Surgical wound infections are among the most common reasons surgical patients are readmitted to the hospital. A study of nearly 500,000 operations, published in the Feb. 3, 2015, edition of the Journal of the American Medical Association, identified such infections as the top cause of unplanned readmissions.1
To address this concern, the University of Iowa Hospitals and Clinics began using predictive analytics software to identify which surgical patients were most at risk for infection and take precautionary steps. By the end of 2014, it had reduced the rate of infections for colon surgery patients by 58% over a two-year period.
“The logic behind predictive analytics is to compare a patient’s history, medical condition, and in-surgery vital signs and complications against a model that associates specific factors with levels of risk for developing specific medical conditions,” says John Cromwell, MD, associate chief medical officer and director of surgical quality and safety at the University of Iowa Hospitals and Clinics. For instance, a patient who was seriously immunocompromised before surgery and then experienced vital-sign fluctuations, serious blood loss and a substantially larger-than-expected surgical wound would be more likely to develop a postoperative infection than someone who was healthy before surgery and had no issues during the procedure.
A Decade in Development
Cromwell’s predictive analytics approach grew out of work he was doing nearly a decade ago, developing a noninvasive device for monitoring patient health. “We realized that the predictive analytics that were being used to process data from that device could be broadly applied to deal with bigger hospital issues such as postoperative infections and unplanned patient readmissions,” he says.
To test this idea, Cromwell and his research team analyzed the surgical outcomes of 1,600 patients, balancing their results against the hospital’s surgical and postoperative procedures, previous outcomes for similar operations and associated known risks of surgical infections. They validated their model against another known population of patient outcomes, then plugged into Dell’s Statistica predictive analytics software.
“This software takes patient history and preoperative data, information from blood pressure monitors and blood oxygen level detectors, and data input during the operation itself,” says Cromwell. “This latter data includes blood loss during surgery, fluctuation in their heart rate and blood oxygen levels, and issues with the wound itself, including any infection at the time of surgery.” Statistica then compares this data against the model to predict the patient’s probability of developing a postoperative infection. “This allows us to proactively treat high-risk patients before they leave the table,” Cromwell says. “It also allows us not to treat patients with low risk rates, because the chances of them developing post-op infections are very slim.”
The potential for the University of Iowa Hospitals and Clinics’ predictive analytics model to reduce post-op infections and readmissions could lead to many benefits, including:
Better health outcomes;
More efficient use of provider resources;
Reduced risk of malpractice suits;
Less stress on patient and family;
Lower patient medical expenses;
Decreased patient time away from work.
The University of Iowa Hospitals and Clinics is proceeding with its use of predictive analytics on a small scale. “Our goal is to expand this application beyond colon surgeries to encompass a range of procedures, and to expand its usage to other hospitals that we are associated with,” says Cromwell. “The potential for predictive analytics to substantially improve patient outcomes is very, very real.”
1 Merkow, R, et al. Underlying Reasons Associated With Hospital Readmission Following Surgery in the United States. JAMA. 2015; 313(5): 483-495; http://jama.jamanetwork.com/article.aspx?articleid=2107788.
James Careless is a freelance writer with extensive experience covering computer technologies.
Sidebar: Predictive Analytics: Maine Knows Who’s Headed to the Hospital
Maine is using a computerized tool—thought to be the first of its kind—that canvasses residents’ electronic health records (EHRs) and uses modeling software to predict who is most likely to end up in the emergency department (ED) or admitted to a hospital, and who might be headed for a heart attack or stroke. This lets care providers intervene early in high-risk cases and, according to the Bangor Daily News, “hopefully prove their computers wrong.”
HealthInfoNet, which runs the computer exchange holding Maine citizens’ EHRs, led development of the predictive analytics tool. It worked with HBI Solutions, which uses Stanford scientists to predict variations in the human genome. HBI believed the same pattern-hunting approach could work with medical information. Its model starts with patients who have already visited EDs, then works backward to evaluate other factors. The initial list it yielded of patients expected to visit EDs within six months was 74% correct. The tool is now being tested by providers.
Predictive analytics aren’t new in healthcare, the newspaper notes, but have largely been based on dated data from insurance claims. The Maine tool uses clinical data no more than 24 hours old. HealthInfoNet’s EHR network includes 32 of 36 state acute-care hospitals, more than 300 outpatient facilities, and providers of behavioral health and long-term care. That enables it to identify patients who use multiple resources across systems. St. Joseph Healthcare reports the tool has already helped reduce readmissions.
HealthInfoNet and HBI are promoting the tool—a subscription to which starts at $14,000 a year—to HIEs in other states. They may find a receptive audience; use of HIEs for public health surveillance is on the rise in the U.S.