By Clinical Informatics News Staff

April 6, 2016 | The fourth annual Medical Informatics World Conference met this Monday and Tuesday in Boston, alongside the ongoing Bio-IT World Conference & Expo. Speakers at the conference addressed a health IT landscape that is quickly progressing from highly specific tools built around convenience—like patient portals, teleconferencing, and electronic prescribing—to systems for aggregating and mining health data to better understand whole cohorts of patients. These analyses can produce surprising insights that hint at new ways to manage large health systems. Yet, as more than one speaker acknowledged, they still rest on an IT architecture that was not designed for the tasks now demanded of it.

The first keynote speaker on Monday morning was especially adamant on this point. Nicholas Marko, Chief Data Officer of the Pennsylvania provider network Geisinger Health System, laid out the case against doing R&D in the electronic medical record (EMR). This has often been the default approach to analyzing big health data, since the EMR contains detailed health and treatment histories on large numbers of patients. But, Marko said, core needs of the EMR are in conflict with the agile analyses that drive good research. An R&D infrastructure, he said, should be able to deploy small pilot projects, iterate quickly, and scale up slowly as ideas are worked out. EMRs, however, need stable structures and rigid governance, making them better suited to occasional, system-wide updates than small iterative changes.

“It’s fundamentally a constraining environment,” Marko said. At Geisinger, he added, within four months his team was able to transfer more EMR data to a custom Hadoop deployment, structured in formats they preferred for research, than they had ever previously been able to access working with the EMR itself.

The second keynote was delivered by Adrienne Boissy, Chief Experience Officer at the Cleveland Clinic. Roles like Boissy’s, concerned with patient satisfaction with the healthcare experience, are relatively new, but rapidly becoming important with the shift from volume- to value-based care. And as Boissy illustrated, IT projects that serve patients’ priorities are often equally valuable to providers. Patients, for example, would rather upload medical images from home than bring in a thumb drive, which also frees physicians from having to spend part of a visit wrangling new data. Providers who don’t stay ahead of the curve on patient experience also face risks. If a hospital doesn’t make patient ratings of their doctors transparent, for instance, people will just look up doctors on third-party sites with no quality controls.

Jason Burke, Chief Analytics Officer at UNC Health Care, returned to Medical Informatics World to build on his 2015 keynote on creating a learning health system. This year, he borrowed concepts from the study of learning in psychology and social science to illustrate the development of health IT systems. Medicine, he said, is largely stuck in the first tier of “cognitive learning,” when concepts are digested in isolation—like when a hospital creates a new metric to track performance. Health IT teams need to advance to “associative” and “autonomous” learning, where meaningful connections are made across concepts and, eventually, become second nature to the system.

“Learning is highly contextual,” Burke said. A heart rate measurement, for instance, is not useful information without knowing when and why it was taken. “We need to do a much better job of contextualizing data and the workflow associated with it. And I would argue that is actually one of the first steps we need to do, because if we get that wrong, everything that happens after that is much more difficult.”

The final keynote speaker, Massachusetts eHealth Collaborative CEO Micky Tripathi, gave a lightning overview of the prospects for EMR interoperability. This issue, of EMRs unable to transfer patient data to other hospitals running other systems, has been a hot one in health IT circles. In sharp contrast to Marko, who joked that achieving meaningful interoperability was as likely as winning the Powerball, Tripathi was optimistic that incentives are finally lining up for fluid data sharing across provider boundaries. He noted that the challenge facing EMR users is not unique in history: in the 1890s, he said, a person who wanted to make telephone calls to ten different locations would need ten different phones. Networks like AT&T, although they were commercial and did not serve the whole public, made the problem orders of magnitude easier to solve by joining phone users across the country in large blocs, leaving fewer gaps. Similarly, a few large networks like Epic and Commonwell are starting to consolidate the EMR market today.

Another reason for hope, Tripathi said, is the new ubiquity of EMRs. While electronic health records have been around for a long time, they’ve only been taken for granted for two or three years—so demand for interoperability is just coming to a boil today. Tripathi also cited government action as a huge factor in driving industry efforts. Proposed regulatory actions, like having the Office of the National Coordinator mandate standards of interoperability, are, he said, “completely bananas. We should all oppose all of it. But the threat of that looming stupidity is driving people to say, ‘We’ve got to solve this problem in the private sector, or they’re going to do it to us.’”

Seeing the Surprising Patterns in Health Data

Throughout the Medical Informatics World Conference, speakers had advice for designing intelligent, interoperable health IT systems. Stanley Huff, Chief Medical Informatics Officer at Intermountain, stressed the need for a new approach to data sharing where data is rarely copied from one network to another. This will involve “a paradigm shift from what I call a classic, ‘messaging’ strategy, to a ‘service-oriented’ strategy,” he said. With projects like FHIR, which is creating a common development language for data analysis tools in the EMR, data sharing can become a matter of running tools wherever the information resides.

This will not only be simpler and less computationally demanding, Huff added, but will also help preserve data quality. “I’ve never been smart enough to know how to keep two copies of data absolutely consistent.”

Valmeek Kudesia, Director of Clinical Informatics at the Boston VA, shared a remarkable system the Department of Veterans Affairs is starting to implement across its health centers. In an effort to compare outcomes across different treatment strategies, the VA is training its EMR system, VISTA, to recognize cases where there are multiple treatment courses available and no clear evidence to prefer one over another. If doctors in these situations consent, the system will randomize treatment—say, offering one patient chemo before radiation and another the opposite. In this way, the huge VA network can gradually accumulate evidence about the standard of care in a huge variety of disease areas.

Elsewhere at the conference, a common refrain was that getting new insights from big health data requires creative thinking.

For example, Craig Schilling, Vice President for Patient Programs at Optum, addressed best practices for improving medication adherence. (Optum makes predictive software in this area.) Patients are typically considered adherent if they take their medications on 80% of days, and some incentive programs reward hospitals for keeping patients above that threshold—so when those hospitals use messaging or reward programs to help patients stay adherent, they often target patients just above or below that 80% rate. Yet Schilling noted that, if you track the same patients’ adherence over multiple years, there’s surprisingly little stability: as many as 30% of “non-adherent” patients become adherent the next year, and vice versa. Because being close to the line is a poor predictor of future behavior, hospitals need to find new metrics to decide which patients are actually likely to fall from adherence, or can be persuaded to become adherent.

Dov Marocco, who works for the Santa Clara Valley Health and Hospital System, a safety net system in the San Francisco area, detailed his team’s efforts to find especially vulnerable patients in an already very vulnerable population. This work, which ropes in patients who get care sporadically and in disparate locations, involves collating data from many poorly connected systems: medical and claims data from clinics, but also information from jails, public health departments, and substance abuse and homelessness databases. But it also requires the thoughtfulness to prioritize among that data. Tracking the number of days patients are hospitalized is very useful, said Marocco, but only if you can create a system that understands that a single 30-day visit is not the same kind of red flag as 30 one-day visits.

Collecting little-considered types of information can also reveal patterns that are meaningful for care. Emil Chiauzzi, Research Director of PatientsLikeMe, and Mikele Bunce, Quality of Care Lead at Genentech, shared results of a joint study surveying almost 4000 chronic disease patients on their “empowerment” and engagement in their own care. By asking about these patients’ relationships with their caregivers, and knowledge of their own diseases and treatments, this project uncovered correlations that few clinicians would have intuited. For example, male patients felt slightly more empowered than female patients across the board. And while empowerment differed between diseases, the division didn’t neatly track factors like a disease’s severity, or how common it is. Instead, disease areas in which patients felt a high degree of empowerment—like multiple sclerosis or Parkinson’s—tended to have better-defined outlooks and mechanisms than low-empowerment diseases like major depression or chronic fatigue.

Another unexpected finding from this study seems directly relevant to the way we approach chronic disease care. “One of the survey questions scored low across all disease areas,” said Bunce. “And that was patients feeling that they had enough support from friends to help manage their disease. Should we be looking at tools to enable friends and family members to better support loved ones?”

Answering questions like this will require more rigorous clinical evidence than can be derived from a survey. But as the speakers at this year’s event showed, the necessary groundwork of powerful health research environments is now being laid in care centers across the country.

The Medical Informatics World Conference will meet again in Boston next year, from May 22^nd to 24^th.