Achieving effective integration of genomic data into knowledge-generating health care systems will require interoperability, said Steven Leffler, the chief medical officer at The University of Vermont Medical Center. Adapting current platforms, reusing existing components of systems that work well, and standardizing the structure of the data, including consumer data, will contribute to these efforts. (See Box 6-1 for actions suggested by individual workshop participants for facilitating the integration of genomics into health care systems.) Once the data are in the appropriate format and can be more easily transferred and used, clinical decision support algorithms can be developed to provide the necessary information at the point of care, said Sam Shekar, the chief medical officer within Northrop Grumman’s information systems sector. The implementation of genomic data in the clinic is not without its challenges. The introduction of new knowledge into the health care system will likely mean that there will be cultural changes related to how information is used, and it will call for behavioral changes in clinical practice as well, said Geoffrey Ginsburg, director of the Duke University Center for Applied Genomics and Precision Medicine. Ensuring that any changes in health care practice benefit all people and do not introduce unintended disparities in health care will be key, said Alexander Ommaya, the senior director of implementation research and policy at the Association of American Medical Colleges.
BOX 6-1
Possible Next Steps Proposed by Individual Workshop Participants
Interoperability of EHRs
- Ensure that the quality of genomic data is clinical grade and that it is in an accessible format so that it can be used for future research and to inform clinical care. (Risch)
- Support regulations that will make EHRs fully interoperable for genomic information. (Leffler)
- Establish data standards for genomics to allow for EHRs to communicate and for genomic data to flow more easily across labs and systems to providers. (Aronson, Fowler, Nolen)
- To demonstrate how the interoperability of systems can be increased, start with specific health problems whose outcomes are likely to be changed with genomic and other clinical data. (Hill)
Clinical Decision Support
- Reach agreement on allele and test code nomenclature to facilitate the development of clinical decision support tools for genomics. (Chute)
- Create warehouses of clinical decision support tools that can be shared and used widely. (Ginsburg)
- Measure outcomes to determine the validity of algorithms used to guide practice. (Moss)
- Develop a core infrastructure to handle clinical decision support and the long-term storage of complex data. (Nolen)
Data Sharing
- Build platforms with reusable components that are scalable and can be implemented anywhere. (Friedman)
- Standardize data so that they can be re-used. (Chute)
- Foster interoperable health care systems to enable genomic data sharing. (Terry)
- Inform the public about data sharing to cultivate a “data donor” culture. (Chute)
- Network data from around the world to increase the size of databases and power of research studies. (Aronson)
- Integrate patient-provided data into health care information technology systems. (Baker)
- Examine whether personally controlled health databanks can make data accessible for sharing while protecting privacy. (Friedman)
- Support research to understand and generate personalized user interfaces and preferences. (Baker)
Implementation
- Engage groups with a particular interest and who value genomics, such as people with undiagnosed or chronic diseases, to demonstrate the full potential of this information. (Terry)
- Measure and track health and health care disparities to determine the impact of genomics-based interventions. (Ommaya)
- Support social science and behavioral research to understand the priorities and values of patients and providers when genomics is introduced in the clinic. (Ginsburg)
To share information seamlessly, EHRs need to be fully interoperable for genomic information and other clinical information, Leffler said. Purchasers of EHR systems can demand that vendors provide this feature, though they have much less leverage if they already have bought a proprietary system. Regulatory bodies also can help push for interoperability, as can entities like the DIGITizE Action Collaborative.
It is important to establish a common set of standards across EHR platform versions and the providers who use them, said Tom Fowler, the director of public health at Genomics England. Currently, not even systems from the same vendor can easily communicate. Standards for data representations, problem lists, medication lists, and other features will make the EHR a more useful tool. The Department of Veterans Affairs has faced challenges in coordinating all of the various forms of data, a participant said. Genome-wide association studies are being conducted on approximately one-third of 1 million veterans; about 25,000 of the samples have undergone exome sequencing, and roughly 2,000 samples have undergone whole-genome sequencing. There is a large amount of data being generated, but the difficulty has been standardizing the information so that it can be studied optimally. Such standards would greatly help coordinating data in different forms.
Adapting platforms built in recent years is one way to take on current problems, said Charles Friedman, the Josiah Macy Jr. Professor and chair of the Department of Learning Health Sciences at the University of Michigan Medical School. For example, establishing health databanks could be a scalable approach to sharing data while protecting privacy. Scalability of platforms is needed, Friedman said, and this can be facilitated by reusing components of systems elsewhere. Just as a bankcard
can work in many different ATMs because people demanded interoperability, widespread public demand for interoperable health care systems could produce change, said Sharon Terry, the president and chief executive officer of the Genetic Alliance. Health databases, for example, could be scaled and maintained for both data sharing capabilities and maintaining privacy. Other potentially scalable programs include the SMART® platform1 (substitutable apps that can be integrated with EHR systems), a data normalization pipeline for phenotyping called PhenoTips,2 and the Clinical Decision Support Consortium.3 “These are just a few examples of platforms that we might be able to incorporate into things that we are doing, saving years, literally, of work,” Friedman said. Ginsburg added that perhaps these software platforms and clinical decision support tools could be put into a shareable warehouse for dissemination. The public also could become more involved in specific policy issues, such as policies on genomic data sharing, Terry said.
Instead of trying to fix the problems with the current EHR platforms, what may be needed is to “create a new [EHR] system that would be universal across all the different health care systems,” suggested Debra Leonard, a professor and the chair of pathology and laboratory medicine at The University of Vermont Medical Center. There would be significant resistance to discarding the current system because of the large amount of investments that organizations have made in it, said Friedman and Fred Sanfilippo, the director of the Emory–Georgia Tech Healthcare Innovation Program. An alternative would be to incorporate platforms that add value into the existing legacy systems, which is why standards are needed to ensure the interoperability of these augmented systems, said Sandy Aronson, the executive director of information technology for Partners HealthCare Personalized Medicine.
By addressing interoperability issues with two or three specific health problems in which there is an argument that genomic and other clinical data can produce game-changing outcomes in a short time, it should be possible for the push for interoperability to gain traction, said Colin Hill, the chief executive officer of GNS Healthcare. This will be particularly true where there are economic incentives. “It’s no accident that oncology is one of the first places where you’re seeing a lot of data sharing,” Hill said. Oncology and the prevention of preterm births are
_________________
1SMART, http://smartplatforms.org (accessed March 2, 2015).
2PhenoTips, https://phenotips.org (accessed March 2, 2015).
3Clinical Decision Support Consortium, http://www.cdsconsortium.org (accessed March 2, 2015).
two examples of areas where patients, providers, and payers are all interested in making progress with genomics. “Some of those areas are emerging, but we need to be thoughtful and careful about the health economics of those areas up front,” he said. “At the end of the day the economics have to make sense.” For many diseases, said Lynn Etheredge of the Rapid Learning Project, the important questions will be how much genetic information is needed, for which patients, and who is going to pay for generating that information.
Genomics is at the forefront of clinical decision support because the data are inherently computable and therefore could be provided as supporting information at the point of care, a workshop participant said. Clinical decision support provides a better opportunity for physicians and patients to use genomic data than either could have when the information is solely contained in paper format. Validating the algorithms that are being used to guide practice will be important, Ginsburg said. Aronson said that today’s in silico prediction algorithms are “extremely noisy,” so that “you wouldn’t want to base a significant clinical decision on one of those algorithms alone.” Validating these algorithms will require measuring outcomes, said Scott Moss, who leads the research informatics research-and-development team at Epic.
Market forces could lead to algorithms from different vendors being compared to see what effect an intervention had, said Andrew Kasarskis, the co-director of the Icahn Institute for Genomics and Multiscale Biology at Mount Sinai Hospital. “To the extent that an organization has a good handle on what its costs are and somewhat structured data on its clinical population, you’ll be able to do some comparisons,” he said. In this case, good data standards would further market comparisons and improve health.
Machine learning is one possible way to produce results to guide clinical decision support. But Ketan Paranjape, the worldwide director of health and life sciences in the Health Strategy and Solutions Group at Intel Corporation, said he was worried about decisions that are based on inadequate databases. Even 50,000 medical records in a system are not enough, he warned. One way to increase the size of databases is to network information from around the world, Aronson noted.
FDA may move to regulate decision support systems, said John David Larkin Nolen, the managing director of laboratory medicine at Cerner Corporation. This is another argument for interoperability, he said. “We have to start talking about core infrastructures because that will make it a more sustainable project, versus a one-off here, a one-off there.” A core infrastructure could handle both the long-term storage of complex data, which does not necessarily belong in an EHR, and decision support.
UNDERSTANDING CONSUMER VALUE AND PREFERENCES
The cultural issues associated with genomics-enabled health care systems have been studied less than the technology issues, Terry said. The consumers of health care will not engage in something unless it offers value to them. One way to start engaging the public, said Terry, would be to start with a subset of people where the value is high, such as those with undiagnosed or chronic diseases and then the reach could be broadened to a larger population. “We’re very engaged in [activities] like comparing plumbers or figuring out the best car to buy, because we have tools to do it and it’s important to us,” she said. The banking and auto industries have focused on designing their products based on what people want because there is a demand from consumers. We need to “activate the public” to become more informed about genomics and create a demand around individual needs, but in order to accomplish that, the right tools need to be available.
Dixie Baker, a senior partner at Martin, Blanck & Associates, noted earlier that individuals provide the information that drives a knowledge-generating health care system and that understanding their preferences for data sharing is key. The value of personalized user interfaces will require research to understand and generate, she said. “We’re trying to engage everyone, and to do that we have to recognize the diversity of people’s values.” As an example of this kind of work, Paranjape noted that he and his colleagues worked with medical ethnographers to figure out what representation of data on a computer screen would be most effective.
An increasingly important issue will be how to accommodate and integrate patient-provided data, whether from wearable monitors, home sensors, or personal genomic tests, into health care information technology systems, said Baker. “Consumers are definitely investing their own
dollars in health products and services,” she said. “This is a very rich area of development and research.”
Moss observed that consumer health data are already flowing into EHRs at many U.S. organizations, and the trend is accelerating, especially as personal monitors become more common. The consumers decide what data go where and who will have access to the data, with control “very much in the hands of the consumer,” he said.
Although there are some concerns about the quality of such consumer-provided data, their potential is great, Paranjape said. As an example, he mentioned a wearable device that can track the tremors of Parkinson’s patients and detect changes caused by medications. And computing will continue to become ever more powerful, which will enable data analysis to be more distributed than it is today, Fowler said. People will be analyzing their own genomic information and coming to the health care system to discuss what they have found and, in some cases, to seek treatment.
The extent to which health data will be open remains an unanswered question, Friedman said. As an example, he mentioned personally controlled health databanks, in which an individual contracts with a health databank to be the custodian of his or her health data (see Chapter 3 for a discussion on PEER). “What goes in is what that person wants to go in … and what gets released is just what that person wants released,” he said. The idea is completely scalable, he added, and the data are open if the patient agrees that the data can be used for a specific purpose. “There can be multiple banks, just as there are multiple financial banks, competing with each other, but all providing the same services using the same standards. I think that’s a very important concept that this community should be aware of.” Aronson commented that such a system also would be useful when a patient moved from one health care system to another.
People may not want all of their health information in a personal health databank. For example, they may not want the databank to contain a record of sensitive health issues from when they were young. But Friedman pointed out that the data in a health bank would not be the only instantiation of the data. A health system could also retain data in an EHR, so people could choose which data to retain in a data bank. Baker added that data coming to a bank would need to be digitally signed to ensure the integrity of the data.
Accounting for individual preferences may require applications of social science and behavioral research. Ommaya said that the Association of American Medical Colleges is involving the implementers of health care systems in the entire development process, from the formula-
tion of questions to the conduct of studies to the dissemination and implementation of results. People need to be acculturated to new systems if they are to be comfortable with those systems and use them, Ommaya said.
Friedman agreed that translating new knowledge into practice requires psychology, communication science, implementation science, and other behavioral sciences. Sanfilippo remarked on the number of M.D./Ph.D. programs that are considering Ph.D.s in the social sciences because of the importance of behavioral research and related disciplines. “That’s a trend that you’re going to see accelerate,” he predicted. Behavioral change involves incentives, disincentives, rewards, and recognition, Sanfilippo added, all of which are being studied by the field of behavioral economics.
Aligning efforts to integrate genomics into the health care system and to use the information effectively will require an understanding of the priorities and values of patients and providers, Ginsburg said. We would like to see action come from this meeting. At the end of the day, he said, we should be thinking about how we can “begin to build this system that is going to support genomics-enabled health care.”