Weill Cornell uses HL7 to help integrate structured genomic data into Epic EHR
One of the frustrating challenges in putting precision medicine to work more widely for clinical care is integrating complex and voluminous genomic data into the EHR.
Instead, in most cases, "we use a very highly interoperable standard for such material called 'PDF,'" joked Beth Israel Deaconess Medical Center CIO John Halamka, MD, when he spoke at the Healthcare IT News Precision Medicine Summit in Boston earlier this month.
That's not necessarily a drawback.
"PDF is not so bad for visualization of data," said Olivier Elemento, associate director of the Institute for Computational Biomedicine at Weill Cornell Medicine in New York. "Clinicians can see the data in a way that's nice-looking and easy to understand,"
But at Weill Cornell, clinicians are able to access genomic information as structured data in the EHR. Staff at the academic medical center have found success integrating actionable test data into the Epic system to enable clinical decision support.
In addition, the hospital has developed "entire app ecosystem" to help manage the workflow of genomic testing data, said Elemento, positioning it well to capitalize on the vast promise of precision medicine.
"I hear a lot about the challenge people encounter, and the process of trying to integrate data into the EHRs and using them in clinical workflow," said Elemento. "At least here at Cornell, it seems like we've taken on the challenges and made some pretty good progress."
That's not to say that the successes have been effortless. "We've experienced some things that don't work – but also thankfully have experienced some things that work."
At Cornell's Caryl and Israel Englander Institute for Precision Medicine, Elemento has spent the past four years focused on integrating complex data into Epic in a structured form that makes genomics more clinically relevant, he said.
One of its programs focuses on advanced metastatic cancers, looking for actionable genetic mutations that could respond well to immunotherapy or innovative drugs.
Using a whole-exome sequencing test, clinicians "evaluate the mutational profile of these patients," said Elemento.
"The test is pretty extensive," he added. "It covers 21,000 genes and is used routinely now for advanced cancer patients now at Cornell and NewYork-Presbyterian."
That "extensive mutational modeling of tumors" leads to a report that is eventually attached to the EHR as a PDF – a data-rich and easy-to-read document that can be consulted by physicians.
But Weill Cornell has also found a method to push that same information into Epic as structured data, said Elemento. "We figured out a way to craft HL7 messages so these messages are accepted in Epic and displayed as tables: you can see mutations, interpretations of mutations."
One key, he said, is to be selective in the specific data that's displayed.
The PDF report contains every mutation discovered by the clinicians, but they're ranked, so to speak. The most important mutations, the actionable ones, are at the top of the report, he said. "In the middle of the report you have other mutations driving the cancer, and then we also put the passenger mutations.
While the report itself displays all those mutations, "it turns out that displaying it all as structured data in Epic is not possible," said Elemento. "We tried it and it didn't work. Epic is not designed to handle the large number of mutations which you could find in a typical patient, many of which actually are not actionable.
"So what we did was push all of these mutations into Epic as structured data, but we only display the actionable data," he said. "All the data is in there. And we can do the data warehousing and analytics. We can do queries based on genomic data for every mutation that's in the EHR – in addition to combining other data for the patient – clinical data, demographics, conditions, diagnosis, meds, prescriptions and so on.
"We push everything into Epic as structured data, but we only display what's most important and relevant for patient care," he added. "But everything is still available as queries. That helps when it comes to things like populating clinical trials."
A Cambridge, Massachusetts-based company called Standard Molecular, which specializes in HL7 and XML integration of discrete data, helped out on the initiative. Its Chief Medical Officer, David Artz, MD, is an associate professor at Weill Cornell and former CMIO at Memorial Sloan Kettering Cancer Center. Before that, he worked at Chicago's Northwestern Memorial Hospital, where he implemented both Cerner and Epic systems.
"David Artz has a lot of experience with Epic, so they've essentially been able to help us connect our workflows with Epic," said Elemento. "Part of it is basically translation – putting it into a format Epic can accept. Essentially all of it is just crafting HL7 messages: The data pushed into Epic as HL7 messages and then can be consumed there."
"Integrating rich discretely detailed genomics data into medical records was a longstanding technical challenge, which we have now solved,” said Artz in 2016 when the Englander Institute for Precision Medicine first announced its success adding computer-readable genomic data to its EHR. The next step, he said, was to share the advancement with other hospitals.
"This project could advance the field of precision medicine beyond Weill Cornell," said Artz. "We’ve created a model for how it could be used elsewhere, especially at other institutions that use Epic."
Beyond its EHR success story, Weill Cornell has also built out its IT infrastructure to make more effective and efficient use of genomic data – an "app ecosystem allows us not only to track samples across the entire process but also essentially be able to do things like ordering tests, ordering for Epic, tracking samples across the workflow," said Elemento.
"Clinical genomics is a complex activity: It involves multiple groups, multiple people, and if you look at the lifecycle of a sample as it goes from biopsy to sequencing to reporting to going into the EHR, there are many many steps. To really facilitate this, IT plays a major role.
In sequencing thousands and thousands of samples, "we need to be able to have essentially real-time tracking," he said. "You want to know where the sample is at in the process. All this stuff is being tracked through apps that allow us to understand all aspects of the process."