Why sepsis prediction models fail: 4 things to know 

Listen

Sepsis prediction models rely on timely, accurate data and artificial intelligence to help clinicians, but they are plagued by issues that can contribute to their failure in clinical settings, according to clinical data engineer Angelique Russell. 

In a June 29 LinkedIn blog post, Ms. Russell, a senior clinical data engineer at Renton, Wash.-based Providence, gave four reasons that sepsis predictive models fail. Her article comes in response to a recent JAMA Internal Medicine study of Epic's sepsis prediction model, which found that the tool only detected 7 percent of sepsis cases missed by clinicians. 

Four reasons that sepsis prediction models fail, according to Ms. Russell: 

1. Many hospitals don't equip their beds with continuous monitoring technology, so patients' vital signs are measured and entered into the database manually. This slows down the reporting process for the predictive model, which then works off data that is not as timely. 

2. Hospitals tend to upcode to maximize revenue, while coders are encouraged to find enough sepsis criteria in "rule out sepsis" orders to justify more expensive bills that may not meet the strict clinical definition of sepsis. Health systems that use bill codes for clinical purposes are not as likely to be successful and can create a feedback loop that leads to overdiagnosis of sepsis. 

3. Sepsis models created at outside organizations may not be generalizable depending on that facility's clinical definition of sepsis and the at-risk populations it treats. This can affect the accuracy of the hospital using the model if its definitions aren't the same. 

4. There is no scientific evidence to guide sepsis treatment that is predicted hours in advance, which some algorithms promise to do. 

Click here to view the full report.

 

Copyright © 2021 Becker's Healthcare. All Rights Reserved. Privacy Policy. Cookie Policy. Linking and Reprinting Policy.

 

Featured Whitepapers

Featured Webinars