Sepsis is a leading cause of death in hospitals. It is a potentially fatal immune reaction to systemic infection brought on by a bacterial, viral, or fungal pathogens quietly entering patients through intravenous lines, surgical wounds, or other lesions. The condition can be subtle and hard to detect at first, but once it gains a foothold it proceeds quickly and can be very hard to treat; the mortality rate for sepsis is between 20 and 60 percent.
Closer, more nuanced monitoring of patients would help doctors identify the risks of sepsis in individual patients. To achieve this, a CITRIS-supported collaboration is developing an online tool that will exploit the UC Davis Medical Center’s data-rich electronic medical records (EMR) so that doctors can evaluate the likelihood that a patient will become septic, and recommend tailored treatments. The tool, called a Clinical Decision Support System (CDSS), will use cutting-edge machine learning algorithms that continually update probabilities based on ever-changing data, such as vital signs, hospital tests, and patient self-reports.
The cross-campus collaboration among the UC Davis Medical Center (UCDMC), UC Davis, and UC Merced College of Engineering represents the evolution of a more basic digital tool that tracks sepsis symptoms and, when detecting two or more, sends an alarm to nearby nurses. Because such a straightforward approach sets off many false alarms, it tends to create “warning fatigue,” says UC Davis physician Timothy Albertson, a co-PI on the project.
“It is like a dashboard light that goes on from time to time in my old Honda Civic. After a while you just learn to ignore it,” says Albertson, who is also Chair of Internal Medicine at UCDMC.
The new CDSS will track a much broader range of indicators than its predecessor, and it will compare and analyze them continuously. The collaborators are developing the tool initially with medical records from 1,492 patients, with the goal of eventually using the much larger database created by the 500 or more patients UCDMC treats every day. Each patient generates thousands of data points each day, creating a gold mine of information, says Ilias Tagkopoulos, Assistant Professor of Computer Science at UC Davis, who is also the principal investigator in the CITRIS grant.
“As more data comes from each patient, the tool integrates it into an updated analysis and informs the staff about what the benefits of doing a particular test or taking a particular course of treatment would be,” Tagkopoulos says.
“When a danger threshold is crossed and the new tool warns a nurse, the specificity of the information will be much greater,” says Albertson. The CDSS will suggest a particular action based on its analysis as well as its probable outcomes. The actual outcomes will then be factored back into the system. With each new test result and symptom, the tool updates its analysis. “It continues to increase its predictive capacity by becoming better, more specific, more knowledgeable” says Tagkopoulos. “It is a tool that learns and adapts to new information.”
The researchers’ main goals are to decrease mortality and improve outcomes for patients. Sepsis cases are very expensive, so it also stands to save patients and insurers a lot of money.
Secondary benefits may prove equally important. Sepsis has a wide range of causes and follows various courses in different patients. Much is still unknown about what happens in the immune system when sespsis catches hold. Widespread inflammation due to immune response sometimes appears to do the most damage. In other cases the systemic infection itself seems to do the most harm. The exact mechanisms at play are still not completely understood. Albertson expects the new CDSS to “turn out strong hints about the pathophysiology of the disease,” he says. And those could lead to better prognosis and treatment.
By identifying new associations between test results, vital signs, symptoms, and treatments that are suggestive predictors of sepsis onset, the tool may well indicate fruitful research paths into the mechanism underlying sepsis.
“When you start chewing on these kinds of huge data sets in real-time, with real people, in real life, meaningful associations are going to pop out,” says Albertson. If the program is adopted into a larger EMR complex, such as Epic, the most widely used EMR system in the United States, the bigger database could unearth even more subtle associations of value to both doctors and researchers.
One way to do that would be to eventually plug genetic analysis and other biomarker identification methods into the tool. Such added data could help doctors evaluate a patient’s risk of sepsis even before hospital admission. “We hope to extend our efforts so that we can identify high-risk groups and individuals whom we should monitor even more closely,” says Albertson.
Finally, says Tagkopoulos, the methods used in the tool could be adapted to other areas of healthcare. “While our tool is being tailored for sepsis, the underlying methods can be applied to a lot of different clinical conditions: acute coronary syndrome, diabetic ketoacidosis, and maybe even cancer,” he says.
We are entering an era when the large amounts of data generated each day can be used to promote understanding of different diseases and to benefit all patients. This CDSS will be among the first generation of tools to truly exploit this potential, says Albertson.