An artificial intelligence (AI) tool achieved high diagnostic accuracy for the identification and prediction of sepsis
1. Bhargava and colleagues evaluated an AI tool’s ability to identify sepsis at presentation in adults with suspected infection.
2. The tool demonstrated high diagnostic accuracy in identifying suspected sepsis and predicting clinical outcomes.
Evidence Rating Level: 2 (Good)
Study Rundown: Sepsis is a serious medical condition for which early treatment is vital for patient outcomes. However, heterogeneity in presentation makes early recognition challenging. Bhargava and colleagues developed and validated the Sepsis ImmunoScore AI tool to identify patients likely to have or progress to sepsis within 24 hours of patient assessment. The study enrolled hospitalized adults with suspected infection, identified through a blood culture order and who had a lithium-heparin (Li-Hep) plasma sample drawn within 6 hours of the first blood culture. The primary outcome was sepsis at presentation or within 24 hours, defined using Sepsis-3 criteria. Secondary outcomes included in-hospital mortality, length of stay, intensive care unit admission, mechanical ventilation, and vasopressor use. The AI tool achieved an area under the receiver operating characteristic curve (AUROC) of 0.81 in external validation. Sepsis occurred in 3.0% of patients categorized as low risk and 69.7% of those categorized as very high risk, with corresponding increases in mortality and other adverse outcomes. This study demonstrated that the Sepsis ImmunoScore tool can stratify risk among patients with suspected infection.
Click here to read the study in NEJM AI
Relevant Reading: Evaluation of Sepsis Prediction Models before Onset of Treatment
In-Depth [prospective cohort]: This study included 3457 adult patient encounters with valid Sepsis ImmunoScore results, with 2366 encounters in the derivation cohort, 393 in the internal validation cohort, and 698 in the external validation cohort. Hospitals represented during tool development were used for internal validation, and different hospitals were used for external validation. The rate of sepsis was 32% in the derivation cohort, 28% in the internal validation cohort, and 22% in the external validation cohort. Diagnostic accuracy was assessed by determining the ability of the Sepsis ImmunoScore and its corresponding risk stratification category (low, medium, high, or very high) to identify patients who meet Sepsis-3 criteria within 24 hours of presentation and secondary outcomes. The tool achieved an AUROC of 0.85 (95% confidence interval [CI], 0.83-0.87) in the derivation group, 0.80 (95% CI, 0.74-0.86) in the internal validation group, and 0.81 (95% CI, 0.77-0.86) in the external validation group. External validation sepsis rates were 3.0% (95% CI, 1.2-6.1%), 12.7% (95% CI, 7.96-19.0%), 36.6% (95% CI, 30.1-42.6%), and 69.7% (95% CI, 51.3-84.4%) across low, medium, high, and very high-risk categories. Corresponding mortality rates were 0.0% (95% CI, 0.0-1.6%), 1.9% (95% CI, 0.40-5.5%), 8.7% (95% CI, 5.7-12.7%), and 18.2% (95% CI, 7.0-35.5%). This study was limited by possible outcome misclassification and reliance on blood culture ordering to identify suspected infection. Overall, this study demonstrated that Sepsis ImmunoScore achieved a high accuracy for identifying and predicting sepsis metrics, which could enable prompt intervention.
Image: PD
©2026 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.
