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 analysis of medical records with a biomarker found in lung fluid, specifically the expression of the FABP4 gene, they were able to improve diagnostic accuracy.
In an observational study involving adults with severe illness, this combined method correctly identified infections 96% of the time. It also outperformed intensive care unit clinicians in distinguishing between infectious and non-infectious causes of respiratory failure. The researchers estimated that if this model had been available at the time of patient admission, inappropriate antibiotic use could have been reduced by over 80%.
“We’ve devised a method that gives results much faster than a culture, and it could be easy to implement in the clinic,” said Chaz Langelier, M.D., Ph.D., associate professor of Medicine and senior author of the study published December 16 in Nature Communications. “We’re confident that it could lead to faster diagnosis and curtail the unnecessary use of antibiotics.”
The team discovered that lower expression levels of FABP4—a gene known for tempering inflammation—can help diagnose infection since infected lung cells express less FABP4 than healthy ones.
The study analyzed data from two groups: one before the COVID-19 pandemic with mostly bacterial infections, and another during the pandemic with primarily viral infections such as COVID-19. Each diagnostic tool—the biomarker or AI alone—was correct about 80% of the time. When combined, their accuracy improved significantly.
To further assess performance, researchers compared AI-driven analysis using GPT4 on a privacy-protecting UCSF platform against three physicians specializing in internal medicine and infectious diseases. Both approaches achieved similar accuracy rates; however, AI relied more heavily on radiology reports while physicians focused on clinical notes.
“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 co-first author. “It shows how AI can complement the work physicians do.”
The research team has published their AI prompts and encouraged other physicians to try them on secure platforms compliant with health privacy regulations.
“Using this is unbelievably simple, you don’t have to be a bioinformatician,” said Hoang Van Phan, Ph.D., co-first author.
Researchers are currently validating this model as a clinical test and plan to expand its application to sepsis diagnosis next.
Other authors from UCSF 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.
Funding for this research came from the National Institutes of Health (R01AI185511; NHLBI R35HL140026) and Chan Zuckerberg Biohub. The authors reported no financial or personal conflicts of interest.
