EVANSTON, Ill.  –  Researchers at Endeavor Health and Northwestern University have developed a new computer program that can help doctors identify Acute Respiratory Distress Syndrome (ARDS), a debilitating and often fatal lung condition that’s common in Intensive Care patients and estimated to impact up to 190,000 Americans per year. Their work was recently published in the journal, Nature Communications.

While common, the condition is difficult to diagnose because doctors have to piece together many types of disconnected information, such as lab results, chest X-rays, other doctors’ notes and heart tests – finding clues of the condition that may go unnoticed. The new program sifts through that information automatically, and in testing, was highly accurate in identifying the condition.

ARDS causes fluid to build up in the lungs’ alveoli (air sacs), leading to dangerously low oxygen levels in the blood stream. It usually occurs after an injury or infection of the lungs, such as COVID-19, and has a mortality rate of more than 40 percent. ARDS is more common in the critically ill and Intensive Care Unit (ICU) patients.

“As an ICU doctor, you’re busy taking care of patients, and it’s hard to comb through all of the data required to make that diagnosis,” said Dr. Curtis Weiss, an Endeavor Health pulmonologist and Co-Director of Critical Care Medicine, who co-authored the paper and is principal investigator on the grant that funded the work.  “That’s where AI and machine learning come in. We think this program could help fill that gap, creating an additional safeguard to help doctors and patients here in Chicagoland and well beyond.”

To develop the program, Dr. Weiss partnered with Northwestern University engineering researchers, Félix Morales, M.S. and Luís Nunes Amaral, Ph.D. who specialize in data science and smart tools for healthcare. Their new algorithm, built on clinical guidelines used by ICU doctors, looks back at patient records to automatically identify those who had ARDS while they were on ventilators.

“If ARDS isn’t caught right away, patients may not receive the life-saving treatment they need as quickly,” said Mr. Morales, a research specialist in NU’s department of Engineering Sciences and Applied Mathematics and lead author on the paper. “An automatic tool like this could catch more cases, helping doctors give the right treatment sooner.”

The team tested their program on medical data from a hospital outside the Endeavor Health and Northwestern systems, and found it correctly diagnosed 93.5 percent of true ARDS cases and only made mistakes (false alarms) about 17 percent of the time. This is an improvement compared to estimates of how often doctors recognize the condition in the ICU.

The algorithm will now be piloted at Endeavor Health. Dr. Amaral, a Professor of Engineering Sciences and Applied Mathematics at NU, said the success of this program speaks to the promise of AI as a tool to help doctors make diagnoses.

“Even for conditions other than ARDS, this is a good example of AI’s potential as a care tool and how it might lead to better and faster treatment,” said Dr. Amaral. “AI done carefully can give doctors superpowers, helping them spot critical conditions faster, more accurately and at a scale no human could manage alone."