Identifying Sepsis Phenotypes Associated with Antibiotic-Resistant Pathogens Using Large Language Models and Machine Learning

Sepsis is a severe and life-threatening condition that occurs when the body's response to an infection causes widespread inflammation, leading to tissue damage, organ failure, and potentially death. It is a major health problem because it is difficult to diagnose and treat effectively, often requiring immediate medical attention. Current guidelines recommend the immediate use of broad-spectrum antibiotics for suspected sepsis to cover all possible pathogens. However, this approach often leads to the overuse of these powerful antibiotics, even when they are not needed. This overuse contributes to the growing problem of antibiotic resistance, where bacteria evolve to become immune to the effects of antibiotics, making infections harder to treat in the future.

This research aims to improve the accuracy of sepsis treatment by identifying cases where narrow-spectrum antibiotics are effective, reducing broad-spectrum antibiotic overuse, minimizing the risk of resistance, and promoting better antibiotic use.

The specific aims of the research are as follows:

• Apply large language models (LLMs) to clinical notes to identify presenting syndromes, recent antibiotic exposures, and recent healthcare exposures in patients with suspected sepsis, and assess the marginal value of these free-text data vs structured data alone to predict antibiotic choice and appropriateness.
• Quantify the impact of overly broad vs appropriate vs overly narrow antibiotics on patient outcomes in patients with culture-positive sepsis.
• Identify phenotypes of suspected sepsis for whom the negative impact of overly broad early antibiotic coverage most likely outweighs therapeutic benefit.

The researchers will use machine learning (ML) and LLMs to retrospectively study electronic health records (EHR) data from nearly 50,000 patients with suspected sepsis. They will analyze both unstructured data, such as doctors' notes, and structured data, like lab results and vital signs, to gather detailed information about each patient's condition. By creating detailed profiles of each patient, they aim to develop predictive models to determine the best type of antibiotic treatment.

This research fills a critical gap in current medical guidelines by promoting smarter, more personalized care for sepsis patients. It aims to improve treatment by helping doctors select the most appropriate antibiotics, reducing unnecessary use of broad-spectrum drugs. By leveraging artificial intelligence to analyze EHR data, the study will identify patterns in antibiotic use across different types of sepsis patients and determine when aggressive coverage is unnecessary—minimizing the risks of both overly broad and overly narrow treatment while ensuring patients receive effective care. The findings will contribute to improved clinical guidelines, supporting more precise antibiotic use, reducing antibiotic resistance, and preventing overuse. Ultimately, this research has the potential to improve patient outcomes and inform data-driven healthcare policies for better sepsis management.