Where Industry Meets Innovation

  • Contact Us
  • sign in Sign In
  • Sign in with certificate
mit campus


Search News

  • View All
  • ILP News
  • MIT Research News
  • MIT Sloan Management Review
  • Technology Review
  • Startup Exchange

ILP Institute Insider

May 9, 2016

Crowdsourcing Clinical Care Research

Leo Anthony Celi builds open health databases for collaborations between physicians and engineers.

Eric Bender

We’re awash in medical data, but doctors get surprisingly little use out of it, says Leo Anthony Celi, physician and data scientist. “Most of the time our decisions are based on personal knowledge as well as biases, and as a result we see a lot of variation in care,” he says.

Leo Anthony Celi
MIT Principal Research Scientist
Institute for Medical Engineering & Science
Celi, who is clinical research director at the MIT Laboratory of Computational Physiology, works to overcome this challenge by boosting the power of data-driven healthcare. “The push towards electronic health records offers a big opportunity to leverage the value of data, and to help us as doctors provide more individualized recommendations in terms of diagnostic tests and treatments,” he says.

Exploiting these health databases effectively requires opening them up to collaborations between clinicians and data scientists, emphasizes Celi, who works in intensive care at Beth Israel Deaconess Medical Center in Boston and is an assistant professor of medicine at Harvard Medical School.

The Medical Information Mart for Intensive Care (MIMIC) database, which archives data from patients admitted to Beth Israel Deaconess’s intensive care unit, offers one case in point.

With MIMIC, “we’re trying to change the culture to bring together data scientists and clinicians to work together and answer research questions posed by the clinicians,” says Celi. “The clinicians know the information gaps that they encounter in day-to-day practice. They lack the methodological expertise that is brought in by the data scientists, who are interested in transforming healthcare but have a superficial understanding of the problems in healthcare.”

Celi and his co-workers often launch these collaborations in “datathons,” medical versions of software hackathons that team up the two groups for brief intense events.

He’s also the founder of Sana, a volunteer organization hosted at MIT’s Computer Science and Artificial Intelligence Lab that aims to spark interdisciplinary medical research in low- and middle-income countries (LMICs).

Sana offers an open-source telemedicine platform created to exploit ubiquitous cell phones. But Celi emphasizes that technology is just part of a solution, which must be defined and driven by colleagues in those countries. “The people who live in the LMICs are the ones who are immersed in the problems and have the best perspectives on the complexities of those issues,” he explains. “It’s preposterous to think that we as MIT faculty and students understand the problems as well as they do.”

Seeking clarity in intensive care
The open-access MIMIC database has drawn more than 2,000 investigators from around the world to ask questions about best practices in intensive care. “Different groups can ask the same questions, and critique and cross-validate each other’s findings,” Celi points out.

In one instance, researchers aim to predict the growing numbers of patients who survive the ICU but die within a year in a nursing home, he says. The goal is to create a tool that will aid clinicians in discussions with the family about whether this outcome would reflect the patient’s wishes.

Celi and his colleagues are partnering with Philips Healthcare, a supplier of electronic health record systems, to scale up MIMIC with intensive-care data from many other institutions, including hospitals in other countries. “Databases like these are not trivial to develop, because you need to bring together data coming from disparate sources,” he says.

Research on MIMIC often begins with datathons that are held over a weekend. Kicking off these events, the clinicians pitch their problems and the data scientists each sign on for a problem. The researchers work together, and usually by the end of the weekend they can show some preliminary findings. “This is a very abbreviated, accelerated way of discovering knowledge,” Celi says.

“Bringing together data scientists and clinicians sounds so easy, but that’s the biggest obstacle in this approach,” he adds. “At MIT we take this collaborative approach to problem-solving for granted. But these groups are typically not used to working with each other, they speak different languages, and they don’t want to be pushed outside their comfort zones. It is even harder to do in other countries.”

Leading locally
Celi’s interest in improving care for underserved populations led him to form Sana. The organization brings together experts in clinical medicine, public health, engineering, computer science, and business to address health problems in countries with fewer health resources, he says.

Sana is built on an open source telehealth platform tailored to cell phones. “When we started, we were focused on this technology,” Celi notes. “Very quickly we learned that technology is the easiest part of the puzzle and that seldom is technology alone sufficient to address the big issues in healthcare.”

“One of the bigger problems that we encountered is a lack of culture of quality improvement,” he says. “We try to nurture this culture, which is a requirement for us to introduce any innovations. And if the people on the ground don’t appreciate what we’re trying to do, and are only drawn to the sexiness of the technology, typically we can’t sustain or scale an innovation. So we stepped back to try to improve the educational piece, so that we all understand the complexity of the problem that we are trying to solve.”

Crucially, partners in these countries choose the problems that are most relevant to them. “They are at the heart of designing the solutions, with our help,” says Celi. “They are also the ones who raise the funding to test the innovation that they have designed. We help them in securing that funding, either from their government or from a foundation, but they are the owners of the project. In that case the accountability for success is much higher, and they are really motivated to move on with a project.”

In a recent datathon in Mexico City, Sana partnered data scientists from the Tec de Monterrey University with geriatricians from a local hospital. The event produced several promising prototype applications for elderly care. For instance, one app would help to provide patients with exercises to control urinary incontinence. Another “Uber-like” app would connect caretakers with families, and a third would allow elderly people to seek others with similar interests, reducing their social isolation.

Partnerships for problem-solving
Industry support is key for assembling robust databases and turning potential solutions into clinical products and services, Celi emphasizes.

“MIMIC would not have been possible without the help from Philips Healthcare,” he says. “We worked with them to archive all the data coming from their system and provide that data for research purposes.”

Clinical records are gathered by many different information vendors, “and we need to bring the industry partners together,” Celi says. “Data integration and analysis are what meaningful use should be all about. Vendors and health organizations hoarding data is a tremendous waste of opportunities that can benefit everyone.”

He also sees a leading role for industry in integrating data from outside the clinic, since behavior often influences clinical outcomes but is rarely well documented in electronic health records.

“You need other sources, such as social media and cell phone use,” Celi says. “That becomes a big challenge because of the issues of privacy and security, which are very daunting. And it won’t be easy to work through all this data, which will be full of noise.”

“People are always looking for the next big discovery in medicine, and typically it’s not one discovery,” he points out. “You need various disciplines working together, various brilliant minds slicing and dicing the database in all sorts of directions, to come up with the best answers. We want to involve more clinicians in this process of knowledge discovery. We want to open the floodgates of research.”