Until recently, valuable data capable of saving the lives of heart attack victims
and babies born prematurely was usually ignored or discarded.
The piles of printouts in the image at left represent 10 hours of data from an electrocardiogram (EKG), a device that helps doctors treat heart disease. (Readings from an entire day would stretch two miles.) Standard operating procedures call for a physician to print out only a small segment of data from an entire day in order to determine a patientâ??s progress following a heart attack. Not surprisingly, these cursory reviews can miss crucial clues.
Three researchersâ??MIT computer scientist John Guttag and cardiologist and MIT computational biologist Collin Stultz (both shown at left, along with University of Michigan computer scientist Zeeshan Syedâ??created a new computer model to analyze EKG data that used to be thrown away. Using data mining and machine learning techniques to sift through the massive quantities of data, the researchers found that three abnormalities in an EKG are correlated with a two to three times higher risk of dying from a second heart attack within a year. Guttag, Stultz, and Syed figure their new computer model will significantly improve todayâ??s risk-screening techniques, which miss identifying about 70 percent of the patients who are particularly likely to have a repeat heart attack.
The challenge is that most patients will recover simply by improving their diet and exercising more. But a small percentage, about 5 percent, need aggressive intervention, such as an implanted defibrillator. â??Our work is designed to figure out who belongs to which population,â? says Guttag.
A similar story is unfolding in premature baby intensive care facilities, such as the one above, where 20 percent of all premature babies contract serious infections while in the hospital.
Modern intensive care units (ICUs) are overflowing with data, but Carolyn McGregor, a computer scientist at the University of Ontario I


Community guidelines

Be sure to review our Community Guidelines. By continuing you are agreeing to our Terms of Service and Privacy Policy.