For the last several years, Healthcare Informatics has made personalized medicine one of its top technology trends, and we are doing it again this year because the stakes are so high. As Christopher Chute, M.D., a Mayo Clinic bioinformatics researcher, told us in 2012: “Many of us believe that genomic information will inevitably transform healthcare beyond recognition. It will be a bigger breakthrough than antibiotics—not immediately, but in the next decade or two.”
Even if the commonplace use of a patient’s personalized genetic risk information to make clinical care decisions is still a decade away, the pace of progress is accelerating, with researchers at academic medical centers studying the integration of genomic data into the electronic health record (EHR) with relevant clinical decision support.
The early efforts of Penn Medicine in Philadelphia are illustrative of the challenges and opportunities of moving into personalized medicine. On the clinical side, all the data is integrated into a data warehouse, explains Brian Wells, associate CIO of health technology and research computing. But on the research side, there are islands of data that are not yet linked together. “We are working to combine that research data, including genomic, biobank and clinical trial data, and link it back to the phenotype data in the clinical data warehouse,” Wells explains. Researchers could find all the patients with a specific gene, and then see those patients’ clinical data; or they could find all the patients with a particular clinical profile and then look at their genetics, he says.
Penn Medicine has created a Center for Personalized Diagnostics to do somatic gene testing of solid tumors and blood-borne tumors. Researchers might look for 30 or 40 genes that they know are highly predictive, clinically significant genes. If they find them, they need to inform the clinicians.
Today, clinicians don’t want a lot of the details about the genetics. “They say, ‘Tell me the clinical significance and the drug I should use based on this patient’s genetic makeup.’ They want targeted answers in plain English,” Wells explains.
Some clinicians want that information to appear as clinical decision support reminders, and some don’t. “We have not built the infrastructure in informatics to take it out of the genetic sequencing process and pipe it right into the EMR [electronic medical record]; and there aren’t really standards in the industry for how you would communicate genetic results. It’s not like typical standards for most lab results,” Wells says.
“At Penn Medicine, we believe it all ought to reside outside the EMR,” he says. “You click on a URL and get a dynamic picture that is constantly changing about what’s significant and what isn’t. You want the genetics to be a snapshot, but the significance is a moving target. That’s why we believe it ought to be a web service or externally provided result that is dynamic.”
Josh Peterson, M.D., M.P.H., director of health information technology evaluation in the Department of Biomedical Informatics & Medicine at Vanderbilt University Medical Center, says his organization has been experimenting with offering clinicians pharmocogenomic information at the point of care along with clinical decision support. First the informatics researchers tried putting the clinically significant drug-gene interaction information in the labs section of the EHR. “Eventually it was put in the patient summary adjacent to a medication list,” he says. Vanderbilt also added a decision support tool within the EHR as well as a surveillance effort by pharmacists who can help interpret the finding for physicians.
The fact that many parts of Vanderbilt’s EHR are homegrown makes modifying it easier, “but we still have challenges getting the right data in the right place at the right time,” Peterson says. “For many of the clinicians, this is fairly new to them, and they say they don’t want to worry about the raw results. They want a distillation of what it means.”
The patient portal also provides a notification to patients that their genetic disposition may have implications if they take certain drugs. Patient education is another area that needs research and improvement, Peterson says.
A HOST OF DATA MANAGEMENT CHALLENGES
Health systems will have to consider what type of infrastructure changes personalized medicine will require, says Andrew Litt, M.D., chief medical officer for Dell Healthcare & Life Sciences. His company is building the IT infrastructure to support an FDA-approved personalized medicine clinical trial using gene-expression-guided therapy for pediatric cancers. The Translational Genomics Research Institute is using its genomics technology to determine the gene expression of children’s tumors and make the data available to teams of specialists who consult on treatment options. Dell built a supercomputer to get the computational time from three weeks to four days, Litt says, and it created a genomics cloud so that researchers at 16 different centers could easily access the data.
Health systems will need new ways to manage all the data involved in personalized medicine, Litt says. “They already have storage issues today and genomic data is an order of magnitude greater,” he says. “Hospital CIOs have to ingest it, store it, and then present it so clinicians can make use of it.” They also need appropriate clinical decision support at the point of care, he adds. “Most current EHRs have no way to present this data, so hospitals have to make a choice whether to build that in or link to an external source. I think it’s more likely they will do the latter. This is not an area of expertise for the EHR vendors. I think for all these reasons, the topic scares the heck out of most CIOs.”
If progress is being made in terms of presenting clinically relevant genetic information at the point of care, many gaps remain in terms of standards, integration, decision support and work flow. Speaking at the AMIA Symposium in Washington, D.C., last fall, Kevin Hughes, M.D., co-director of the Avon Comprehensive Breast Evaluation Center at Massachusetts General Hospital in Boston, described some of these gaps in greater detail. In the ideal world, he said, a clinician would pull structured data out of the EHR to support a genetic consultation. That would include access to decision support and risk algorithms to determine what might be needed for an individual patient. The clinician could send genetic test requests as structured data, including a structured family history, and get back a structured result, which could help feed a rapid learning health system.
“We don’t have any of this,” Hughes told the AMIA audience. Genetic lab tests are being sent back and forth on paper, he added. All that information is being stored as free text in the EHR, where it becomes unmanageable. He noted that although his EHR at Mass General may show in its notes section that a patient tested positive for the BRCA1 mutation, the clinical decision support section says the patient has no increased risk of breast cancer. “In the absence of structured data, the decision support has no idea that this patient is a mutation carrier,” he said. “Not only are EHRs not interoperable,” he added, “they can’t even talk to themselves.”
“We need clinical decision support, we need knowledge bases, and we need a rapid learning health system, and unless the data is standardized we will not get there,” Hughes stressed. “Health IT solutions must collect, receive and transmit standards-based family history and genetic data.” Guidelines and knowledge bases must be machine-readable and deployed as web services. “Closed, proprietary systems that are not interoperable are holding us back.”