Epic’s sepsis early warning system, which notifies clinicians of potential signs of sepsis in their patients, sent out an overload of alerts during the COVID-19 pandemic that may have contributed to alert fatigue, a new study suggests.
The Epic Sepsis Model is a tool in Epic’s electronic health record (EHR) system that uses predictive analytics to identify warning signs of sepsis in patients. Hundreds of U.S. hospitals use the feature, the company previously told Fierce Healthcare.
But researchers at the University of Michigan found a dramatic jump in the number of alerts received by clinicians in the early stages of the pandemic, which they said may have burdened providers already overwhelmed with increasing patient acuity.
Across the 24 hospitals analyzed in the study published in JAMA Network Open, the total number of sepsis alerts per day increased 43% in the three weeks after the hospital’s first COVID-19 case, even as total hospital census decreased by 35%.
The University of Michigan paused Epic’s sepsis alerts entirely in April 2020 in response to complaints about over-alerting, researchers said.
RELATED: Sepsis hospitalizations cost Medicare $41.8B in 2018—and that number’s likely to grow
At least 1.7 million people in the U.S. develop sepsis each year, and nearly 270,000 people die from it, according to the Centers for Disease Control and Prevention. The condition comes with a wide array of symptoms such as increased heart rate, fever and confusion, which can make it difficult to weed out sepsis cases from the vast number of problems that may cause the same symptoms.
The study didn’t examine whether Epic’s model was accurately predicting sepsis cases. However, the researchers said that even if the alerts were accurate, they may not have been "entirely appropriate” in the pandemic’s context.
Most sepsis workflows are built around bacterial rather than viral sepsis, researchers said; bacterial sepsis is the most common form of sepsis, but viruses like COVID-19 can cause sepsis, too.
The study also found that the percentage of sepsis alerts in the hospitals rose from 9% to 21%. However, it concluded the total increase in the number of alerts was more reflective of the alerting burden since the proportion increase could be explained by the decreases in elective surgeries and higher patient acuity.
RELATED: Epic’s widely used sepsis prediction model falls short among Michigan Medicine patients
The researchers said the findings indicate a need to carefully monitor artificial intelligence algorithms after they're deployed, especially in the event of a dramatic shift in hospital resources and patient acuity like that experienced during the COVID-19 pandemic.
Epic did not respond to requests for comment in time for publication.
The EHR vendor’s sepsis model came under fire earlier this year, too. A June study of the model’s effectiveness at the University of Michigan Medical School in Ann Arbor found its output to be less reliable among their patients than Epic had claimed, signaling a need for external validation of sepsis models.
In response to the findings, Epic said the researchers’ threshold wasn’t appropriately tuned for real-world clinical use and that the company’s formula and model inputs were available for systems administrators to test its reliability.