Researchers at the University of California, San Francisco (UCSF) have developed a new approach to diagnosing lung infections in critically ill patients. By combining generative artificial intelligence (AI) analysis of medical records with a biomarker found in lung fluid samples, the team was able to identify infections such as pneumonia with a high degree of accuracy.
The study involved analyzing data from two groups of critically ill adults: one group recruited before the COVID-19 pandemic, mostly with bacterial infections, and another group recruited during the pandemic, mainly with viral infections including COVID-19. The combination of AI and the biomarker resulted in correct diagnoses 96% of the time. This method also more accurately distinguished between infectious and non-infectious causes of respiratory failure than intensive care clinicians.
Chaz Langelier, M.D., Ph.D., associate professor of Medicine at UCSF and senior author of the study published on December 16 in Nature Communications, stated: “We’ve devised a method that gives results much faster than a culture, and it could be easy to implement in the clinic. We’re confident that it could lead to faster diagnosis and curtail the unnecessary use of antibiotics.”
A key component is the FABP4 gene biomarker developed by Langelier’s team in 2023. The gene is less expressed in infected lung cells compared to healthy ones, helping diagnose infection.
Each diagnostic method—using either just AI or just the FABP4 biomarker—was about 80% accurate when used alone. When combined, their performance improved significantly. The researchers noted that if this model had been available when patients were admitted, inappropriate antibiotic use could have been reduced by over 80%.
To evaluate how well their approach worked compared to human doctors, three physicians specializing in internal medicine and infectious diseases reviewed patient records alongside GPT-4 AI operating on a privacy-protecting platform developed at UCSF. Both approaches made about the same number of correct diagnoses; however, while physicians focused more on clinical notes, the AI placed greater emphasis on radiology reports from chest X-rays.
“It was almost showing a cultural difference, if you can say that about an AI,” said Natasha Spottiswoode, M.D., DPhil., assistant professor of Medicine and first author on the paper. “It shows how AI can complement the work physicians do.”
The research team has published their AI prompts for other physicians to try using HIPAA-compliant platforms. Hoang Van Phan, Ph.D., also a first author on the paper, added: “Using this is unbelievably simple, you don’t have to be a bioinformatician.”
Further validation is underway as they consider applying this diagnostic model to sepsis.
Additional authors include Emily Lydon, M.D., Carolyn Calfee, M.D., MAS; Victoria Chu, M.D., MPH; Adolfo Cuesta, M.D., Ph.D.; Alexander Kazberouk, M.D., MBA; Natalie Richmond, M.D.; and Padmini Deosthale MS—all affiliated with UCSF.
Funding for this work came from grants provided by the National Institutes of Health and support from Chan Zuckerberg Biohub.
No financial or personal conflicts of interest were reported by any authors.
