June 5, 2017

Machines can already predict who's going to the hospital with 82 percent accuracy. What's next?

Daily Briefing

    Machine-learning algorithms have the potential to improve patient outcomes and reduce costs, Yannis Paschalidis, a professor of engineering and the director of the Center for Information and Systems Engineering at Boston University, writes in Harvard Business Review.

    Get your readmission reduction toolkit

    Paschalidis highlights a machine-learning project that can predict hospitalizations for heart disease and diabetes with an accuracy rate of about 80 percent—giving clinicians a chance to intervene and prevent hospitalizations.  

    The project—a joint effort between Boston University and Boston-area hospitals, including the Boston Medical Center and Brigham and Women's Hospital—uses algorithms to evaluate anonymized EHR data, such as admissions, procedures, vital signs, medications prescribed, lab results, diagnoses, and demographics.

    The researchers found they could predict readmissions for heart disease and for diabetes about one year ahead of time with an accuracy rate of up to 82 percent, Paschalidis writes. According to Paschalidis, the accuracy rate exceeds "what is possible with well-accepted risk scoring systems such as the one that emerged from the famous Framingham Heart Study." The Framingham study can predict hospitalizations with an accuracy of roughly 56 percent, according to Paschalidis.

    Further, Paschalidis writes that the researchers "found that feeding the factors used in the Framingham 10-year risk score into more sophisticated machine-learning methods still leads to results inferior to ours (an accuracy rate of about 69 percent)." According to Paschalidis, this implies that researchers can garner "superior prediction results" by using a complete EHR record—which can include up to 200 factors—rather than "a few key factors."

    Moreover, an algorithmic approach can be scaled to a large population relatively easily, Paschalidis writes.

    Health care could see "enormous" benefits by applying machine learning, Paschalidis contends. He cites estimates from the Agency for Healthcare Research and Quality that estimated 4.4 million admissions in one year could have been prevented by using machine learning, which would have saved $30.8 billion. About half of those costs were for patients with heart disease or complications from diabetes, Paschalidis adds.

    Further, faced with pressure to assume more financial risk, "hospitals are increasingly making analytics and new technologies an integral part of hospital operations," Paschalidis notes. He points out that researchers are developing algorithms to help doctors make diagnoses.

    "These advances are only the tip of the iceberg," Paschalidis states, adding, "(I)magine what is possible if we can tap into [the] trove of personal data."

    Acknowledging that there are "critical concerns" when it comes to the "privacy, security, and reliability of new systems and methods," Paschalidis calls for efforts to work to "strengthen" existing methods and regulations—as opposed to "retreating from this new era" (Paschalidis, Harvard Business Review, 5/30).

    Population Health Analytics 101

    Join us on June 21 for a webconference on Population health Analytics, an integrated health solutions platform that brings together underlying data assets from different programs into a user-friendly interface.

    During the one-hour training, one of our Population Health experts will guide you through the new platform and what it entails.

    Register for the Webconference

    Topics

      X
      Cookies help us improve your website experience. By using our website, you agree to our use of cookies.