Artificial intelligence (AI) has been used increasingly in almost every aspect of our lives. The medical and healthcare sector is no exception. Doctors have recently adopted AI to improve areas like medical image reading and administrative tasks. While these are impressive, AI is proving even more helpful as it takes on tasks like reading physicians’ notes and accurately predicting outcomes to various patient scenarios. The AI assisting the healthcare sector is called NYUTron.

NYUTron is a Large Language Model (LLM), a form of AI that learns from text without relying on specially formatted data. Instead, it uses specialized computer algorithms to predict outcomes based on the patterns it has been taught to recognize. Therefore, LLMs need a lot of data to increase the accuracy of the anticipated results. NYUTron was trained on millions of clinical notes from the records of over 387,000 patients who received care from NYU Langone hospitals between January 2011 and May 2020.

News about NYUTron gained attention after a study on its predictive value was published in the journal Nature. The lead author, Eric Oermann, a neurosurgeon and assistant professor, spoke about the importance of predictive value in eradicating the stress of reorganization and formatting, as is the practice with the non-AI predictive models used over the years. The team is building on medical notes as a data source, including patient progress notes, discharge instructions, and radiology reports. Everything doctors have seen and discussed with patients over that period resulted in a 4.1-billion-word language cloud.
Designed by a team at NYU Grossman School of Medicine, NYUTron is currently used at various hospitals affiliated with the university across New York. The team hopes the scope of use will expand rapidly to make AI a standard part of health care. They emphasized that the tool only exists to complement the work done by physicians, not to act as a substitute for them.

The team tested NYUTron against humans, and while it was better at predicting outcomes than most physicians, it still fell against the most experienced of them. The most senior physician, who is very famous and known to produce superhuman performance, did better than the model, to the surprise of everyone. Oermann emphasized this result to promote the fact that technology and medicine may not always produce superhuman results but will bring up the baseline.

Currently, the model is capable of predicting 79 percent of patients’ length of stay at hospitals, 80 percent of patients who were readmitted, 85 percent of those who did in the hospital, 87 percent of when patients were denied coverage by insurance, and 89 percent of cases where additional conditions accompany a primary disease. The expectation is for the numbers to improve as the model is fed more data from various subjects and demographics.

The creation of NYUTron has raised the bar on how AI can be used in medicine and healthcare. The model’s potential includes the ability to predict the risk of infection, identify the proper medication to get, extract billing codes, and much more. It will free up physicians to spend more time with patients and work more efficiently