Researchers from Endeavor Health and Northwestern University have developed an artificial intelligence tool designed to help doctors identify acute respiratory distress syndrome (ARDS), a severe and often fatal lung condition that is frequently underdiagnosed in critically ill patients. The AI system has demonstrated 93% accuracy in identifying historical ARDS cases using data already available in electronic health records, including lab results, imaging, and clinical notes. The tool is now preparing for real-time pilot testing in hospitalized patients at Endeavor Health. ARDS occurs when the lungs become severely inflamed, causing fluid to leak from blood vessels into the air sacs, impairing oxygen exchange. This leads to life-threatening breathing difficulties, often described as "drowning on dry land" because the lungs fill with fluid despite the absence of external water. The condition has a mortality rate as high as 46%, and survivors often face long-term complications such as lung scarring or cognitive deficits due to prolonged oxygen deprivation. ARDS can be triggered by various critical illnesses, including sepsis, pneumonia, and severe viral infections like COVID-19. During the pandemic, otherwise healthy young adults succumbed to ARDS, highlighting the condition’s unpredictable and deadly nature. Dr. Curtis Weiss, a pulmonologist at Endeavor Health and co-director of critical care medicine, has been working on improving ARDS recognition since 2018 and led the team that developed the AI tool. Unlike generative AI models such as ChatGPT, this system does not create new content or hallucinate data. Instead, it analyzes existing clinical information to detect patterns associated with ARDS. The goal is not to replace doctors but to serve as an automated alert system that flags potential cases, prompting clinicians to consider ARDS as a diagnosis. One of the main challenges in diagnosing ARDS is that its symptoms overlap with other conditions, such as congestive heart failure, which also causes fluid buildup in the lungs. However, the underlying causes differ significantly. In ARDS, the fluid results from lung inflammation, while in heart failure, it stems from the heart’s inability to pump effectively. The treatment for each condition varies: ventilator settings and patient positioning differ, and placing an ARDS patient on their stomach can improve oxygenation, whereas this position may worsen outcomes in heart failure patients. Weiss noted that ICU physicians face information overload, managing multiple critically ill patients daily, which can lead to missed diagnoses. The AI tool aims to overcome this by integrating disparate clinical data points—oxygen levels, chest X-rays, underlying conditions—into a cohesive assessment, helping doctors recognize ARDS earlier. Currently, the tool has been tested only on past cases. The next phase involves real-time use in active patient care to predict ARDS before a formal diagnosis is made. The team is willing to accept a higher rate of false positives to ensure no true ARDS cases are missed. “I’d rather treat a few patients who don’t have ARDS than fail to treat someone who does,” Weiss said. The focus remains on solving the critical problem of underdiagnosis and improving survival rates through early detection.