CHICAGO – The second round of pilot tests in the Sync for Genes program showed that the exchange of discrete, machine-computable clinical genomic data is possible. It also served to demonstrate how much more work remains in integrating genomics into electronic clinical workflows in pursuit of true precision medicine.
The federally funded Sync for Genes program seeks to standardize methods for communicating genomic information from next-generation sequencing and other tests to clinicians and patients in a usable and consistent manner.
Sync for Genes is an extension of the Sync for Science data standardization effort, which itself is part of the National Institutes of Health's All of Us research program.
Phase 1 of Sync for Genes demonstrated the usefulness of the FHIR Genomics standard for interchanging genomic data. It did not include any integration with electronic health records or patient portals. Phase 2, which kicked off in 2018 and ran for about a year, represented the first example of Sync for Genes in action in live healthcare settings.
Specifically, Phase 2 tested interchange of health data following Health Level Seven International's (HL7) Fast Healthcare Interoperability Resources (FHIR) Clinical Genomics standard, a component of FHIR version 4.0.1. The testing took place before HL7 published that version Oct. 30, and feedback from the pilot participants informed development of the standard.
The US Department of Health and Human Services' Office of the National Coordinator for Health Information Technology (ONC) this spring released its final report on the second phase of pilots.
"This has helped to advance our understanding of how to share genomic information at the point of care for research in a standardized way and it has shown to be a good model to see the reality of how these standards can get adopted and used," said ONC Chief Scientist Teresa Zayas Cabán.
Zayas Cabán, ONC Project Director Stephanie Garcia, and Sync For Genes Project Technical Director Robert Freimuth, from Mayo Clinic's Center for Individualized Medicine, also discussed Phase 2 in a paper published in April in Applied Clinical Informatics.
The paper explored confusion about how to bundle multiple FHIR resources into a single XML message, as well as four areas of concern that merit further study: FHIR proficiency, support from technology developers, the scope of genomic information integration projects, and bridging expertise in health IT and clinical genomics.
It also contains some sample code for other organizations to see how the pilot sites adapted their information systems for this kind of data exchange.
Pilot sites included the Utah Newborn Screening Program, New York-based Weill Cornell Medicine, Lehigh Valley Health Network in Allentown, Pennsylvania, and the National Marrow Donor Program.
Each site had a different use case, including newborn screening, cancer treatment, pharmacogenomics, and bone marrow matching, and involved a variety of stakeholders, including a health information exchange.
"We were gathering feedback throughout the process and informing HL7 throughout Phase 2," said Garcia.
"We were able to glean information about the ability to [move data] effectively and provide feedback to HL7 about what worked and what could use some refinement, as well as have some sort of broader lessons learned for this community to take on," added Zayas Cabán.
For Phase 2, ONC and its Sync for Genes partners created what is called Scenario 10 for FHIR Clinical Genomics, which focuses on diagnostic reports. This scenario was intended to pinpoint the types of modifications each site needed to implement FHIR Clinical Genomics resources. Under Scenario 10, participants sent XML FHIR messages to other test servers, verified the integrity of the file upon receipt, then attempted to consolidate all three MDx reports into a single report.
According to the Sync for Genes report, the FHIR Clinical Genomics specification was "ambiguous and lacks clear guidance when there are multiple ways to represent a concept in FHIR." This hinders interoperability by creating "inconsistencies" between implementations of the standard, said the authors, which included ONC staff and pilot participants.
The pilot participants identified several gaps in the standard, including the bundling issue.
They suggested that standards developers may not be completely aware of the "complexities" associated with clinical genomics. "Currently, FHIR does not provide best practices for the representation of complex workflows, such as one diagnostic report referencing another diagnostic report," the report said. "It is currently unclear how FHIR would be able to support a reference within a new diagnostic report to previous reports."
The authors also said that FHIR documentation did not adequately explain how the standard might indicate whether included genomic results are merely a part of a larger test such as a whole-genome or whole-exome sequence. Furthermore, "observations" in a FHIR resource to represent multigene panels may not show which genes are being referenced, they said.
"This issue is of particular interest to the genomic community as diagnostic technologies, interpretation, and genomic science are still evolving, and the roles different genes have in disease presentation can change over time," according to the report.
The pilots also found that FHIR Genomics was unclear about how to capture metadata about genomic information, including which regions of the genome were tested. In the current Release 4 of FHIR, "location" means a physical space such as a laboratory or facility, not the location of genetic features on a reference sequence, they said.
The participants also observed missing or misaligned semantics, lack of stakeholder diversity in the standards development community, and inadequate understanding of the complexity of genomics among technology developers and implementers.
Garcia said there is a working list of items provided to HL7 for working groups to address as they go forward. "One that did get approval [from HL7] shortly after the conclusion of the project was the ability to bundle different FHIR resources," Garcia said.
The National Marrow Donor Program (NMDP), which took part in Phase 1 in the area of tissue matching, worked with the Stanford Blood Center in Palo Alto, California to see if they could share human leukocyte antigen (HLA) and killer-cell immunoglobin receptor NGS data, and then integrate this information into a patient portal.
After the program collects patient samples by buccal swab, it orders NGS-based HLA genotyping testing. At minimum, the lab sequences exons 2 and 3 of HLA genes, and may run a whole-genome sequence. Test results are uploaded to a portal in Histoimmunogenetic Markup Language (HML) format, and are converted to FHIR as part of a "transaction bundle" that includes the final report, data supporting the genotyping decision, allele identification, and exon sequencing data, according to the Sync for Genes report.
This information goes into a repository in FHIR format for later retrieval when matching potential donors to patients in need of marrow, as well to support research.
Prior to Phase 2, the NMDP developed an HML-to-FHIR conversion tool, and verified during the pilot phase that the tool worked for FHIR release 3 and 4, following the HL7 implementation guide for clinical genomics.
However, project leaders noted "insufficient provenance resources" to track the original source of data and MDx reports for clinical genomics usage, the report said. NMDP is building some of those to support its own conversion tool so it will be able to send real patient genomic data through its HML gateway in the future.
The Utah Newborn Screening Program, run by the Utah Department of Health, is building a whole-exome sequencing platform for second- and third-tier molecular testing in babies who had abnormal biochemical test results. During the pilot phase, this program developed a proof-of-concept model for standardizing and sharing raw genomic data with care providers.
By following the FHIR Clinical Genomics standard, the screening program tested its ability to send MDx reports as FHIR messages to appropriate institutions and clinicians using an application programming interface. ONC said that the test was successful in sending FHIR messages and genomic test information as discrete, machine-readable data.
To get that outcome, the Utah Newborn Screening Program had to work through a variety of technical challenges, including seeking guidance on which FHIR resources were most appropriate for attaching VCF and other files at an adequate level of granularity to support reanalysis if necessary. They also found that some FHIR resources were not designed with genomics in mind, so the program needed guidance on how to represent genomic concepts in FHIR messages and harmonize certain definitions of data elements in clinical genomics.
"Perhaps the most significant and pressing challenge for the NBS team were issues regarding health IT developers' readiness and ability to adopt FHIR," the report said. For example, the program initially lacked adequate IT infrastructure to store large genomic files.
Weill Cornell Medicine, a participant in the US National Institutes of Health's All of Us cohort research program, has been applying precision medicine to cancer care. Its Sync for Genes use case was pharmacogenomics decision support for metastatic cancer, as well as trial recruitment.
The academic medical center was looking to feed clinical decision support systems with discrete genomic data from next-generation sequencing tests to support care in its Englander Institute for Precision Medicine, according to Medical Director for Informatics Sameer Malhotra, who was the pilot project's lead.
Weill Cornell integrated tests into its Epic Systems with Exome Cancer Test version 1.0 (EXaCT-1) — the first whole-exome sequencing test for clinical oncology approved in New York state — and its successor, EXaCT-2. For the Sync for Genes pilot, the Weill Cornell-affiliated NewYork-Presbyterian Hospital used EXaCT-1.
In the narrative of the pilot, a clinician requests through the EHR an EXaCT-1 test for a patient diagnosed with metastatic bladder cancer to look for variants that might be responsive to known chemotherapies or immunotherapies. The EHR sends the order to a molecular pathology lab, which processes the test and reports results back into the same system as variants that are pathogenic, likely pathogenic, and of unknown significance; variants of unknown significance go into an external repository to save EHR space.
The provider then can run queries of knowledgebases against the variants identified in search of information relevant to that specific patient's treatment.
Data flows between systems — the EHR, laboratory information systems, genomic knowledgebases, and payer databases — from a connection built with a FHIR API. Weill Cornell also is building an app to bring in information from public resources such as ClinVar and map that data to FHIR specifications.
The 21st Century Cures Act and the anti-information-blocking rule it spawned encouraged the use of APIs and standards such as FHIR to encourage the interoperability of health information to support coordinated care and, indirectly, the practice of precision medicine.
In the pilot, the New York City health system looked at only a small subset of the FHIR Clinical Genomics standard. Early on in Phase 2, Weill Cornell Medicine decided to focus on one test for cancer genomics, looking at how to represent somatic variants in the EHR. "It was more about seeing how we can further the FHIR standard and make use of that for the use cases," Malhotra said.
"Broadly speaking, we really were interested in this is [because] the genomic data that's coming out of these labs is not static. The knowledge evolves with time, and consequently, the clinicians and patients may need to be notified," he explained.
Variants thought to be pathogenic have turned out to be normal as new research appears, so the science has to evolve.
"We couldn't just have data coming from the lab and sitting in the EHR in a static fashion," Malhotra said, so Weill Cornell wanted to use FHIR to connect current knowledge with current genotype-phenotype data.
Also motivating Malhotra was the rapid increase in the volume of sequencing tests done at the health system in the last five to 10 years.
"As we have been doing more and more sequencing, we realized that the results are not just being seen by the oncologists, but are often seen by other providers," he said. "We have way too much knowledge out there and we need to bridge the gaps for clinicians who are not essentially specialists in this."
He said that FHIR was a good standard for matching molecular reports to knowledge sources to help explain variants to those who are not experts in genomic medicine.
Malhotra did find some shortcomings in FHIR Genomics.
Release 3 was a little immature, he said, but being part of Sync for Genes gave Malhotra the opportunity to get involved in several HL7 working groups, which led to the development of new resources.
"I think R4 addresses a lot of the initial shortcomings that were realized around how exactly to share sequence-level data," Malhotra said.
At Allentown, Pennsylvania-based Lehigh Valley Health Network, another Epic user, the organization's cancer center integrated genomic sequencing into EHR and clinical workflows for pharmacogenomics purposes.
Lehigh Valley leaned on HL7 to move genomic data into the EHR alongside phenotype information, and redesigned its oncology workflow to deliver clinical alerts through Epic at the point of care, based on drug-gene pairs.
"What we wanted to do was use some form of genomic data, move it into the EHR directly in the workflow of providers, and expose it to decision support so that the providers would be able to make decisions in real time," said former Lehigh Valley CMIO Donald Levick, still a practicing pediatrician there.
Lehigh Valley relied on sources including FDA labels and Clinical Pharmacogenetics Implementation Consortium (CPIC) guidelines for pharmacogenomics.
The health system did not use FHIR, but instead sent messages in the HL7 v2.2 format, based on the capabilities of its EHR in 2018, and programmed three oncology drug-gene pairs and related lab tests into Epic. "As we got into the pilot and investigated the facts around using FHIR in Epic and for what we wanted to do with it, we found that our existing HL7 interface was appropriate and actually smoother than trying to put FHIR in place," Levick said.
LVHN created three forms of an alert for HLA-B*5701, which predicts hypersensitivity to HIV antiviral abacavir. When a physician orders the drug, the system responds based on the patient's HLA-B*5701 status, according to Shannon Fenstermacher, the organization's lead pharmacy analyst.
The system automatically fires a warning to the prescriber not to continue the medication order if the patient is positive. If the patient's genetic test is pending, an alert says to await the result before proceeding with the prescription. In a case where the patient has not been screened for HLA-B status, the EHR advises the physician to order the test and includes a link to place the order.
For the month of December, there were 318 web service calls on the network when physicians ordered the drug, and the system sent 20 alerts indicating that the patient in question had the variant, Levick said.
"We knew it wasn't going to be a high-volume alert, nor did we want it to," Levick said. "We wanted to make sure it's working."
Lehigh Valley also created targeted medication warnings for hyperkalemia, but has not added other pharmacogenomic alerts, largely because further genomics work has been put on hold during with the COVID-19 pandemic.
Barring another surge in COVID-19 cases in the Lehigh Valley, the health system plans on upgrading to the November 2019 release of Epic, one that offers native support for genomics, though not for FHIR Genomics, Fenstermacher said.
"As we bring the Epic genomics module into play, I think we're going to have to reevaluate how that is going to receive discrete genomic data from the sources that we have through our lab and our reference lab," Levick said. "At that point, we'll decide what the best way to move that data is, whether or not we would require FHIR."
Prior to the pandemic, LVHN had been working with its outside genetic testing partners to prepare data for migration into Epic, Fenstermacher said. Some were only able to provide PDF files, not discrete data. The point of Sync for Genes Phase 2 was to exchange discrete data, and the LVHN-owned Health Network Laboratories is able to send discrete data, she said.
Weill Cornell's Malhotra wishes that Phase 2 would have had more collaboration with labs, with senders and receivers alike adhering to the FHIR clinical genomics standard. "That would have probably made the turnaround to get it into production a little faster, versus us working unilaterally on the receiving end," he said.
Indeed, The Sync for Genes Phase 2 report noted that outcomes depended on whether lab and other ancillary information systems could implement FHIR. For this reason, an upcoming Phase 3 will focus on laboratory connectivity.
Zayas Cabán said that ONC has identified potential sites for Phase 3, but has not announced any details. "Phase 3 is exploring the potential for standardizing genomic data being generated by the laboratories, so we're trying to look at the entire process," she said. "It will continue to expand on the adoption of the [FHIR] specification."
Meanwhile, ONC and the Centers for Medicare and Medicaid Services are leading the implementation of the "information blocking" rule in the 21st Century Cures Act, which Sync for Genes fits into.
"What the final Cures Act rule does is actually create the ecosystem and the technology infrastructure that will enable some of the sharing to happen in a standardized way," Zayas Cabán explained. This will require the use of FHIR R4, which is the first release that includes the specifications for clinical genomics.
"Once the [information blocking] rule is implemented, it will even out the playing field in terms of the infrastructure that is needed in order to make this happen in a standardized way," Zayas Cabán said. She also acknowledged that like any aspect of health IT interoperability, clinical genomics integration is a long-term proposition.
"We will continue to help advance the work that is needed and enable some of the testing that is needed to be able to provide feedback to different communities in terms of what's really working, what's not working yet, and what can be done about it," she said.