To Take Advantage of FHIR, mCODE Effort Developing Common Data Model for Oncology

In December 2017 I had a conversation with Keith Campbell, director of informatics architecture at the U.S. Veterans Administration, about the relationship between Fast Healthcare Interoperability Resources (FHIR) and semantic interoperability. At the time, he said that focusing on FHIR is like starting to build a skyscraper on the third floor. The ground floor, he stressed, is creating semantic interoperability so that health systems don’t have to keep creating maps between different terminologies.

On Nov.  12, Brian Anderson, M.D., chief digital health physician of MITRE Corp., gave a fascinating presentation in the Harvard Clinical Informatics Lecture Series about the effort he is leading to develop a common data model in oncology called mCODE (minimal Common Oncology Data Elements). This initiative is identifying cancer data elements that would be essential for analyzing treatment across electronic health records (EHRs) and cancer practices to improve quality and care coordination. MITRE is developing an mCODE-based FHIR implementation SMART-on-FHIR application that will extract mCODE data in computable formats and deliver reports to providers and patients—empowering shared decision making.

The effort involves most if not all of the big players in oncology such as the American Society of Clinical Oncology (ASCO) as well as several large health systems, including Kaiser Permanente, Partners Healthcare and Intermountain. Anderson, who previously worked as informatics department head at athenahealth and on a task force of the Office of the National Coordinator for Health IT, said MITRE has several pilot projects in the works and it is in talks with Epic and Cerner to make capturing the agreed-upon data elements part of their next releases.

MITRE is working with oncologists to come to an agreement about the elements that need to be part of every cancer patient’s record. Once agreed upon, the goal is for every health system to find a way to capture it and have an agreed-upon way to structure data in the EHR so it becomes semantically interoperable in what MITRE refers to as an oncology standard health record (SHR). “Part of the reason it has been difficult to show success from spending on interoperability spending is that data is not harmonized as much as it needs to be to help us develop insights,” Anderson said. “If we have minimal oncology common data elements, then providers can use FHIR APIs to share data and to develop a learning health system.”

MITRE brought together a group 80 physicians, including surgical and radiological oncologists, to hammer out categories. They came up with 75 key data elements. The clinicians MITRE worked with agreed that of all the data captured about patients in EHRs — clinical notes, genomic profiles, side effects, MRI imaging, cancer staging — disease status is the most important. “At the end of the day, as a physician, most of what I care about is: is the patient getting better? Are they tolerating chemo?” Anderson said. Those elements are often in blobs of free text. It is hard to find. Natural language processing is capturing some of this information, but it is not 100 percent accurate. “What we have attempted to do is have an agreed-upon way four or five variables to describe disease status,” he explained.

One question Anderson gets is why no one has done this before. He said there was an earlier effort in the early 2000s called CA-BIG. It failed, in part, he believes, because it attempted to create a data dictionary of all clinical elements and didn’t take a use case approach. He said mCODE  is working on a use case-driven way of implementing this at the point of care. “We want to demonstrate the utility and the value that can be realized at other health systems.”

The lists of health systems involved in the effort has grown to include Penn Medicine, Geisinger, Rush University, ThedaCare, Trinity Health, Kaiser, Intermountain, UCSF, Mayo Clinic, and more. Thought leaders consulted include the FDA’s Amy Abernethy, John Halamka, Robert Miller, Stan Huff, Margaret Van Meter, Larry Shulman and Aneesh Chopra, according to Anderson’s presentation.

 Anderson says MITRE is talking to other specialty societies such as cardiology and infectious disease about taking the same approach to semantic interoperability, so they can start comparing apples to apples in their learning health system efforts.

 One mCODE pilot project involves a partnership with vendor Nuance, engaging its expertise in speech recognition to see if its Dragon Medical One and Ambient solutions can decrease the burden of collecting mCODE data elements.

 Another pilot project involves ways to increase utility of EHRs for clinical trials of pharmaceuticals. “My hope is that mCode will be the foundation to unlock innovation and create efficiencies around real world data in clinical research,” Anderson said. “It could be a game changer for unlocking real-world clinical data.”