Slashdot Mirror
88% Of Medical 'Second Opinions' Give A Different Diagnosis - And So Do Some AI (mayoclinic.org)
First, "A new study finds that nearly 9 in 10 people who go for a second opinion after seeing a doctor are likely to leave with a refined or new diagnosis from what they were first told," according to an article shared by Slashdot reader schwit1:
Researchers at the Mayo Clinic examined 286 patient records of individuals who had decided to consult a second opinion, hoping to determine whether being referred to a second specialist impacted one's likelihood of receiving an accurate diagnosis. The study, conducted using records of patients referred to the Mayo Clinic's General Internal Medicine Division over a two-year period, ultimately found that when consulting a second opinion, the physician only confirmed the original diagnosis 12 percent of the time. Among those with updated diagnoses, 66% received a refined or redefined diagnosis, while 21% were diagnosed with something completely different than what their first physician concluded.
But in a related story, Slashdot reader sciencehabit writes that four machine-learning algorithms all performed better than currently-used algorithm of the American College of Cardiology, according to newly-published research, which concludes that "machine-learning significantly improves accuracy of cardiovascular risk prediction, increasing the number of patients identified who could benefit from preventive treatment, while avoiding unnecessary treatment of others."
"I can't stress enough how important it is," one Stanford vascular surgeon told Science magazine, "and how much I really hope that doctors start to embrace the use of artificial intelligence to assist us in care of patients."
But in a related story, Slashdot reader sciencehabit writes that four machine-learning algorithms all performed better than currently-used algorithm of the American College of Cardiology, according to newly-published research, which concludes that "machine-learning significantly improves accuracy of cardiovascular risk prediction, increasing the number of patients identified who could benefit from preventive treatment, while avoiding unnecessary treatment of others."
"I can't stress enough how important it is," one Stanford vascular surgeon told Science magazine, "and how much I really hope that doctors start to embrace the use of artificial intelligence to assist us in care of patients."
Pretty much this. Cardiovascular risk is one of the best studied disease states known. Which is probably why they studied it. Even then, the 'AI' algorithms only improved risk stratification around 5% - nothing to sneeze at but hardly earth shattering.
OK, now, for extra credit lets risk stratify middle age hypertensive diabetics who are depressed.
Like the typical 'real world' patient. I'd just love some help here but the underlying data just doesn't support it. Which is kinda surprising since we've been studying these folks for a while. Simple medical problems are simple. Typical medical problems are not.
Faster! Faster! Faster would be better!
I was thinking more of the opposite. A doctor who is asked for a second opinion knows the patient does not want the same diagnosis. He knows that the patient is shopping around for the "best" diagnosis. He knows that the only way he is likely to be able to start treatment is to give a different diagnosis. I would say that medical diagnosis are complicated things that patients are not likely to be able to gauge correctly. If a patient is better at diagnosis than a doctor, and better able to tell if the correct one has been given, why even use doctors?
Troll is not a replacement for I disagree.