Authors
Bhavna Singla, Priya Gupta, Anum Fatima, Shivam Singla, Sunita Kumawat, Vimi Bansal, Daniel E Cook, Taha Khalid
Published in
Cureus. Volume 18. Issue 7. Pages e113193. Epub Jul 22, 2026.
Abstract
Sepsis remains a major cause of morbidity and mortality worldwide, and delayed recognition continues to compromise timely intervention. In recent years, artificial intelligence (AI) has been increasingly applied to continuously updated clinical data to facilitate earlier detection of sepsis; however, the quality, interpretability, and clinical readiness of these models remain uncertain. This systematic review evaluated real-time AI models designed for early sepsis prediction using dynamic hospital data. A structured search of PubMed/MEDLINE, Scopus, and Web of Science identified studies published between January 2015 and June 2025. Eligible studies included adult hospitalized populations, employed machine learning or deep learning approaches using sequential or continuously updated data, and reported predictive performance for sepsis onset detection. Eight studies met the inclusion criteria. These studies were conducted across intensive care units, emergency departments, and multicenter hospital systems, with sample sizes ranging from several hundred to more than 500,000 patients. Model architectures included gradient boosting methods, neural networks, recurrent survival models, and deep learning prediction platforms. Reported discriminatory performance was moderate to high, with area under the receiver operating characteristic curve values generally ranging from 0.83 to above 0.95, and several studies demonstrated clinically meaningful lead times before sepsis onset or treatment initiation. More recent investigations increasingly incorporated external validation, transfer learning, false-alert mitigation strategies, and explainability methods, such as feature attribution and Shapley additive explanations analysis. Risk of bias assessment using the Prediction model Risk Of Bias ASsessment Tool indicated that most studies had a moderate overall risk of bias, primarily due to retrospective design, heterogeneous sepsis definitions, and limitations in analytical reporting. Current evidence suggests that real-time AI shows considerable promise for early sepsis recognition; however, prospective validation, calibration assessment, workflow integration, and demonstration of consistent patient benefit remain essential before widespread clinical implementation can be justified.
PMID:
42632992
Bibliographic data and abstract were imported from PubMed on 23 Aug 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 5
- Comments 0