Predictive software can identify risk of dangerous diseases
By Maggie Van Dyke
If an infectious disease outbreak or an attack using an agent such as anthrax were to occur in Chicago, it most likely first will be noticed in emergency rooms throughout the city. Swift identification of the cause of an incoming patients’ illness could be crucial to public health and safety personnel being able to intervene in time to save lives.
The emergency department at Rush University Medical Center has developed and implemented state-of-the-art technology to help protect the public during a health crisis: a suite of software tools called Guardian. Developed by Rush emergency physicians and researchers, Guardian can predict the presence of a variety of illnesses in patients arriving at the emergency department, whether they’re ordinary flu-like illnesses, emerging tropical diseases such as Zika or Ebola virus, or the effects of biological or chemical agents such as anthrax or sarin.
Reflecting its full name — Geographic Utilization of Artificial Intelligence in Real-Time for Disease Identification and Alert Notification — Guardian can make this assessment up to hours before physicians complete a final diagnosis for the patients.
“The Guardian system provides a feedback loop to clinicians,” says Dino P. Rumoro, DO, MPH, chairperson of Rush’s Department of Emergency Medicine. “If Guardian thinks that a patient whom I think has pneumonia might have anthrax, it could page me and say, ‘Have you considered a chest CT scan to look for evidence of anthrax?’ That page might be all I need to say, ‘You know, that’s not a bad idea.’”
Readiness is all
Fortunately, the Chicago area never has suffered a bioterrorism attack and hasn’t experienced any deadly infectious disease pandemics in recent years. Health crises don’t schedule appointments in advance, though, so Rush is primed to respond if such emergencies occur.
In 2012, Rush opened an expanded, specially designed and equipped emergency department as part of the nation’s first civilian-based, chemical and biological advanced emergency response center. Located at the base of the new hospital Tower building that Rush opened at the same time, the 60-bed state-of-the-art facility easily can be converted to handle surges of casualties, and the center’s layout and airflow can be controlled to isolate patients and to prevent the spread of infectious agents.
Rush also has a containment suite available for hospitalized patients with deadly infectious diseases and a set of protocols in place to enable clinicians to provide care for such patients without risk to others in the hospital. Rush built the suite and developed its practices in response to the worldwide Ebola outbreak in 2014 after the city of Chicago asked Rush to be part of a network of local hospitals prepared to treat patients with Ebola.
‘Make the computer think like a clinician’
Guardian continually runs in the background in Rush’s ED, helping to identify at-risk patients and disease trends. As the doctors and nurses enter information about a patient into Rush’s electronic medical record, Guardian’s analyzes the data in real time.
“We’re trying to make the computer think like a clinician,” Rumoro says.
Doing it required the development of a computerized algorithm, or decision tree, that sorts through and adds up numerous clinical variables — e.g., patient-reported symptoms, age, past medical history, blood pressure readings, blood test results — before determining whether a patient has a high risk of a certain illness. Some of the variables are assigned a greater weight in the algorithm — just like a physician would place more significance in certain clinical findings than others before making a diagnosis.
The system has become so good at tracking flu-like illness that Guardian identified the arrival of this year’s flu season several weeks before the U.S. Centers for Disease Control and Prevention and the Chicago Department of Public Health issued a formal flu warning. “For five years now, we’ve been out-predicting CDC with public flu reporting by identifying the outbreak weeks before an official public health announcement was issued,” says Rumoro, who leads the Guardian development team at Rush.
Born out of disaster
The inspiration for Guardian can be traced to the 1995 heat wave in Chicago that killed 739 people. By the time area hospitals and public health officials realized there was a crisis, it was too late.
“The emergency physicians weren’t putting two and two together,” Rumoro recalls. “We worked in air-conditioned environments and we got into our air-conditioned cars to go to our air-conditioned homes.”
Interested in whether EDs could have predicted the crisis by paying closer attention to heat-related illnesses (e.g., heat stroke), Rumoro worked with a CDC researcher to collect data from 33 area hospitals. The effort came before hospitals had computerized medical records systems, and the researchers had to leaf through big log books manually.
The effort was worth it: “We saw that hospitals started to see an increase in heat-related illnesses one week before the first person died,” Rumoro says. “We also found that when there were more than two heat-related illnesses for every 100 emergency room patients, it was a predictor of a heat-related crisis and impending mortality.”
Understanding dictation and misspellings
Thus began Rumoro’s interest in developing a disease surveillance system that EDs can use to detect emerging health threats. Guardian is the result of more than a decade of work by Rumoro and his research team, which includes PhD experts in statistics, industrial engineering and computer programming. Guardian’s development has been funded by the U.S. Department of Defense, which resulted in a partnership with Pangaea Information Technologies, Ltd. Panagea is a joint partner in Guardian Health Technologies, which is a wholly-owned subsidiary of Rush.
Guardian is equipped with advanced capabilities, including natural language processing, which enables the system to read the notes that physicians and nurses type or dictate into a patient’s electronic health record. As a result, busy ED staff members do not have to take time away from their patients to enter additional data into the record via drop-down menus or electronic forms.
In addition, Guardian is equipped with artificial intelligence, which helps the system continuously improve. “When the system misses a key word, it could be because a doctor misspelled it,” Rumoro says. “So we teach the system that this is another possible way to spell this word.”
Watching for tropical disease and bioterrorism
Guardian can be programmed to detect almost any type of illness or problem. To date, Rush has used it to identify West Nile virus, meningitis, Zika, flu-like illness, gastrointestinal illness and a variety of biological threat agents.
The program has also proven helpful in tracking potential bioterrorist attacks. During the 2012 NATO Summit in Chicago, the Rush team programmed Guardian to identify ED patients who might have been exposed to biological or chemical agents.
Eventually, Rumoro hopes to see all Chicago area EDs adopt Guardian to help in identifying and responding to health crises. Until that happens, Rush is using the system for research and public health reporting.
Helping get patients into the hospital, and back home
In addition to predicting disease risk, a new application of Guardian can forecast whether a patient will need to be hospitalized 15 minutes after the patient has arrived in the ED, even though the admission into the hospital may not take place until hours later. The accuracy of this early predictive analytic tool is between 80 and 90 percent and can help to expedite bed placement and patient flow form the ED into the hospital.
Recent work on an extension of this capability into the hospital setting eventually could help Rush and other hospitals better manage patient flow by communicating with floors in the hospitals to determine when a patient will be discharged, making room for a patient from the ED.
“Arranging to discharge a hospitalized patient requires multiple inputs from nurses, doctors and the admitting or patient bed office, that if coordinated could not only reduce waiting times in the ED but also could allow for more precisely scheduled, better planned discharges from the inpatient units,” Rumoro says.
To allow for such coordination, Rumoro and his team currently are working to expand Guardian to follow patients though their hospital stay. “Guardian would be able to predict when an intensive care unit patient is ready to move to a less-intensive bed on a medical unit. Or it would be able to say ‘This patient is ready to go home tomorrow,’” he says.
“It is not always easy to get a patient admitted to a specific hospital unit for special care, whether the patient comes from the ED, is referred from the outside, or from another place in the hospital,” Rumoro continues. “This new Guardian application has the potential to expedite the process, and the financial implications of better managing admissions and discharges is great. The reduction in variable expenses achieved through such efficiency can add to a hospital’s bottom line.”