Validation of a Machine Learning Diagnostic Tool for the Prediction of Sepsis and Critical Illness

BackgroundSepsis is a life-threatening condition that demands prompt treatment for improved patient outcomes. Its heterogenous presentation makes early detection challenging, highlighting the need for effective risk assessment tools. Artificial Intelligence (AI) models have the potential to accurately identify septic patients, but none have previously been FDA-authorized for commercial use. This study outlines the development and validation of the Sepsis ImmunoScore, the first FDA-authorized AI-based software designed to identify patients at risk of sepsis. MethodIn this prospective study, adult patients (18+) suspected of infection, as indicated by a blood culture order, were enrolled from five U.S. institutions between April 2017 and July 2022. The participants were divided into an algorithm development cohort (n=2,366), an internal validation cohort (n=393), and an external validation cohort (n=698). The primary endpoint was the presence of sepsis (Sepsis-3) within 24 hours of test initiation. Secondary endpoints included hospital length of stay, ICU admission within 24 hours, mechanical ventilation use within 24 hours, vasopressor use within 24 hours, and in-hospital mortality. ResultsThe Sepsis ImmunoScore demonstrated high diagnostic accuracy, with an AUC of 0.85 (0.83-0.87) in the derivation cohort, 0.80 (0.74-0.86) in internal validation, and 0.81 (0.77- 0.86) in external validation. The score was categorized into four risk levels for sepsis with corresponding likelihood ratios: low (0.1), medium (0.5), high (2.1), and very high (8.3). These risk categories also predicted in-hospital mortality: low (0.0%), medium (1.9%), high (8.7%), and very high (18.2%) in the external validation cohort. Similar trends were observed for other metrics, such as hospital length of stay, ICU utilization, mechanical ventilation, and vasopressor use. ConclusionsThe Sepsis ImmunoScore demonstrated high accuracy for identification and prediction of sepsis and critical illness that could enable prompt identification of patients at high risk of sepsis and adverse outcomes, potentially improving clinical decision-making and patient outcomes. DescriptionSepsis is a life-threatening acute condition that requires accurate and rapid identification to guide proper treatment. This study outlines the development and validation of the first FDA-authorized AI-based software to identify patients at risk of having sepsis.