NASHVILLE, TENN., SEPTEMBER 9, 2020—Predictive analytics can drive healthcare performance improvement and program innovation by anticipating areas of need, helping organizations determine where to direct resources to optimize care outcomes. To build momentum and advance the science in this important area, the Agency for Healthcare Research and Quality (AHRQ) recently fielded a new challenge competition, “Bringing Predictive Analytics to Healthcare,” designed to assess the reliability and validity of varying predictive models. Ursa Health, a data analytics software and professional services company, ranked third in the competition.
According to Gopal Khanna, director of the AHRQ, “We know that decisions makers sometimes must make policy decisions quickly and can’t wait for traditional research methods that use more rigorous analysis of information. Predictive analytics and innovative use of data will offer a solution in those cases. Other industries have used predictive analytics and related methods successfully. But the healthcare industry is lagging behind. Until now.”
AHRQ fielded the challenge in 2019, inviting applicants to use machine learning and related methods to estimate vital healthcare outcomes, hospital inpatient utilization, and average lengths of stay for selected counties in the United States. AHRQ provided applicants with access to customized analytic files that included information on hospital inpatient discharges for years 2011 to 2016, which applicants used to predict the outcomes for the following year, 2017.
Applications were rigorously evaluated based on the reliability and validity of their predictive models. Reliability was defined as how closely the models predicted actual utilization rates for 2017, while validity was defined as how well the models performed on earlier years of data.
Explained Colin Beam, Ph.D., director of advanced analytics at Ursa Health and architect of the award-winning solution, “A wide variety of factors drive patient utilization, so it was important to assemble data from many different sources. We used health and environmental data from the Centers for Disease Control and Prevention and the University of Wisconsin’s Area Deprivation Index, and worked to capture some of the geographic variation in utilization with data from the U.S. Department of Housing and Urban Development. We then applied a gradient boosting algorithm for our predictions, which typically works well when applied to many diverse data sources.”
”Predictive analytics are a powerful tool enabling healthcare organizations to target finite resources to the patients and programs most likely to produce the best outcomes,” explained Robin Clarke, M.D., Ursa Health’s chief executive officer. “The AHRQ challenge nicely demonstrated how developing high-performing models requires not only advanced programming but also clinical nuance to appreciate the complex factors that drive healthcare in the United States.”
Winners included HCA Healthcare-NC Division (first place), Premier, Inc. (first place), Children’s Hospital (Aurora, Colo., second place), and Kalman & Company, Inc. (third place).