Published on
Updated
Reading 1 min.
Researchers at the Langone School of Medicine in New York have developed software capable of accurately anticipating the risk of death, readmission to hospital and other medical complications encountered by patients who have been hospitalized. The results of its effectiveness have just been revealed in the journal Nature.
Called NYUTron, this predictive model was developed from millions of medical observations from the files of more than 300,000 patients passed through hospitals attached to New York University Langone, between January 2011 and May 2020.
Several billion words analyzed
This included written reports from doctors, notes on the evolution of patients’ condition, medical imaging examinations, but also the recommendations given to patients before they were discharged from the hospital. In total, the software had to analyze 4.1 billion words. The researchers wanted to know if artificial intelligence was able to predict the evolution of the state of health of patients based on a database of medical notes written by doctors.
A tool with formidable precision…
NYUTron has been tested in real conditions. The authors of the study asked him to analyze reports from a hospital in Manhattan and compare them to those from a hospital in Brooklyn (patient populations are different in these two hospitals). The tool was very effective since it identified 95% of patients who died during their hospitalization and 80% of patients who were readmitted less than a month after their discharge. It also predicted the duration of hospitalization for 79% of patients, the refusal of reimbursement of care by their insurance for 87% of patients and identified 89% of cases in which the patient suffered from additional pathologies.
… but who doesn’t do better than a very experienced doctor
On the other hand, the tool did no better than a very experienced doctor since the latter gave predictions “even better than the software”, recognized Eric Oermann, co-author of the study.
Even if artificial intelligence will never be able to do the work of a doctor, it could in the near future guide practitioners in their decisions by providing them with a wealth of information that has been sorted and analyzed beforehand.