"AI Doctor" Better at Predicting Patient Outcomes, Including Death

2023-06-08

Artificial intelligence has proven itself to be useful in medical imaging interpretation, even demonstrating its ability to pass medical licensing exams for doctors. Now, a new AI tool has shown its ability to read doctor's notes, accurately predict patient mortality risk, readmissions, and other important outcomes for patient care. Designed by a team at NYU Grossman School of Medicine, this software is currently being used in NYU affiliated hospitals and aims to become a standard component of healthcare. A study on its predictive value was published in the journal Nature on June 7th. Dr. Eric Oermann, a neurosurgeon and computer scientist at NYU and the lead author of the study, told AFP that while there are non-AI predictive models in the medical field, they are rarely used in practice due to the need for tedious data reorganization and formatting. However, "a common thing in the medical field is that doctors write down what they see and discuss with patients," he said. "So, our basic insight was, can we use medical notes as our data source and build predictive models on top of that?" This large-scale language model, named NYUTron, was trained using millions of clinical notes from 387,000 individuals' health records who received treatment at NYU Langone Hospital between January 2011 and May 2020. These records include any notes written by doctors, such as progress notes, radiology reports, and discharge instructions, forming a corpus of 4.1 billion words. One of the main challenges faced by the software is interpreting the natural language used by doctors, which varies greatly between individuals, including their choice of abbreviations. By reviewing the occurrences of the notes, researchers were able to calculate how accurate the software's predictions were. They also tested the tool in real-time environments, such as training it on records from a hospital in Manhattan and observing its performance in another hospital in Brooklyn with different patient demographics. Overall, NYUTron found that 95% of patients died in the hospital before discharge, and 80% of patients were readmitted within 30 days. Its predictive performance was better than most doctors and non-AI computer models currently in use. However, the team was surprised to find that "the most senior doctor, who is actually a very famous doctor, performed exceptionally well, even better than the model," said Oermann. "The key point of technology and medicine is not that it will necessarily provide results beyond human capabilities, but that it does improve the baseline level," Oermann said. NYUTron also accurately estimated the actual length of hospital stays for 79% of patients, insurance denials for 87% of patients, and the presence of other conditions accompanying the primary disease for 89% of patients. Oermann stated that AI can never replace the relationship between doctors and patients. Instead, AI will help provide more seamless information at medical points so that doctors can make wiser decisions. (Source: https://techxplore.com/news/2023-06-ai-doctor-patient-outcomes-death.html)

RECENT AI NEWS

RECENT AI TOOLS