AI and a Hidden Lung Biomarker Could Catch Deadly Infections 96% of the Time

Researchers at UC San Francisco have developed a groundbreaking method that combines artificial intelligence with a lung biomarker to diagnose deadly pneumonia and other lung infections with 96% accuracy. This innovative approach could revolutionize how doctors identify these infections in critically ill patients while dramatically reducing unnecessary antibiotic use by more than 80%.

How Does This New Diagnostic Method Work?

The breakthrough combines two powerful tools: generative artificial intelligence analysis of medical records paired with a biomarker found in lung fluid samples. The biomarker focuses on a gene called FABP4, which reduces inflammation and is expressed less in infected lung cells compared to healthy ones. When researchers tested each method separately, both the FABP4 biomarker and artificial intelligence were correct about 80% of the time, but together they achieved remarkable 96% accuracy.

The study examined data from two groups of critically ill patients: 98 recruited before the COVID-19 pandemic (mostly with bacterial infections) and 59 recruited during the pandemic (mostly with viral infections, including COVID-19). The artificial intelligence component used GPT4 on a privacy-protecting platform developed at UC San Francisco.

What Makes This Approach Different From Current Methods?

Traditional diagnosis of lung infections like pneumonia is notoriously difficult, yet these infections rank among the world's top killers. The new method offers several key advantages over current diagnostic approaches:

• Speed: Results come much faster than traditional culture methods, which can take days to provide answers
• Accuracy: The 96% diagnostic accuracy significantly outperforms individual methods and matches physician expertise
• Antibiotic Stewardship: The model was much more judicious in diagnosing pneumonia compared to doctors who prescribed antibiotics for most patients in the study
• Clinical Implementation: The approach could be easy to implement in hospital settings without requiring specialized bioinformatics expertise

"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, MD, PhD, an associate professor of Medicine and senior author of the study published in Nature Communications. "We're confident that it could lead to faster diagnosis and curtail the unnecessary use of antibiotics."

Could This Change How Doctors Diagnose Lung Infections?

When researchers compared the artificial intelligence analysis to how three physicians specializing in internal medicine and infectious diseases analyzed the same medical records, both achieved similar diagnostic accuracy. However, the artificial intelligence showed interesting differences in approach—it gave more weight to radiology reports about chest X-rays, while physicians focused more on clinical notes.

"It was almost showing a cultural difference, if you can say that about an AI," said Natasha Spottiswoode, MD, DPhil, assistant professor of Medicine and one of the first authors. "It shows how AI can complement the work physicians do."

The research team has made their artificial intelligence prompts publicly available in their published paper, encouraging physicians to try them on their own HIPAA-compliant artificial intelligence platforms. "Using this is unbelievably simple, you don't have to be a bioinformatician," explained Hoang Van Phan, PhD, a bioinformatician and first author of the study.

The team is currently validating the model as a clinical test and plans to tackle sepsis next—the most common cause of hospital death and another condition that's famously difficult to diagnose accurately. This research was supported by the National Institutes of Health and the Chan Zuckerberg Biohub.