Roupen Odabashian, Oncologist at Abbotsford Regional Hospital and Cancer Centre, shared on LinkedIn:
“Yesterday I spent an hour with a man in his eighties who has one cancer spot, in his brain, in the speech area. He can barely get words out now.
He wants to stay alive because his wife is sick and he is the one who takes care of her. He does not want chemotherapy because he remembers what it did to him last time. His family wants everything done.
Tell me which model solves that.
I read the JAMA Perspective arguing that autonomous AI will exceed AI aided physicians. The numbers in it are real and I am not here to dispute them. But two things are worth saying plainly.
It is a Perspective, not a trial. And it went viral in large part because of whose name is on it.
Put Khosla on almost anything about AI replacing doctors and it travels.
Here is the part I want to argue with instead.
The paper measures medicine as diagnosis and decision making. That is maybe ten percent of my day. The other ninety percent is not in the literature at all.
Another patient this week, heavy smoking history, previous pancreatic cancer removed surgically, now new spots in the lungs. Could be lung. Could be pancreatic. We ran the full genetic workup and we still do not know.
He and I have to decide anyway. There is no paper for that.
A third has had bladder cancer and breast cancer. Every trial I would want for her excludes her, because a second primary disqualifies you. The evidence base was not built with her in it.
So when we say AI selects ‘the most appropriate treatment,’ I want to ask who defined appropriate. It was defined by prior literature and prior trials. Trials that excluded my patient. Trials with their own biases baked in.
Sometimes the right move is a less appropriate treatment, because it is the one that fits the life the person actually wants to live.
The patient decides that, not the guideline.
And I should be clear: if AI takes this part of my job, I am fine with it. I am not defending my job.
It is heavy to deliver prognosis all day. I love this work and I am already moving toward machine learning because I want to work on curing cancer rather than only treating it.
What I actually wonder about is behavior, not capability.
When Airbnb started, the idea that you would let a stranger sleep in your house sounded insane. It changed because people got paid. The money moved the behavior.
Health has no equivalent lever. Patients already come to me having asked ChatGPT, and I encourage that. Then they sit down and we talk anyway. Because what they came for was not the answer.
It was to be helped to decide, by someone who will still be there when the decision goes badly.
I do not know what changes that. Maybe something does. But the paper is about accuracy, and the thing standing between these tools and my exam room is not accuracy.”
Title: Will Autonomous AI Exceed AI-Aided Physicians as the Best Medical Care?
Authors: Ezekiel J. Emanuel, Abe Baker-Butler, Neal Khosla, Vinod Khosla.
Read The Full Article.
Other articles about AI in Oncology on OncoDaily.