One hospital’s embrace of big data
By Barb Darrow
The University of Pittsburgh Medical Center is a microcosm (albeit a large one) of the big data problem facing medical organizations. Hospitals are under intense pressure to cut costs, but they are simultaneously expected to modernize their IT and maintain their patient care.
Until now, most medical organizations were focused on getting their electronic medical record (EMR) systems running. These systems digitize the paper-intensive world of traditional medicine. The problem is that many organizations, including UPMC, picked the best EMR for certain medical specialties and now must deal with a welter of systems that don’t talk to each other very well. The worst part? EMRs represent only a portion of the data hospitals generate.
The goal is to somehow create one complete electronic record for patients that includes all their data (images, pharmacy records, clinical notes, self-reported patient information), regardless of underlying system and to combine that information with relevant financial, genomic, and research data to provide a holistic view within the EMR, said Lisa Khorey, the VP of enterprise services and data management at UPMC.
Dr. Rasu Shrestha, the VP of medical information technologies, is intimately familiar with the challenges of managing vast amounts of medical data, as well as the opportunities for hospitals that do it well: He is both an M.D. and the IT guy who has to make it happen.
UPMC and its affiliates make up a $10 billion entity, with 55,000 employees at more than 20 academic and regional hospitals. Its health plan has close to 1.5 million members and a network of 125 hospitals. It encompasses both the health care payer (insurance company) and provider (hospital). Those two pools of data represent what Shrestha calls the “yin and yang” of accountable care.
The EMR work started years ago and is fully deployed, but this phase of bringing all the data together in a unified, meaningful, record at the point of care continues. UPMC’s contract with Nuance Communications, for example, was signed last year and is a ten-year commitment.
The stakes are huge. For one thing, better patient care depends on this secure flow of pertinent patient data to the clinician. If UPMC’s Medicare Advantage plan meets certain quality and performance standards set by the Centers for Medicare and Medicaid Services (CMS), it could reap an additional $40 million to $50 million in pay-for-performance bonuses. Effective use of big data is a prerequisite to that payoff. Systems need to be in place to make sure those quality metrics can be measured and achieved.
UPMC also fields 31 clinical systems for various specialties: Cerner for inpatients, EpicCare for ambulatory care, Allscripts for affiliates and Varian for oncology. The doctor for any given patient might need information from many of those sources to get to the full story.
Of course, there are non-IT issues in all of these moves. Older clinicians are not necessarily thrilled with using new technology, but those barriers are falling as younger, tech-proficient doctors come up the ranks.
UPMC is certainly not the only hospital struggling with these issues, but a Nuance executive that works with many health care organizations said UPMC is well ahead of others in terms of moving beyond EMRs.
Doctors now try to take a more holistic view of their patients, and that requires the ability to pull together data from different sources. Imaging data is separate from surgery notes, which is separate from pharmacy data.
“If we look at big data, the idea is how to interconnect multiple points of data across the broad, biological continuum,” Shrestha said. “If the patient is diabetic, you don’t just see an endocrinologist looking at the liver in terms of liver function tests or any scans but across the biological spectrum of organs and then down to a cellular level. We look at pathology slides, reports on molecular imaging and down to the genomic levels.”
As the data gets more granular, the data set gets bigger. “We start talking about gigabytes per image. We have to get ready for this tsunami of data,” he said.
For large health care organizations, data proliferation means doctors have to spend time corralling the information. “We play detective more than anything, piecing together information as we interact with the patient,” Shrestha said in a recent interview. “We’re trying to make sense of all this by figuring out how to get to where I, as a clinician, get to be more a clinician and less a detective.”
Three big buckets of data
Data volume is one thing, but the diversity of data is another. Shrestha puts it in three buckets. First there is all that imaging: CAT scans, PET scans, MRI scans. UPMC has close to 2 PB of total archival data at this stage and nearly 1 PB of imaging data.
The second bucket is traditional structured data sitting in relational databases. That represents about 20 percent of all non-imaging data. This includes structured data both within the EMRs and in standard databases that underlie key operational systems like accounts payable and receivable, the transaction systems that run the hospital.
The third bucket — the remaining 80 percent of non-image data — is unstructured. This includes postoperative notes, radiology reports, discharge summaries. Some are written or typed. Much is spoken or dictated and has to be transcribed and digitized.
UPMC is still evaluating some of the technology that it will use, including IBM/Cognos (s ibm) and Oracle (s orcl) databases and related tools. It is using DBMotion to handle much of the EHR integration, and Nuance’s’ natural-language-processing technologies to deal with the voice-to-digital transcription process at the front end and act as a searchable back-end repository for that data once it is processed through the Clinical Language Understanding (CLU) engine.
UPMC has aggregated data from those 31 provider-based systems, and there have already been payoffs, Shrestha said. Because doctors can see not only what has been prescribed across all EMR systems but also the claims information, they know which prescriptions have been filled. “This is all about filtering the noise to get data that is actionable. In this case we won’t treat the patient with a pseudo condition of acute abdominal pain but get to the root of his problem, which is probably opiate addiction,” Shrestha said.
But there could be bigger payoffs down the road, as machine learning can be applied to historical patient data to extract medical insights that would otherwise never be found. “If you take all those patient records and run algorithms and machine learning tools against a 20-year set, it will find things we wouldn’t have found otherwise. Some will be valuable,” Khorey said.
UPMC plans to launch a digital pathology PACS (picture archiving and communication system) from Omnyx — a company formed by UPMC and GE Healthcare) that will generate more imaging data. These systems basically put that biological or chemical data from blood tests, cell cultures and so on into digital format. That is the next data tsunami UPMC is prepping for.