The University of Pennsylvania in Philadelphia has a large clinical enterprise, Penn Medicine, and a highly regarded School of Medicine. But integrating data from the two for research purposes had always been a challenge. Even though Penn Medicine has had a clinical data warehouse for several years, the research data resided in islands of siloed databases using myriad data formats.
But the 2014 launch of PennOmics, a research data warehouse serving the hospitals of the University of Pennsylvania Health System and the Perelman School of Medicine, has given physicians and academic researchers access in one platform to massive amounts of de-identified, aggregate patient data from electronic health records and cancer genomics data from six formerly stand-alone systems. For its vast potential to contribute to the field of translational medicine, the editors of Healthcare Informatics have selected Penn Medicine as the co-third-place winning team in the 2016 Innovator Awards program. As with all such projects, there were critical decisions and cooperative efforts required to make this broad, enterprise-wide initiative happen. Here is Penn Medicine’s story of innovation.
In 2011, Michael Restuccia, the vice president and CIO of Penn Medicine, and Brian Wells, the health system’s associate vice president of health technology and academic computing, held a series of meetings with clinical researchers to understand their challenges in terms of technology. “What we heard about was a lack of access to integrated data,” says Wells. “They were unable to join their research data with the already consolidated health system data. As we talked, it crystalized in our minds that we needed a tool that could bring it all together.”
One of the driving forces was the team that managed the tumor registry. Although they had a robust registry, they had no way to query it, analyze it, and discover things within it. “That was a big driver,” Wells adds. “We brought in the tumor registry, a lot of genetic data, biobank data and clinical trials data. We needed a platform where we could join that data together, de-identify it and offer self-service access.”
Restuccia calls the PennOmics effort a good example of the information systems group enabling research and patient care capabilities based on what researchers tell them is important. “We heard loud and clear there was a void in having a repository to store the data in an integrated manner.”
(From L to R:) Brian Wells; Michael Feldman, M.D., Ph.D.; Mike Restuccia
Once the need was identified, the next step was to approach a senior IT council made up of six representatives from the health system and six from the school of medicine. “We bring recommendations to this group, and they determine funding, priorities and pace,” Restuccia explains. One of the first key decisions that had to be made was whether to build the data warehouse themselves or work with a vendor’s solution. “Our preference here at Penn is to build. We like to have control,” Restuccia says, “but in this instance, building it would have required significant investment in bioinformaticians’ time, energy and know-how, and that excess capacity just didn’t exist. Meanwhile, Oracle had a pretty strong data model for what our identified need was. So we decided to buy the solution vs. developing it in-house.”
The next step was to form a PennOmics governance committee made up of researchers working in fields such as genetic sequencing and high-performance computing. “We met monthly and they helped us clarify important data sources and data types and gave feedback to Oracle about their product and what it needs to do,” Wells explains. A few researchers let the project team load their genetic data to do a proof-of-concept to prove that the data-loading process worked and that researchers could query it on the back end.
Penn Medicine has a new Center for Personalized Diagnostics, a clinical laboratory that does somatic tumor testing of targeted genes that are clinically significant and actionable. A big task was loading the center’s data into the warehouse and making sure it flows correctly. Now data about all the patients sequenced flows into the PennOmics warehouse every two weeks.
Today PennOmics is proving of critical value to researchers studying clinical effectiveness by doing retrospective analytics on aggregated data. For example, a researcher studying a population of patients with the BRCA1 gene mutation can study treatment regimes and outcomes. Or by evaluating individuals with similar medical profiles, a physician might conclude that one blood pressure medication would be more effective than another. Using PennOmics, physicians can ask questions of the data, such as “find all lung cancer cases with EGFR-activating mutations that have failed primary EGFR therapy, with disease-progression presenting as new metastatic disease.”
“That all feeds back into the decisions you make in the clinical area about what is the best drug or treatment,” Wells says. “It all ties together. That is translational medicine at its heart.”
Penn Medicine was the second customer to sign up for a cloud-based service called the Oracle Health Sciences Network that gives its partners the capability to query Penn Medicine’s de-identified data and find populations to meet their needs. Researchers can determine if there are enough patients at Penn to support a specific clinical trial. “What you are doing for external sponsors and internal investigator-initiated trials is preventing them from starting a trial that is going to cost a lot of money and never yield enough patients to have accurate results,” Wells says. For instance, in a recent collaboration with a private-sector bioanalytical lab, Penn Medicine’s Office of Clinical Research was able to quickly identify patients for two upcoming clinical trials.
Self-Service Data Queries
The researchers themselves say PennOmics is making their investigations faster and more self-service. Scott Damrauer, M.D., a vascular surgeon, says that previously he would have to send a request to the Penn Data Store team and have a programmer there search SQL databases. The process could take weeks to months. “Now with PennOmics I can log on and search the data myself in minutes. It’s allowed individual investigators to query the data directly,” he says. For example, he recently used PennOmics to assess the feasibility of a sponsored clinic trial in which a positive troponin following vascular surgery was one of the primary outcomes. By determining the number of patients who had troponin lab values in the 30 days after undergoing an index procedure and then exporting the lab values for analysis, he was able to determine the likely event rate. This information was important in helping to determine the power of the study and the necessary enrollment. His team has subsequently secured the study contract and started the trial.
Michael D. Feldman, M.D., Ph.D., associate professor of pathology and laboratory medicine, says more data is gradually being moved into the PennOmics platform in the area of precision medicine. “We have a reliable pipeline now for moving our next-generation sequencing data for oncology into PennOmics and are beginning to see researchers depositing their data from samples that they have run on clinical specimens back into PennOmics,” he explains. “We are in the process of linking our biosamples. Each step begins to execute a long-term vision that allows Penn Medicine to bridge the bench to bedside, where we can identify patients and biosamples to test hypotheses being generated in animal models. That is the real value here: using this to help Penn become an even better learning health system.”
Feldman says Penn is even beginning to link clinical practice and outcome data to genetic variance within tumors. As Penn delivers lung cancer care, it can examine the utilization of care if the patient’s tumor has a specific variation. “You can dissect health system processes and refine those because you now have the ability to look at that business data in combination with genetics,” he says. “It will allow us to refine how we deliver care based on genetic variation within disease groups. I think that is a pretty novel way of thinking about precision medicine.”
A Valuable Recruiting Tool
Besides significant time savings in identifying patient cohorts for clinical trials and clinical care treatments, Penn Medicine says PennOmics serves as a valuable tool for the recruitment of biomedical and research informatics experts. “It is definitely a draw,” Wells stresses. “We are in the process of recruiting a chief research information officer. Many of the candidates are attracted to Penn because we have already laid the groundwork for this type of work.”
Feldman agrees. “Being able to demonstrate to job candidates how tightly the health system and the school of medicine are working together in this translational space makes it so much easier to convince somebody that this exists and is real,” he says. “It is not just smoke and mirrors.”
Restuccia says PennOmics is part of a larger effort to create common data platforms for the School of Medicine as his team has already done on the health system side. Previously, decision-making about IT tended to be decentralized, so laboratory information management systems from many different vendors were used, and some were even homegrown. “The approach now is to get everything in one system vs. having dozens of little systems out there,” he adds. “That allows for aggregating the data, and searching it in a more cohesive, secure manner. Our approach is the same now on the School of Medicine side as it was on the health system side.”
One of the challenges in the first year after launching PennOmics was just getting the word out to internal researchers about its capabilities. Communications in academic medicine can be difficult, Restuccia says. “It is not like a health system where everything filters down from a CEO.” Wells recalls a recent meeting with neurological researchers: “We got applause at the end of the meeting. They were shocked that this existed. They don’t think it can be done. We tell them it is possible and sometimes they still don’t believe us.”
As part of an internal marketing effort, the PennOmics team held a contest in 2015 challenging researchers to use its Cohort Explorer tool to perform creative and purposeful research of Penn Medicine patient data in genomics or clinical categories. The top submission in each category received $10,000 towards professional expenses.
Although PennOmics started out with a small user base when it was first introduced in 2014, it has grown to several hundred consistent users.
“Our ongoing challenge is to continue to drive value out of big data,” Restuccia says. Looking ahead, one area of focus for 2016 is adding imaging metadata to PennOmics, Wells says. “Researchers will be able to ask: Do I have a patient with this disease, this DNA, and this modality in our imaging system?”
Another effort has been around natural language processing (NLP) to search unstructured data. “Although we have a search tool that has been helpful in accelerating searches,” Restuccia says, “we haven’t put in place the NLP yet, but that is going to come rather soon. That will be a big step toward driving value.”