Dignity Health fights sepsis with natural language processing, targeted alerts

At a Dignity Health hospital in Northern California, a woman who had just given birth was showing early signs of septic shock.

Pic of Joseph Colorafi
Joseph Colorafi

Although clinicians noted her elevated heart rate, many of the other indications remained undetectable. Instead, an algorithm picked up on those abnormalities and alerted the nurse on duty. After conferring with a physician, the patient was transferred to the ICU for treatment.

That was the real-life anecdote relayed by Joseph Colorafi, M.D., vice president and chief medical information officer at Dignity Health during a presentation at the SAS Health Analytics Executive Forum on Wednesday. Colorafi, a former internist, said there was “nothing more worthwhile” than the work his system has done using analytics to recognize symptoms of sepsis and prevent situations where “patients are going over the waterfall, and far downstream.”

RELATED: Sepsis drives more readmissions than medical conditions tracked by CMS

Using real-time data, natural language processing and automated alerts, Dignity Health is improving the process of identifying sepsis as soon as possible—when survival rates increase significantly—along with specific outcomes associated with the infection, including mortality and ICU length of stay.

Rather than using its EHR system, Dignity developed a clinical decision support software called the Dignity Health Insights Program that operates on top of the EHR, allowing it to capture information from patients across each hospital and root out real-time changes to patient conditions.

Although Cerner and Epic are viewed as leaders in the EHR industry when it comes to integrating sepsis tools, Colorafi said the constraints of an EHR system don’t allow for in-depth analytics that independently constructed algorithms provide.

RELATED: 44 children’s hospitals unite to combat sepsis

A key part of the tool is a CMS core measure known as “time zero,” which outlines specific actions clinicians must take within three hours of identifying a suspected infection. An electronic dashboard designed by the hospital counts down from 180 minutes and a color-coded chart indicates what parts of the sepsis bundle have been completed.

Better compliance means better outcomes. For every 10% increase in compliance with the sepsis bundle, mortality reduces by 3%, Colorafi said, adding that that compliance ceiling hovers around 60%-65%.

“That’s the confirmation we want to drive, drive, drive the bundle compliance rate because we know what it’s going to do to mortality,” he said.

RELATED: Hospital Impact—With innovation, the end of sepsis is possible

Across 12 hospitals, sepsis mortality rates have dropped 5%, according to Colorafi, with some hospitals dropping as low as 13%. The system also has the capability of retroactively testing new studies, including recent research that indicates delirium or decreased renal function could also be indicators of a sepsis.

Dignity has mimicked this approach to drive clinical decision support initiatives for five or six other diseases that benefit from real-time alerts and tap into the 95% of EHR data that never makes it past the encounter phase.

“If you don’t add natural language processing into some projects, that remains as dark data,” Colorafi said.