Detecting Delirium

Newswise — Delirium is one of the great hidden dangers for seniors seeking medical assistance in a hospital emergency room.

Unable to accurately communicate what’s wrong with them, patients who are delirious have an increased risk of numerous health complications, including longer hospital stays and greater likelihood of death. The problem is not uncommon—studies suggest as many as two-thirds of emergency room patients over the age of 65 experience delirium at some point during their stay.

But properly diagnosing it can be difficult for health care professionals, especially on busy days when they have lots of patients to attend do in the ER. Delirium symptoms can mirror dementia, and can come and go during the day, so nurses and doctors have to test for it often. As a result, it’s underdiagnosed and often untreated.

But researchers at the University of Iowa have developed a new algorithm that could help health care professionals more easily identify patients who are at risk of delirium. Nick Street, professor of business analytics at the Tippie College of Business, says the algorithm isn’t intended to positively identify whether a patient has delirium, but can quickly point out those who are at greater risk of experiencing it.

“It can tell health care professionals if a person should be tested immediately, be kept an eye on for 24 hours, or if this person is OK,” said Street. “It’s not intended to be a replacement for a doctor or a nurse, but it can help evaluate and provide more information to help those professionals make a decision.”

Since it’s a machine learning algorithm, it uses the data that’s been input over time and is able to improve its own accuracy as new data is entered.

Street and his research team built the algorithm by analyzing the health records of more than 28,000 patients older than 65 who were admitted to the emergency department at the University of Iowa Hospitals and Clinics between 2014 and 2020. They looked at 651 factors from each patient, ranging from such demographic information as age and body mass index, to medical factors like temperature, respiratory rate, chronic illnesses, and what types of medications the patient is taking.

They then entered whether the patient had a true positive test for delirium within 72 hours of admission so the algorithm could determine which factors were most telling for delirium in which circumstances.

Brianna Mueller, a doctoral student in the Tippie College of Business’ business analytics department who is on the research team, said those factors differed based on each patient’s circumstance. But the factors that seemed to have the greatest connection to delirium were whether the patient had dementia, dysphagia, or brain injuries, urinary tract infections, epilepsy, age, or had suffered a stroke.

Street said the team will continue refining the algorithm and use it to evaluate the effectiveness of different interventions that would reduce the risk of delirium. They also plan to modify it for use in intensive care units.

Street’s and Mueller’s study, “Evaluating the performance of machine learning methods for risk estimation of delirium in patients hospitalized from the emergency department,” was published recently in the journal Acta Psychiatrica Scandinavica. It was co-authored by Ryan Carnahan of the University of Iowa College of Public Health and Sangil Lee of the University of Iowa Carver College of Medicine.