United StatesHealthcare and life sciences2019 – ongoingMichigan Medicine
Epic Sepsis Model alerts
Michigan Medicine paused Epic sepsis alerts after nurses reported excessive alerts during COVID-19
Problem
Michigan Medicine used automated sepsis warnings to help clinicians identify patients who might need treatment. The model calculated risk from information in the electronic health record.
Approach
Epic Sepsis Model alerts began on two hospital units in March 2019 and expanded to a third in August. In April 2020, Michigan paused the alerts after nurses reported excessive alerting during the first COVID-19 wave. A subsequent multicenter study linked the pause to rising alert volume, while explicitly stating that it had not evaluated alert accuracy. A separate account by clinical AI researchers, including the chair of Michigan's AI governance committee, described the deactivation as a response to changes in the patient population and the relationship between fever and bacterial sepsis.
Results
Takeaway
Nursing reports exposed a change in the model's operational burden. Michigan's governance committee could disable the alerts when the patient population changed.
Categories
- Size
- Large enterprise
- Industry
- Healthcare and life sciences
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Vendor solution
- Goal
- Quality and risk
- Evidence
- Independent study
Sources
Source summaries
The authors documented Michigan's April 2020 pause after nursing reports of excessive alerts. Their multicenter analysis measured alert volume rather than diagnostic accuracy and discussed how COVID-19 changed the demands placed on sepsis workflows.
The study described the model's use of electronic health record data and Michigan's deployment dates. Alerts started on two units on March 11, 2019, and expanded to a third on August 12, 2019.
The authors described Michigan's deactivation of Epic's sepsis model during COVID-19 as a response to dataset shift. They reported that the hospital's clinical AI governance committee decommissioned its use after spurious alerting.
Full details are in the original sources.