CHICAGO – Rush University Medical Center has been sequencing cancer genomes for clinical decision-making for close to a decade. Only since the spring of 2019 has the academic health system been able to get discrete, structured data from test results into its electronic medical record to inform downstream clinical decision support and secondary analytics.
The University of Pennsylvania Medical Center has had a similar experience. Both Penn and Rush are customers of Epic Systems, the EMR vendor that a year ago released an optional genomics module for its core platform.
"Our goal is that eventually everybody will want to and be able to incorporate computable genomic data into patient records, but not everybody is ready to take that step yet," said Peter DeVault, Epic's vice president of genomics and interoperability.
About 50 organizations have licensed the optional set of features and 10 have gone live, though the usage number can be a bit fuzzy, according to DeVault.
"We have so many groups that have been doing this for many years, not necessarily with our genomics module. Some are using features that are already available, already part of the Epic EMR, and others that are incrementally taking advantage of some of the new features that we have," he explained.
Verona, Wisconsin-based Epic does not issue press releases, and generally relies on its users to disseminate news. Rush started publicizing its use of the Epic genomics module about a month ago, while Penn has been talking up its experience at conferences of late. DeVault did not disclose the names of other early adopters.
Epic for several years has had customers incorporate genetic data into patient records, but everyone seemed to be doing so in different ways. "What we wanted to do with the genomics module is standardize that as much as we can without inhibiting discovery and innovation," DeVault said.
"We're trying to help liberate genomics from the PDF report," DeVault said. That means converting flat results documents into discrete, computable, data that resides in the patient record and helps users act upon that data. In doing so, the vendor has built new data structures for genetic variants into the patient record and has adopted version 2.5 of the Health Level Seven International (HL7) communications standard, the first to include support for genomics.
"This is the first time that we are able to get the results into Epic in a structured way such that they can be used for downstream decision support and analytics and patient care," said Mia Levy, the Sheba Foundation director of the Rush University Cancer Center and codirector of the Rush precision oncology program.
"In the past, the best we were able to do was to get a copy of the report as an image file and upload it into Epic, and you'd be lucky if it was indexed."
Genomic data, of course, is more complicated than other types of information in patient records.
"Ironically the genomic information starts out in a digital format," Levy noted. "It has to be transferred into a PDF in order to get into the medical record when it was digital to begin with."
Levy said that the module "facilitates our ability to have a complete roundtrip workflow" for care teams. The EMR now has electronic pathways for order entry, delivery to the lab, and return of results.
Rush uses Tempus for its next-generation cancer sequencing and genomic profiling. But prior to the EMR upgrade last year, test results would come back as PDF files, so a bioinformatician would have to process those files to make them suitable for analytics.
"Now at Rush we have that structured information which has been curated by a molecular pathologist at Tempus," said Chief Research Informatics Officer Casey Frankenburger, who was head of computational biology at Tempus until he returned to Rush in December 2018.
Users still see PDF reports in the EMR, but can click to get structured data and recommended courses of action. The interface also provides access to the Tempus portal so clinicians can see how a specific patient's gene expression compares to population cohorts.
This hybrid setup means that Rush now gets back both raw and curated data. "We can export that data for both operational and research analytics," Frankenburger said.
In customizing the Epic installation, Frankenburger and the clinical oncology team mostly worked on how information is displayed to each type of user. Tempus programmers worked on coding results to the HL7 2.5 specification and Rush validated that work.
Frankenburger said that his plan was "meeting the physicians where they're at" to help them learn this burgeoning field of genomics.
He created order sets within Epic, groups of tests that oncologists might order according to the type of tumor, such as a PDL1 assay for immunotherapy or a DNA mismatch repair test in addition to the core 596-gene Tempus xT gene panel. "It's all bundled in the way that they typically order," he said.
If results from different tests come back at different times, the EMR then knows to tell the care team to wait on additional reports.
Physicians are actually not the primary beneficiaries of the technology. To avoid physician pushback and complaints of alert fatigue, Rush so far has made front-line genomic clinical decision support available to nurses, genetic counselors, and pharmacists, but not oncologists. "And then if there is actually a signal there, we then move it along to the physician directly," Levy said.
That could change as the program matures and the leadership team evaluates early results, she noted.
"If there is any question by the physician, they can always submit that to the tumor board for discussion," Frankenburger added.
The test results come in as structured data and curated analyses, with pathogenic variants and important, actionable findings presented first. For example, if the test turns up a BRCA mutation, the EMR alerts the care team that the patient should be set up with a counselor.
"Even beyond that, we created best-practice alerts," Frankenburger said. In the BRCA instance, a genetic counselor will automatically get a message. The system also flags records showing biomarkers related to clinical trials available at Rush.
The EMR also sends alerts for rare mutations which have a useful drug available that the oncologist might not immediately think about. Epic, like many of its competitors, has built pharmacogenomics into its EMR because there is considerable demand for that service among hospitals just launching clinical genomics programs.
Rush also has been trying to match up test results with medical knowledge on somatic mutations to create actionable alerts.
Penn has taken a similar path, adopting clinical decision support as part of wider effort to integrate genomic data into the EMR in order to bolster clinical genomics. "This is only one aspect of an overall push that we're doing to be able to better integrate it and be able to utilize genomic data," said Katherine Nathanson, deputy director of the Abramson Cancer Center at Penn's Perelman School of Medicine.
Nathanson said that all genetic data on patients goes into the EMR in a way that makes the information easily searchable. Penn also has set up a specific tab in Epic for precision medicine.
Penn relies on Ambry Genetics for its sequencing, and has a direct pipeline from the Ambry laboratory that feeds into the genomics module in Epic.
"When we place orders in Epic, it goes directly to Ambry Genetics into their Ambry portal to trigger the testing, taking the insurance information with it for precertification on their end. And then when they send their reporting back, it hooks directly in and goes in to directly back into our EMR," Nathanson said.
Currently, Ambry sends the reports as PDFs, but Penn is now working with the testing lab to be able to receive discrete data in the genomics module, which will then allow genetic indicators to inform the clinical decision support engine. The goal is to have the second phase complete by summer.
Penn has built standard operating procedures for whenever manual entry is required for the genomics module. "We need really consistent SOPs so that everybody is putting in the data exactly the same way," said Nathanson, a medical geneticist specializing in BRCA-related research and the former chief oncogenomics physician at Penn until she got her current job.
"The SOPs are critical because when we move to phase two with Ambry, everyone has to agree on how the data will populate," she said.
Convincing clinicians to change their habits is always fraught with resistance, but Nathanson said that the move to the precision medicine tab for all genetics information was "incredibly positively received." The same was true for the integration of the Ambry portal, which also predated the adoption of the Epic genomics module.
In customizing its own Epic build, Penn had to create an exception to its policy of automatically releasing test reports to patients through the EMR portal within three days of receipt. Genetic test results from Ambry currently are manually released after a counselor signs off.
Much work remains for all of Epic's clinical genomics customers.
The software company is building a dedicated, cloud-based sequence server to store large genomics files rather than putting raw sequencing data into patient records. The EMR itself has a "home" for genetic information, including what DeVault called "actionable variants," meaning variants with thresholds backed by clinical evidence, including pharmacogenomics and risk factors for cardiovascular disease and cancer.
With the sequence server, Epic is currently prototyping several bioinformatics tools, starting with the ability to look up variants in a knowledgebase.
"One of the things that's especially exciting about what Rush has done is having that live interface for discrete oncology, somatic genetic results," DeVault noted.
He also said that Epic is prototyping variant look-up in the sequence server with ClinVar. "But we're doing that in a neutral way so that eventually you can use that with other knowledgebases," DeVault said. That capability should be out later this year.
Rush and Penn are not using the new genomics module for pharmacogenomics just yet, though DeVault said that PGx is the most common clinical application of genomics among Epic users. The vendor has incorporated the Clinical Pharmacogenetics Implementation Consortium's Level A guidelines into its core EMR.
Mayo Clinic and Geisinger Health System have recently had journal papers published about their use of pharmacogenomic clinical decision support in Epic. The EMR supported PGx before the release of the HL7-based genomics module.
Epic does not yet support Fast Healthcare Interoperability Resources (FHIR) Genomics, a component of FHIR version 4.0.1, which HL7 published Oct. 30. FHIR Genomics was still in draft form when Epic released its current genomics module.
FHIR Genomics is from the same HL7 clinical genomics workgroup that produced the genomics module in HL7 version 2.5. Epic has started working on incorporating FHIR Genomics because its customers have been asking for it, DeVault said.
"To us, it's largely a matter of a customer's preference, a way of getting data from one place to another," DeVault said. "From that point of view, neither standard has particular advantages."
He said that HL7 interfaces work well for making permanent connections between the EMR with laboratories, while FHIR is nimbler for ad hoc data exchange. "What we're looking forward to doing with FHIR is being able to better accommodate the kind of landscape we're starting to see where one healthcare organization might work with a dozen different genetic laboratories, sometimes just for one test apiece."
That will come in a later release. DeVault did not discuss the timeline for that.
Frankenburger and Levy of Rush are in the process of writing up an application note about comparing the HL7 2.5 standard to FHIR Genomics in the context of precision oncology, for submission to the Journal of the American Medical Informatics Association.
"It seems to me that the FHIR specification is a little better fit for cancer genomics," thanks to additional structuring, according to Frankenburger. "That may be somewhere that we move forward to in the future."
Levy said that genomic variants are "very well structured" in HL7 2.5. "But these interpretive reports contain a lot more than just which alterations were found." She has some issues with the ability of Epic's implementation of HL7 2.5 to express information about relevant and contraindicated therapies are not.
"We do get all the data. The data is there. It's just not as well organized as it could be for secondary use, so we're hoping that future versions will enhance our ability to use that information," Levy explained.
Frankenburger noted that Rush is able to query everything in genomic reports imported into Epic, but he and Levy would like to see more sophistication in that area.
This year, the Rush precision oncology program will be assessing the value of the clinical decision support and looking to bolster the portfolio of clinical trials to "increase the actionability of the sequencing for our patients," she said.
She expects to be able to offer more types of molecular tests, including circulating-tumor DNA assays. And Levy would like to increase the panel of patients in the precision oncology program, which stood at about 1,700 as of mid-February.
"As our numbers continue to grow, we're looking forward to being able to increase the secondary use for research," Levy said.
And then there is the issue of interoperability across healthcare provider organizations.
Last month, Epic CEO Judith Faulkner controversially emailed hospital executives, asking them to oppose a proposed US Department of Health and Human Services rule intended to spur interoperability and punish providers and IT vendors for "information blocking." Faulkner, who also sent a copy to HHS Secretary Alex Azar, encouraged healthcare leaders to sign the letter. Several dozen have already signed on.
News reports have suggested that HHS could finalize the rule any day now. The agency occasionally issues major health IT regulations around the time of the annual Healthcare Information and Management Systems Society (HIMSS) conference, the largest industry gathering of the year. The 2020 meeting starts March 9.
Epic has long had a reputation for supporting interoperability on its own terms, even though it does follow industry standards for data exchange. The vendor has been pushing Care Everywhere, its interoperability platform that is said to work with any brand of EMR that follows the same standards.
DeVault did not comment on Faulkner's letter, but said that the company always focuses on making data interoperable when developing new features, including in genomics.
"The foundation of our whole genomics module is how do we get a good representation of genomic data from the laboratory setting into the EHR, and from there, how do we share it with other people on the patient's care team?" he said.
Epic supports many standards to characterize variants, including Logical Observation Identifiers Names and Codes (LOINC) for laboratory data, the SNOMED-CT ontology, and reference knowledgebases like ClinVar and the Single Nucleotide Polymorphism Database (dbSNP).
Penn Medicine has the same instance of Epic at several hospitals and community clinics. One affiliate with its own build of Epic, Penn Medicine Lancaster General Health in Lancaster, Pennsylvania, has expressed interest in adopting the genomics module, but Nathanson said that having different versions hampers interoperability.
Penn also must find a way to be interoperable with community oncology practices that use a different brand of EMR, she added.
At Rush, most of the external electronic data sharing happens with other Epic users, in part because the vendor dominates the Chicago market. Epic counts Advocate Aurora Health, Northwestern Memorial HealthCare, University of Chicago Medical Center, Loyola University Health System, and NorthShore University HealthSystem as its customers, comprising every major provider organization in the area.
Rush is still working through how to incorporate community cancer centers, however. "It makes it a lot harder when you're going from an institution that doesn't have Epic. There are still no solutions for health information exchange that work," Levy said.
DeVault said that Epic has tried to make its data structures generic enough to accommodate outside knowledgebases while also keeping multidimensional characteristics of genetic variants intact.
The vendor executive noted that interoperability summary reports for the Care Everywhere platform support genomic indicators, following a concept that Geisinger and Epic have dubbed the "genetic phenotype."
According to the Geisinger case study in Frontiers in Genetics, the genetic phenotype links "variant and gene knowledge to a defined patient characteristic or disease whose risk is associated with genetic variant(s) for which information can be delivered to clinicians."
For example, the article discusses metabolic phenotypes derived from genotypes. This, DeVault said, means that the Epic engine might characterize a patient as a CYP2C19 poor metabolizer to make information more readily actionable rather than simply stating which SNPs a patient has within a gene.
"The industry as a whole is not quite there yet in terms of standards for things like genetic phenotypes," DeVault noted. "In some areas like pharmacogenomics, we are getting there."