Over the years, the prevalence of infections among critically ill patients admitted to the hospital has been high. The risk of mortality and length of admission in critical patient care is established to be associated with the immediacy of appropriate antimicrobial administration, especially within the first hour of admission. This clinical gap for timely antimicrobial prescription necessitates the need for initial prescriptions of targeted empirical antimicrobial therapy in critically ill patients.
To address this problem, our focus aims at building a robust retrieval-augmented generation (RAG) – language model that assists in accurate empirical antimicrobial prescription therapy by retrieving context-aware, evidence-based recommendations based on past antimicrobial resistance (AMR) patterns, patient data, and clinical notes, and stands in line to further inform streamlining and adjustment of therapy once microbiological results become available. Our solution would involve using a dataset-based clinical decision support system (CDSS) that conducts case-matching, recommendations and infection inferences by retrieving patterns from open AMR registries (with datasets provided and outsourced), and parsing to a generative language model for a detailed and enhanced markdown of the accurate empirical antibiotic prescription therapy.
Our solution aims at developing an accurate and scalable CDSS for the immediate delivery of appropriate antibiotic therapy in critically ill patients, to reduce patient mortality and length of stay in the hospital.