Experts from Facebook and genetics labs team up to help doctors make personalized predictions about their patients.

http://www.technologyreview.com/news/518916/a-hospital-takes-its-own-big-data-medicine/

On the ground floor of The Mount Sinai Medical Center’s new behemoth of a research and hospital building in Manhattan, rows of empty black metal racks sit waiting for computer processors and hard disk drives. They’ll house the center’s new computing cluster, adding to an existing $3 million supercomputer that hums in the basement of a nearby building.

The person leading the design of the new computer is Jeff Hammerbacher, a 30-year-old known for being Facebook’s first data scientist. Now Hammerbacher is applying the same data-crunching techniques used to target online advertisements, but this time for a powerful engine that will suck in medical information and spit out predictions that could cut the cost of health care. With $3 trillion spent annually on health care in the U.S., it could easily be the biggest job for “big data” yet. “We’re going out on a limb—we’re saying this can deliver value to the hospital,” says Hammerbacher.

Mount Sinai hopes data will let it succeed in a health-care system that’s shifting dramatically. Perversely, because hospitals bill by the procedure, they tend to earn more the sicker their patients become. But health-care reform in Washington is pushing hospitals toward a new model, called “accountable care,” in which they will instead be paid to keep people healthy.

One person he’ll be working with is Joel Dudley, director of biomedical informatics at Mount Sinai’s medical school. Dudley has been running information gathered on diabetes patients (like blood sugar levels, height, weight, and age) through an algorithm that clusters them into a weblike network of nodes. In “hot spots” where diabetic patients appear similar, he’s then trying to find out if they share genetic attributes. That way DNA information might add to predictions about patients, too.

A goal of this work, which is still unpublished, is to replace the general guidelines doctors often use in deciding how to treat diabetics. Instead, new risk models—powered by genomics, lab tests, billing records, and demographics—could make up-to-date predictions about the individual patient a doctor is seeing, not unlike how a Web ad is tailored according to who you are and sites you’ve visited recently.

That is where the big data comes in. In the future, every patient will be represented by what Dudley calls “large dossier of data.” And before they are treated, or even diagnosed, the goal will be to “compare that to every patient that’s ever walked in the door at Mount Sinai,” he says. “[Then] you can say quantitatively what’s the risk for this person based on all the other patients we’ve seen.”