Using AI to detect the risk of delirium in patients

Researchers at Johns Hopkins University said they are developing ways to use artificial intelligence to detect warning signs of delirium in hospital patients.

The AI algorithms have proven to be successful in early tests, according to findings published in the journal Anesthesiology.

The early results carry intriguing implications in improving treatment and detecting patients who could develop delirium, said Robert Stevens, associate professor of anesthesiology and critical care medicine at the Johns Hopkins University School of Medicine and senior author of the study.

“Being able to differentiate between patients at low and high risk of delirium is incredibly important in the ICU because it enables us to devote more resources toward interventions in the high-risk population,” Stevens said in a Johns Hopkins news release.

Patients with delirium are at greater risk of death, longer hospital stays, or a long-term reduction in cognitive function, researchers have found.

As many as 80% of older patients treated in intensive care units can develop delirium, and roughly a quarter of all older adults admitted to the hospital have been found to have delirium, according to a Nature study published in 2020.

Johns Hopkins engineering students helped develop the AI models used to predict delirium risk, according to the university’s release.

Researchers developed a “static” AI model that predicted risk from a snapshot of patient data just after admission, the university said. They also developed a “dynamic” model that examined patient information, such as pulse and blood pressure, over a period of days to assess the risk of delirium over the next 12 hours. The models were tested by examining more than 100,000 ICU stays at a Boston hospital, Johns Hopkins said.

Both models proved successful in projecting patients that would get delirium. The dynamic model identified patients at risk for delirium up to 90% of the time, while the static model successfully projected patients that would develop delirium 78.5% of the time.

Stevens says he is testing the algorithms on historical patient data at Johns Hopkins Medicine ICUs, and he plans to design a trial to test the models on patients who are admitted to an ICU. He envisions using AI models to try to predict risks of heart failure and embolisms.

“For a lot of these physiological transitions, we think that there are early warning signs that may not be obvious to a clinician but can be picked up on using the kinds of artificial intelligence-supported pattern analysis that we used here,” Stevens said in the news release.

Researchers are finding promising results in the use of AI. Mayo Clinic researchers have used AI-enabled echocardiograms to identify patients at greater risk of stroke.

Despite a great deal of hype, the widespread use of AI in medicine is just beginning, but healthcare leaders nonetheless predict that artificial intelligence has the potential to transform treatment and patient care.

Michael Howell, Google’s chief clinical officer and deputy chief health officer, said during a panel in September, “AI will do things we didn’t think were possible.”

Read more:

Artificial intelligence, healthcare, and questions of legal liability