A team from the Seattle-based Institute for Systems Biology won the $70,000 first prize for matching unstructured data from medical journals with structured data from genomic and proteomic registries to study thousands of patients with more than 25 varieties of cancer, according to YarcData. Their work used YarcData's new uRiKA graph analytics appliance to help understand how each type of cancer disrupts various biological networks and to identify drugs approved for other conditions that might help treat certain cancers.
In second place were researchers from the University of California at Santa Barbara, who probed clusters of disease precursors for clues that might shed light on the causes of autism. Dr. Abraham Flaxman, an assistant professor at the University of Washington's Institute for Health Metrics and Evaluation, took third for developing a computer model to predict patient risk and mortality in the 30 days after a heart attack.
"The winning entry demonstrated classic features of big data problems: a large volume of complex information from a variety of disparate sources and a need to manage data input and output at a high velocity. The team's submission exemplified the power of graph analytics for gaining insight into difficult knowledge-discovery problems," one contest judge, Dr. Bruce Hendrickson, senior manager for computational sciences and mathematics at Sandia National Laboratories in New Mexico, said in a statement from YarcData.
In an interview with InformationWeek Healthcare third-place winner Flaxman was more humble about his achievement. "I'm not here to tell you I've solved the problem of predicting 30-day mortality [for individual patients]," he said. "This is where computer scientists have been overpromising for years."
[ IBM's cognitive computing platform could do much more than win at "Jeopardy." See IBM's Watson Could Be Healthcare Game Changer. ]
However, using the uRiKA technology and records from the federal Agency for Healthcare Research and Quality's state inpatient databases, Flaxman applied the K-Nearest Neighbor algorithm to help estimate mortality risk for populations based on patient age, sex, condition and hospital.
The uRiKA appliance essentially is a Web interface, supporting SPARQL queries, for supercomputer technology that otherwise would be difficult to harness the power of, according to Flaxman, who wrestled with a supercomputer at the U.S. Department of Energy's Pacific Northwest National Laboratory in Richland, Wash. "It was extremely promising but it had a high bar for being able to use it," Flaxman said of the lab's computer.
Flaxman, a mathematician by training, said he picked 30-day risk for heart attacks because there is a wealth of data available and because it is important to hospitals. Indeed, the federal Medicare insurance program for Americans 65 and older no longer reimburses hospitals for preventable readmissions within 30 days of initial discharge for heart attack, heart failure and pneumonia.
Estimating mortality at the population level essentially provides a view of how well hospitals are allocating their resources, Flaxman explained. For example, if a facility has a high number of deaths of patients presenting with heart attacks, the graph analysis can help determine if the mortality rate is due to a large group of patients who already were at high risk of dying.
Flaxman said his next step in refining his research will be to define exactly what it means to do a good job in adjusting mortality data for risk and then finding the best ways to perform risk adjustment. "Graph analytics seems like a really promising approach," he said.
Regulatory requirements dominate, our research shows. The challenge is to innovate with technology, not just dot the i's and cross the t's. Also in the new, all-digital The Right Health IT Priorities? issue of InformationWeek Healthcare: Real change takes much more than technology. (Free registration required.)