Penn Medicine uses predictive analytics to reduce sepsis mortality

BOSTON—Penn Medicine set out to decrease its sepsis mortality index (SMI) since the rate of incidence was increasing and the organization’s SMI was higher than the median of the University HealthSystem Consortium. Christine VanZandbergen, Penn Medicine’s associate CIO of clinical applications, discussed the deployment of an automated, real-time intervention that would help providers intervene earlier at the Big Data Healthcare Analytics Forum on Nov. 20.

Systemic inflammatory response syndrome (SIRS) criteria—temperate, heart rate, white blood cell count and respiratory rate--offers some guidance in identifying sepsis early. If a patient has four or more points, sepsis is presumed. But, “even if they don’t have sepsis, the patient is pretty sick,” she noted.

They first looked at a retrospective dataset of 193 points with scores of 4 or more. There was a positive predictive value of about 23 percent for death, transfer to ICU or rapid response team. On average, these events occurred over 48 hours after the score of 4 was first met. “We felt there were an additional 20 percent of patients who had sepsis and would be identified by this trigger and could benefit from early diagnosis and treatment,” VanZandbergen said. The early warning system was likely to also identify patients who had new episodes of bleeding, cardiac ischemia or pulmonary emboli.

They tried changing the threshold values to increase the positive predictive value “but this came at a price in terms of sensitivity. We would miss five deaths.”

The organization conducted a “silent” one-month pilot with a goal of validating the retrospective analysis in a live environment.

Analytics and predictive modeling is a “small piece” of the overall effort, she said. “End-user tool development is really where we’ve gleaned our success.” After the team built the tool and decided on an operational response, they identified the three critical people on the care team: the nurses, primary physician and coordinator role. “That [coordinator] role served as the glue. It’s different at each site but the process is the same.”

The pilot also validated that a score of 4 was an acceptable positive predictive value, she said. Operations teams determined they could handle the volume with current staff. They excluded certain patient populations, had alerts fire once per visit and limited the new process to patients admitted at the go live or after to gradually ramp up and not overwhelm the staff.

The tool was turned on in 2012 and “we now have the benefit of two years of data,” said VanZandbergen. It’s been a reliable system in terms of notification, she said, with the three critical team members notified via text message. The coordinator is easiest to contact because providers have to self-identify when they come on shift. “If they don’t do that, we can’t send them a text page. We have institutionally supported devices but some carry their own devices.” They get a pop-up when they log into the system.

The early warning system asks providers to validate that the vitals are correct and give an opinion on whether the patient is indeed septic. The coordinators perform follow-up tasks and answer questions such as whether everyone showed up and did so within 30 minutes, was the patient septic and if so, did you already know and did the alarm have value. Thirty percent said it did and “we felt good about that,” she said. “We’ve seen sustained value to this.” That includes a firing rate that has decreased by 20 percent.

The perceived effectiveness of the early warning system is “incredibly important to us in our longevity to this type of work.” Forty percent of clinicians said the system alerted them to new information and 46 percent of nurses said the same. More than one-third (34 percent) of physicians said the system is helpful while 40 percent said the same.

The organization has achieved its goal with a SMI of .92 which is lower than its predicted mortality rate. “Our next goal is to get to the top 5 performers.”

Meanwhile, long-term care and rehab have asked for the system. “We’re pretty encouraged by that.”

Beth Walsh,

Editor

Editor Beth earned a bachelor’s degree in journalism and master’s in health communication. She has worked in hospital, academic and publishing settings over the past 20 years. Beth joined TriMed in 2005, as editor of CMIO and Clinical Innovation + Technology. When not covering all things related to health IT, she spends time with her husband and three children.

Around the web

The tirzepatide shortage that first began in 2022 has been resolved. Drug companies distributing compounded versions of the popular drug now have two to three more months to distribute their remaining supply.

The 24 members of the House Task Force on AI—12 reps from each party—have posted a 253-page report detailing their bipartisan vision for encouraging innovation while minimizing risks. 

Merck sent Hansoh Pharma, a Chinese biopharmaceutical company, an upfront payment of $112 million to license a new investigational GLP-1 receptor agonist. There could be many more payments to come if certain milestones are met.