New York City-based NYU Langone Health is using an artificial intelligence model that can predict the chances of a patient's death and readmission.
The new AI is a large language model that can "learn" from text without needing specifically formatted data, according to a June 7 system news release. New York City-based NYU Grossman School of Medicine researchers developed a model called NYUTron that used clinician notes from electronic health records to make assessments about patient health.
A study, published June 7 in Nature, trained NYUTron using millions of clinical notes from 336,000 men and women who received care at NYU Langone between January 2011 and May 2020. The notes included radiology reports, patient progress notes and discharge instructions. Language was not standardized among physicians, but the AI could interpret abbreviations unique to a particular writer, according to the release.
NYUTron identified 85 percent of those who died in the hospital, estimated 79 percent of patients' actual length of stay and successfully assessed the likelihood of additional conditions and complications, as well as the chances of an insurance denial.
"These results demonstrate that large language models make the development of 'smart hospitals' not only a possibility, but a reality," study senior author Eric Oermann, MD, said in the release. "Since NYUTron reads information taken directly from the electronic health record, its predictive models can be easily built and quickly implemented through the healthcare system."
Researchers said the AI model may speed up workflow and allow physicians more time to spend with patients.