting regulatory requirements—include research that takes into account demographic information as well as corollary tests related to specific treatments. This eliminates gaps in records that public health agencies often must fill with phone calls to already burdened healthcare organizations, Dixon notes. In return, the community data that physicians receive from public health agencies will be robust enough to offer what Dixon dubs “population health decision support.”
5. Make Healthcare IT Vendors Articulate SOA Strategy
Dr. Mark Dente, managing director and chief medical officer for MBS Services, recommends that healthcare organizations “aggregate clinical data at whatever level you can afford to do it,” then normalize that data (as others explain above). This capability to normalize data sets in part explains the growth and success of providers such as Kaiser Permanente and Intermountain Healthcare, he says.
To do this, you need to create modules and apps such as the ones D’Amore describes. This often requires linking contemporary data sets to legacy IT architecture. The MUMPS programming language, originally designed in 1966, has served healthcare’s data processing needs well, but data extraction is difficult, Dente says.
Service oriented architecture is the answer, Dente says, because it can be built to host today’s data sets—as well as tomorrow’s, from sources that organizations don’t even know they need yet. (This could range from personal medical devices to a patient’s grocery store rewards card.) Challenge vendors on their SOA strategy, Dente says, and be wary of those who don’t have one.
6. Use Free Public Health Data For Informed Strategic Planning
Strategic plans for healthcare organizations often resort to reactive responses to the competitive market and a “built it and they will come” mentality, says Les Jebson, director of the Diabetes Center of Excellence within the University of Florida Academic Health System. Taking a more proactive approach requires little more than a some programming know-how.
Using Google Maps and free public health data, the University of Florida created heat maps for municipalities based on numerous factors, from population growth to chronic disease rates, and compared those factors to the availability of medical services in those areas. When merged with internal data, strategic planning becomes both visually compelling (critical for C-level executives) and objective (critical for population health management), Jebson says.
With this mapping, for example, the university found three Florida counties that were underserved for breast cancer screening and thus redirected its mobile care units accordingly.