The following is a special contribution to this blog by Suchi Saria, a 2011 Computing Innovation Fellow who recently joined the faculty of Johns Hopkins University in computer science as well as health policy and management. Suchi co-led the organization of the second annual symposium on Meaningful Use of Complex Medical Data (MUCMD) in Los Angeles, CA, with Randall Wetzel, professor of anesthesiology at Children’s Hospital Los Angeles.
Our growing health care need is one of the largest looming crises of our time. In the United States, per capita spending in health care constitutes the highest in the world and almost twice that of the country ranked second. However, our quality of care (measured by metrics such as life expectancy) is ranked below more than two dozen other countries. With the recent passage of health care reform, more of our interactions with the health system are digitized; patient symptoms, treatment decisions, and follow ups are stored indefinitely. This provides us with a fantastic lens into the health system — we can observe individual diseases, the paths of treatment decisions, and identify opportunities to improve quality of care. For example, last year at MUCMD, I presented work that showed how, using data routinely collected in the NICU, one could predict very early which infants are likely to have major complications downstream. But healthcare data are complex. To develop valuable clinical tools, it is important to know the data-generating mechanism (e.g., what biases exist in recording an observation), to understand the clinical environment (e.g., the need, who will use these tools, how will these integrate within the clinical workflow), and to develop methods that can flexibly model the complexity of this data. With these goals in mind, we organized MUCMD 2012 (more following the link…).
The second annual symposium on Meaningful Use of Complex Medical Data (MUCMD, pronounced MUKH-med) was held in Los Angeles August 9-12 at Children’s Hospital Los Angeles. About 110 attendees from over 30 different academic and research institutions, including companies like Medtronic, Siemens, HCA Healthcare, Practice Fusion, JPL, and Lockheed Martin, participated. Talks were divided across those discussing infrastructure and analysis, as well as interactive sessions to foster collaborations. Below I briefly discuss some of the talks and sessions to give a flavor of MUCMD. See the program schedule for the complete description.
Heather Duncan from Birmingham Children’s Hospital, U.K., and Peter van Manen from McLaren Electronic Systems, a U.K.-based company that provides systems to all of the competitors in the Formula One World Championship and NASCAR, jointly presented an exciting talk that showed how telemetry is used to tightly monitor and make decisions during a car race. They then inspired the audience to work towards building such a system for monitoring patients in the ICU.
Several of the talks and sessions focused on learning about problems within the current clinical environment. Randal Wetzel led a session with three other pediatric intensivists discussing problems within the pediatric ICU. Warren Sandberg from Vanderbilt University described the operating room and their efforts in integrating technology to reduce medical errors within the OR.
In a follow-up session, we broke up in mixed groups of size 10 to brainstorm more broadly about problems in and around the health system that could benefit from good engineering solutions. The need for well-developed software that smoothly integrates with clinicians’ workflows came up as a recurring theme.
On the infrastructure front, Lucila Ohno-Machado from UC San Diego discussed iDASH, a framework for sharing data between institutions. For methodologists who are interested in trying out ideas across multiple institutions, collaborating with an investigator who is at an iDASH site is a great way to start. Kenneth Mandl from Children’s Hospital Boston discussed an initiative to develop a platform called SMART which provides an interface layer to electronic medical records (EMRs) such that calls against multiple EMRs can be standardized. Once such a platform receives widespread adoption, it will make development of plug ‘n play medical apps much easier. Yolanda Gil from USC’s Information Sciences Institute discussed various workflow systems to allow researchers to keep track of experimental workflow (e.g., what values the parameters were set to, how data were transformed, which classifier was used, and so on). I could see this being especially useful for interdisciplinary research in which all of the knowhow that is assumed within a community and typically not specified in papers needs to be communicated to researchers outside the community who may be consumers of the tools. Josh Wills from Cloudera discussed when researchers should consider using Hadoop over a traditional relational database in computing with clinical data.
On the analysis front, we had several exciting talks from computer scientists, informaticians, and statisticians. These talks included prediction within the ICU by Pete Szolovits and Mark Wainwright, how to model missing data in clinical time series by Benjamin Marlin, recent developments in NLP for clinical data by Noemie Elhadad, and constraint-based clustering for analyzing multiple sclerosis patient characteristics by Carla Brodley. Susan Murphy presented analysis of randomized clinical trial data to develop rules for assigning patients to therapies within everyday practice such that treatment efficacy is maximized. Joydeep Ghosh showed an approach for integrating aggregated state level healthcare data to improve prediction for the individual (quite neat!). And a subset of the talks explored new fronts for bringing in CS research into health data research. For example, Mary Czerwinski discussed her work in HCI in developing an interface for tracking moods. Gert Lanckreit presented on crowdsourcing and its potential use for developing medical games. Katrina Ligett discussed differential privacy and whether it can bring in new tools for solving problems related to sharing of medical data.
I want to close this post by mentioning three fairly experimental break-out sessions which several of the attendees seemed to have really enjoyed. Josh Rosenthal did a hands-on session on understanding the economic value of a problem to prioritize which problems to solve. He went through why health care startups fail (a lesson that translates equally well to research) and tips for systematically avoiding these pitfalls.
In another session, Hector Corrado Bravo led an interesting discussion on high-level questions that frequently come up in interdisciplinary research. For example, a question that generated a lot of lively back and forth was, “Which is more effective — a team of multidisciplinary researchers or a team of multi-disciplinary researchers?”
The symposium ended with a hack-a-thon on Sunday. With real data from Practice Fusion, Medtronic, and Children’s Hospital Los Angeles, the hack-a-thon allowed attendees to get their hands dirty with real data. Practice Fusion is making their data available on an ongoing basis through a Kaggle competition.
We’d like to thank our generous funders including Practice Fusion, Kaggle, Perminova, JPL, the National Library of Medicine (NLM) within the National Institutes of Health (NIH), and Children’s Hospital Los Angeles (CHLA) for making it possible to host this event. We’d especially like to thank David Kale who contributed loads of hours into organizing and coordinating with a team of CHLA staff to make all of the superb local arrangements. We plan to host MUCMD again next year. Keep checking the MUCMD website for details. And all talk slides from this year will be put up at this link within the next few weeks.