Defeating Infectious Agents — 14

All these trials are using natural phages, but the next phase is getting started with multiphage cocktails, phage genome engineering, and synthetic phages. While I have emphasized the use of phages to overcome microbial resistance, there are many potential applications for noninfectious diseases. Examples include phages to treat Klebsiella pneumoniae or Escherichia coli in inflammatory bowel disease, Enterococcus faecalis in alcoholic hepatitis, to manipulate the gut microbiome to improve immunotherapy for colon cancer, and as a vector to deliver drugs or vaccines. The successful use of phage delivery of CRISPR base editors to alter the sequence of Escherichia coli and Klebsiella pneumoniae in the mouse gut microbiome bodes well for both a future application of phage and another way of overriding antimicrobial resistance.

Sometimes sepsis, an extreme reaction to some infection, kills due to antimicrobial resistance, but mostly treatment failures are from missing the diagnosis or suboptimal medical care. With nearly 50 million cases and 11 million deaths each year around the world, and 350,000 deaths in the United States, this life-threatening, organ dysfunctional syndrome is the third-leading cause of death in US hospitals. Can technology help here? An AI tool was widely adopted throughout the United States to predict the risk of sepsis, but subsequent assessment showed its performance was poor. This was ascribed to the phenomenon of dataset shift, whereby a mismatch occurs from the dataset a model was developed from compared with the one on which it was deployed. Since that failure was diagnosed in 2021, there have been improvements in AI detection of sepsis that significantly reduced time for starting antibiotic therapy and were associated with an 18 percent reduction in deaths when physicians responded to alerts promptly. In 2024, the FDA approved the first AI model that predicts the risk of sepsis by Presnosis, a sepsis immunoscore that uses twenty-two clinical and laboratory parameters for likelihood of developing this condition within twenty-four hours.

Accurate and timely detection of hospital-acquired infections from central venous access or urinary catheters was fostered by two large language model AIs (GPT-4 and Mistral). Biomarkers and whole-blood gene expression that help make the diagnosis or partition high-risk patients for developing sepsis are also being validated, and they would add to the current algorithms that integrate the patient’s electronic health record, labs, and vital signs. It’s fair to say that we still have a long way to go to get morbidity and mortality down from sepsis, but use of AI and omics may contribute. There’s another strategy that is rarely used but could make a big difference thanks to the latest technology—sequencing the blood or body fluid.