Artificial intelligence (AI) has proven useful in reading medical images and has even shown that it can pass the licensing exam for doctors. now a new AI tool has demonstrated the ability to read medical notes and accurately anticipate the risk of death of patients, their readmission to the hospital, and other outcomes important to their care.
(See: The company that has become a ‘superpower’ thanks to AI)
Designed by a team at the NYU Grossman School of Medicinethe program is currently used in all hospitals affiliated with this university in New York (United States), with the hope that it will become standard in health.
The study on its predictive value was published June 7 in the journal Nature. Lead author Eric Oermann, an NYU neurosurgeon and computer scientist, told the AFP that while the predictive models without AI have been around medicine for several yearshave been rarely used in practice because the data you need requires cumbersome rearranging and formatting.
(See: The West seeks common standards for Artificial Intelligence)
However, “One thing that is common in medicine everywhere is that doctors write notes about what they see in the clinic, what they have discussed with patients”said. “So our basic insight was: Can we start with medical notes as a data source and then build predictive models on top of them?”he added.
The huge language model called NYUTron was trained with millions of medical notes taken from medical records of 387,000 people who received care at NYU Langone hospitals between January 2011 and May 2020.
AI and health
iStock / Bogota City Hall
These records included those written by doctors, patient progress notes, radiology reports, and discharge instructions, totaling 4.1 billion words. One of the key challenges for the program was to interpret the natural language that doctors writewhich varies widely between individuals, even because of the abbreviations each uses.
By looking at records of what they got, the researchers were able to calculate how often the program’s predictions were accurate. Besides tested the tool in live environmentsand they trained it on records from a hospital in Manhattan and then saw how it performed at one in Brooklyn, with different patient demographics.
(See: Why it would be good for AI to replace humans in jobs)
In general, NYUTron identified a staggering 95% of people who died in hospital before being discharged, and 80% of patients who would be readmitted in 30 days. The tool outperformed most of the doctors’ predictions, and also outperformed current models that do not use AI.
of people who died in the hospital before being discharged were identified by NYUTron.
However, to the team’s surprise, “the most experienced of the doctors, who is in fact very famous, performed superhumanly, better than the model”Oermann said. In addition, he indicated that “The sweet spot between technology and medicine is not that medicine must always necessarily deliver superhuman results, but rather that it really offers that starting point”.
NYUTron also correctly estimated the length of stay of 79% of patients89% of cases in which patients were denied coverage by their insurance and 89% of cases in which the patient’s primary illness was accompanied by additional conditions.
(See: ‘Artificial intelligence could lead humanity to extinction’)
AI will never be a substitute for the doctor-patient relationship, says Oermann. Instead, it will help “provide more information to physicians at the time of care so they can make more informed decisions”.
AFP
