I’ve just come back from the "Discovery and Innovation in Health IT" workshop. You can find more on the workshop’s web site. You can also follow the twitter stream by using the #dihit hash tag. This workshop pulled together about 100 people – mostly researchers from academia, government and corporate labs – to lay out the key health IT research challenges going forward. I’ve attempted to summarize some of the key themes and points that were raised, but I’ve by no means captured it comprehensively and I’m certain that I’ve taken a lot of license in my interpretation – so please add your voices – I’d love to get other perspectives.
The workshop started with some tone-setting presentations that illustrated the gap between the potential of health IT and where we are now. Zak Kohane, in a brief introduction, noted that back when he was a grad student (more than a few years ago), clinical decision support was said to be right around the corner and while it’s still not reached its promise, Apple’s managed to generate 86,000 apps in its app store, many of which focus on health, in less than a year. “Why?” he asked. Bill Stead, recapping the work he led for the National Research Council’s report on Computational Technology for Effective Health Care, laid out the challenge starkly, saying (and I’m paraphrasing) that the path we’re on with health IT will not solve the problems in our system and could even make them worse. Most hospitals have implemented IT in ways that rate a 5 on a scale of 100, where 100 is ideal and the best are at 15-20, he went on. Today’s health IT systems “chain us” to the realm of transactions not decisions and often focus on post-hoc documentation. “Would you ever get into an airplane where the pilots go through the checklist after they’ve taken off?” he quipped. He focused a lot on the need for cognitive support, showing a hockey stick graph of the number of facts that will be relevant to a given clinical decision over time (this theme reappeared several times over the two days). The number is expected to reach 1000 by 2020, while the number of facts that a human can contemplate while making a decision remains stuck at um, five.
Other presentations, notably from Dietrich Stephan of Navigenics and Craig Feied from Microsoft, extrapolated from recent progress in genomics, proteomics and systems biology to sketch out a future of increasing precision in our understanding of diseases (hence the hockey stick of medical knowledge). In these visions, the increasingly ability to subclassify diseases at the molecular level, to understand the biological processes that cause them and to design laser-like therapies to target them, leads to precision diagnoses, early detection and better understanding of which therapies will work and which won’t.
These presentations provided the background for breakout discussions that consistently returned to the same themes. There was a striking concordance in the visions that informed those discussions. The broad vision is characterized by:
- A broadened scope of health data. While health IT has traditionally focused on the data generated by care processes in health care settings, the participants imagined a future where health IT used the data that emanate from where health happens –where we live, learn, work and play. Collected from the flow of life, data on diet, exercise, sleep, pain, mood, meds taken,will enable richer understanding of people’s health and more personalized care. (For followers of Project HealthDesign, this should sound familiar.) Mash this up with environmental exposures, ranging from toxins to corner stores and we could learn even more. A key point was to gather all these data and store them in raw form, as interpretations is them might change over time, as new knowledge is developed.
- Data liquidity. Data flow from location to location, from storage to application. This is not so much about pure interoperability as an ability to provide access to data (and some descriptors that put it in context) so others can use it and add value to it.
- An explosion of knowledge. As the talks by Stead, Stephan and Feied highlighted, the sheer amount of medical knowledge that can and often must be brought to bear on clinical decisions will expand tremendously – beyond the limits of human cognition.
- Multiple uses, multiple users. As the data both grow and become more liquid, they will serve new applications and multiple users – i.e. clinicians, patients, caregivers, researchers, public health officials. The range of apps would be broad and unpredictable.
- Data-driven personalized care. Scientific advances promise more tailored therapies and greater access to data from people’s day-to-day lives will enable finer-grained assessments for diagnoses and rapid, more precise feedback on the results of therapies. These data and this knowledge will also lead to personalized tools that can predict health (or disease) trajectories and simulate different scenarios.
- Rapid learning. As Bill Rouse noted in his presentation, randomized controlled trials are the most expensive way to learn. The vast amount of liquid data opens the door to a potential explosion of highly efficient research that’s grounded in day-to-day experience.
This vision has a number of implications:
- Separate the data from the applications. Achieving much of the above vision requires robust approaches for storing, securing and providing access to the data and then enabling the development of applications that are independent of the data storage system. Opening up the application layer is necessary to promote innovation in how the data are used (and combined with increasing knowledge) to improve people’s health.
- Finding signals in the data. The vision is highly dependent on having very sophisticated and powerful data mining techniques to extract signals, patterns and correlations from vast amounts of data – data that are often and incomplete. Imaging is another challenge and opportunity – and innovation will be needed in extracting information from images – whether the images are MRI scans or cell phone snapshots of people’s meals.
- Consolidating the information. Latanya Sweeney spoke of the “challenge of consolidation” for providing decision support. With so much data (and information) available, the challenge will be how to prioritize, consolidate and present the right information, with the right context, to the right person at the right time.
In some ways, what was striking about this vision was what it was not. This was not a vision of interoperable EHRs, all using controlled vocabularies and connected through health information exchanges. It was instead much more reminiscent of the search engine approach to the Web (perhaps this came from having so many computer scientists in the room). This was a vision not of a carefully designed system – but rather came from a recognition that health data will be unevenly collected, variably coded, and semantically unstable. But as long as the data are liquid and the tools for analyzing them are powerful, different users will be able to get what they need. This came together for me when Bill Stead commented, as an aside, that the Internet works because it is not a system.
I’ve tried to summarize what I saw as a high degree of concordance around a vision. There was, however, less discussion on the implications of this vision for health and health care in the U.S. More on that in my next post. In the meantime, I’d love to hear some reactions to the vision.
]]>