Can An Algorithm Predict Hospital Deaths?

Whether we realize it or not, most of us enter into some company’s algorithm every day. This includes online shopping choices, Google ads, and email alerts. Algorithms use information to try and establish patterns and make predictions. Banner Health, a hospital group based in Arizona, tried to create an algorithm to predict infections in patients. It turns out the program did not work as intended; however, it did have an effect on predicting which patients were most seriously ill.

Infections are responsible for up to half of all deaths in U.S. hospitals. Sepsis can be caused by bacterial, viral, or fungal infections. It can develop inside the body from pneumonia; abdominal, blood, or kidney infection; or it can be spread from another infected patient. Sepsis can progress to severe sepsis and septic shock, eventually leading to death.

Sepsis symptoms may include fever, increased heart and breathing rates, difficulty breathing, pain, and a change in mental status. Eventually, the body begins to shut down, impairing blood flow to the vital organs, leading to tissue death and organ failure. The goal is to treat sepsis during the earlier stages before it becomes more serious. One problem for hospitals is that the symptoms of sepsis can look like other illnesses.

Dr. Hargobind Khurana worked with Banner Health in Phoenix, Arizona to develop a computer program that could warn hospital employees when patients might be at an increased risk of sepsis. After a few years of use, it did not accurately predict sepsis as intended; however, it was able to identify sicker than average patients.

“It’s hard to create a good alert,” said Dr. Khurana. “And it’s hard to get buy-in from doctors and nurses because it’s ‘just another thing’ to do. How do we keep that balance of not just expecting them to do more work, but how do we make sure the patient is taken care of?”

Many doctors and nurses are wary of these automatic alerts. According to a report by the Joint Commission, a hospital accreditation group, hundreds of alerts can sound off per patient every day, giving healthcare workers alarm fatigue. The doctors and nurses may have to manually turn off the alarm, evaluate the patient, and document the incident.

After five years of the sepsis alert system being used on more than 300,000 patients in two dozen hospitals, Dr. Khurana and his team evaluated the data. They found that only about a quarter of the alerts accurately found sepsis. However, as a side effect of sepsis mimicking other illnesses, the alerts did flag patients who were, on average, sicker than others. The program identified almost 90% of all the patients who died in the hospital.

If you or your loved one has been injured because of a medical error, the Gilman & Bedigian team is here to help. We are fully equipped to handle the complex process of your medical malpractice claim. Our staff, including a physician and attorneys with decades of litigation experience, will focus on getting you compensation, so you can focus on healing and moving forward.