[ad_1]
Scientists from the University of British Columbia and BC Cancer have developed a new AI that can accurately predict how long a cancer patient will live, just by reading a doctor’s notes.
According to their findings, published this week in peer-reviewed journal JAMA Network Open, the AI model was able to forecast whether patients would survive another six months, 36 months or 60 months with more than 80 per cent accuracy.
After following up on the patients at regular intervals after the initial test, the robot was correct in predicting when people had less than six months to live 86 per cent of the time, when patients would die within 36 months 84 per cent of the time, and when patients had less than 60 months to live 84 per cent of the time.
The AI uses a technology called natural-language processing, the same tech powering viral AI chatbots like ChatGPT. This was used to analyze oncologists’ notes after a patient’s initial consultation visit, the first step following a cancer diagnosis. No other information was provided to the AI.
Over a five-year period, the researchers put their AI to work scanning through the notes of 47,625 patients suffering from the disease in six BC Cancer locations. The model was able to accurately provide expected lifespan, despite myriad different types and stages of cancer.
The researchers have not compared their model’s accuracy with that of human doctors, said John-Jose Nunez, lead researcher on the study and a clinical research fellow with the UBC Mood Disorders Centre and BC Cancer, told the Star. That will be one of the team’s next projects.
Determining a patient’s survival expectation is a notoriously difficult task. Studies have shown some cancer specialists are correct merely 20 to 33 per cent of the time when dealing with long-term prognoses, with almost half the incorrect predictions being more than six months off. Other studies found doctors had a roughly 74 per cent accuracy rate when predicting cancer survival times within a year.
“Sometimes oncologists will overestimate how long someone has left. This can be related to just feeling that emotional connection with the patient, wanting them to do well,” said Nunez who noted that he is not an oncologist. “Sometimes they also underestimate, they might know that a particular cancer type is a pretty difficult one” and let that sway their prediction, Nunez continued.
Having a robot do the work instead could cut out human bias and potentially come to a more objective conclusion, Nunez said. He added that AI is generally far better than humans at processing large amounts of data quickly and “make a prediction based off a ton of little moving parts that we as humans may not be able to.”
Nunez says his AI could give oncologists an important tool, helping them to understand when more treatment is necessary and when to prepare for a patient’s end of life.
“I just really want to emphasize, I think models like this are going to help all spectrums, all ends of cancer care,” Nunez said.
There are still questions to be answered: for example, Nunez said it’s unclear exactly how the AI is making its predictions. In the future, his lab will study which specific words the language model examines to come to its result. He also wants to expand the study to cancer patients across Canada, not just B.C.
“AI is actually already been being incorporated in health care,” said Nunez, adding that he uses AI to dictate his notes at the hospital. “It’s going to be a gradual process over decades of more and more artificial intelligence being used in health care.
“But to some extent, the future is already here.”
Anyone can read Conversations, but to contribute, you should be a registered Torstar account holder. If you do not yet have a Torstar account, you can create one now (it is free)
does not endorse these opinions.
[ad_2]
You can read more of the news on source
