MEDICAL: Artificial Intelligence in EHRs


By White Hat Anonymous

Epic Systems, the country’s leading e-health record company, says an algorithm it developed can accurately flag sepsis in patients 76% of the time. The life-threatening disease, which arises from infections, is a major concern for hospitals: One-third of patients who die in hospitals have sepsis, per the CDC. 

  • Generally, the earlier sepsis is diagnosed and treated, the better a patient’s chances of survival—and hundreds of hospitals use Epic Systems’s sepsis prediction model, The Verge reports. 

The problem: According to a study published this week in JAMA Internal Medicine, Epic Systems may have gotten the success rate wrong: The model is only correct 63% of the time—“substantially worse than the performance reported by its developer,” the researchers wrote. 

  • Part of the issue can be traced to the algorithm’s development, Stat News reports. It was trained to flag when doctors would submit bills for sepsis treatment—which doesn’t always line up with patients’ first signs of symptoms. 
  • “It’s essentially trying to predict what physicians are already doing,” Dr. Karandeep Singh, study author.

See the source image

When reached for comment, Epic Systems told us the researchers’ hypothetical scenario lacked “the required validation, analysis, and tuning that organizations need to do before deployment,” adding that the JAMA study’s findings differed from other research. 



Bottom line: Algorithms can augment healthcare, but the life-or-death nature of their use requires serious due diligence.

ASSESSMENT: Your thoughts are appreciated



%d bloggers like this: