Artificial Intelligence (AI) has demonstrated superior capabilities in interpreting medical images and has even proven capable of passing medical licensing examinations.
A new AI tool has recently shown its ability to read doctors’ notes and accurately predict patients’ risks of mortality, readmission, and other critical outcomes in their care.
This new AI software accurately estimates 79% of patients’ actual hospital stay duration. (Image source: dailymail.co.uk).
According to a recent research report published in the journal Nature, the new AI software called NYUTron was developed by a team of scientists from NYU Grossman School of Medicine and is currently being applied in hospitals affiliated with this university across New York (USA).
The lead author of the study is Eric Oermann, a computer scientist and neurologist at NYU. According to Oermann, while predictive models not based on AI technology have long existed in medicine, they have rarely been implemented in practice due to the cumbersome organization and formatting of required data. However, he noted: “Doctors often take notes on what they observe in the clinic, what they discuss with patients. So, our direction was whether we could start using these medical notes as our data source and then build predictive models based on that.”
The NYUTron large language model was built on millions of clinical notes from the medical records of 387,000 individuals cared for in hospitals in New York from January 2011 to May 2020. This data encompasses everything written by doctors, such as notes on patients’ disease progression, X-ray results, or discharge papers… All of this combined forms a text corpus of 4.1 billion words.
One of the main challenges for this software is interpreting natural language based on what doctors have written. The way doctors express themselves varies significantly, even in their habitual abbreviations.
By reviewing the records of what has occurred, researchers were able to calculate NYUTron’s correct prediction frequency. Overall, NYUTron accurately identified 95% of patients who died in the hospital before being discharged and 80% of patients who would be readmitted within 30 days. NYUTron also accurately estimated 79% of patients’ actual hospital stay duration, and 87% of cases where patients’ insurance claims were denied… This software demonstrated superior predictive ability compared to most doctors and the non-AI computer models currently in use.
Oermann believes that AI will never replace the relationship between doctors and patients. However, he asserts that AI can help “provide additional information to doctors seamlessly at the time of patient care, enabling professionals to make more informed decisions”.
