Artificial intelligence (AI) has been making significant strides in various medical applications, including reading medical imaging and passing doctors’ licensing exams.
Now, a new AI tool developed by a team at NYU Grossman School of Medicine has demonstrated the ability to read physicians’ notes and accurately predict patients’ risk of death, hospital readmission, and other important outcomes for their care.
The software, known as NYUTron, is currently being used in affiliated hospitals throughout New York and aims to become a standard part of healthcare. The predictive value of the AI tool was assessed in a study published in the journal Nature.
Lead author Eric Oermann, a neurosurgeon and computer scientist at NYU, explained that while non-AI predictive models have existed in medicine for some time, they were not widely used due to the cumbersome reorganization and formatting of the required data.
The team’s approach was to leverage the wealth of information contained within physicians’ notes, which are consistently written and provide valuable insights.
NYUTron was trained on millions of clinical notes from the health records of 387,000 patients who received care at NYU Langone hospitals between January 2011 and May 2020. These notes encompassed various types of records written by doctors, such as patient progress notes, radiology reports, and discharge instructions, resulting in a vast corpus of 4.1 billion words.
One of the significant challenges for the software was interpreting the natural language used by physicians, which can vary greatly among individuals, including the abbreviations they choose.
Researchers evaluated the accuracy of the software’s predictions by analyzing historical records.
They also tested the tool in live environments by training it on records from one hospital and assessing its performance in a different hospital with diverse patient demographics.Overall, NYUTron successfully identified 95% of patients who died in the hospital before discharge and accurately predicted 80% of patients who would be readmitted within 30 days.
?