Roupen Odabashian, Hematology/Oncology Fellow at Karmanos Cancer Institute, and Podcast Host at OncoDaily, shared a post on LinkedIn:
”Sometimes AI can cause more harm than benefit to patients.
I was in the emergency department recently seeing a patient with diverticulitis, which is inflammation in part of the bowel.
This patient had been on prednisone two weeks earlier. According to the transfer documents, he was started on prednisone for arthritis, though I was not sure of the exact etiology of his arthritis.
While managing his diverticulitis and comorbidities, the patient’s daughter and wife asked me whether prednisone was the cause. I paused and explained that prednisone can increase the risk of diverticulitis, but it is very difficult to say it directly caused the episode. Diverticulitis usually develops from chronic pouches in the bowel that become inflamed when food gets stuck in them.
Then something striking happened. The daughter showed me a screenshot taken from her sister’s phone. In the image, a Google search result—powered by AI, answered “yes, prednisone can cause diverticulitis.”
The problem is clear: patients are no longer checking sources. They are simply trusting whatever AI outputs. This can be dangerous in medicine.
Without context, AI can mislead. In this example, the first summary answer said “yes,” but scrolling further showed mixed results depending on the source.
Prednisone can increase the risk of diverticulitis, but no physician can say this patient developed diverticulitis because of prednisone. Don’t believe everything you read! Context is still important.”
More posts featuring Roupen Odabashian.
