This study aims to prospectively validate the GRADY prediction models, which use machine learning algorithms to estimate the risk of gram-negative bacteremia and sepsis in intensive care unit (ICU) patients based on routinely collected vital signs and laboratory data. Sepsis, a life-threatening condition associated with high ICU mortality, requires early diagnosis and treatment-yet current diagnostic methods relying on blood cultures are time-consuming. Existing scoring systems such as SOFA, SIRS, and NEWS2 often lack sufficient sensitivity and specificity in early sepsis detection. Unlike traditional tools, the GRADY models seek to provide earlier and more accurate risk stratification. This study will compare the clinical performance of GRADY models against standard scoring systems and explore their integration as early warning tools to support rapid intervention and improve outcomes in critical care.