Researchers at Baltimore-based Johns Hopkins University have created an artificial intelligence tool to predict the risk of a patient developing "ICU delirium," a condition in which patients admitted to the intensive care unit develop bouts of confusion, inattention, paranoia and, in some cases, hallucinations.
The team developed two AI models trained on a dataset of 200,000 ICU stays, according to a Jan. 26 Johns Hopkins news release. The first static model predicted ICU delirium risk based on factors such as age, the severity of illness, current medications and other diagnoses.
The second dynamic model predicts delirium risk over the coming 12 hours through monitoring a patient's pulse, blood pressure and temperature readings.
When tested on data sets from a Boston hospital covering more than 100,000 ICU visits, the static model accurately predicted delirium 78.5 percent of the time, and the dynamic model was accurate 90 percent of the time, the release s