• Susan P McGrath, Irina Perreard, Todd MacKenzie, and Michael Calderwood.
• Analytics Institute and Department of Anesthesiology, Dartmouth-Hitchcock Medical Center, Lebanon, NH, USA. Electronic address: Susan.p.mcgrath@hitchcock.org.
• Am J Emerg Med. 2022 Jan 1; 51: 239-247.

BackgroundSepsis remains a leading cause of death among inpatients. Scoring systems designed to identify inpatients with sepsis currently have limited effectiveness. This single institution, retrospective, case-control study aims to improve sepsis decision support tool performance using temporal analyses of sepsis-specific and general deterioration scoring systems.MethodsSequential Organ Failure Assessment, National Early Warning Scores (NEWS), and Modified Early Warning Scores were calculated using four years of inpatient data. Sensitivity and specificity analyses compared performance of each score, calculated as a function of both various score cut-off values and time before sepsis diagnosis using established proxies for identifying clinical suspicion for sepsis.ResultsNEWS had the best sensitivity-specificity performance (AUROC 82.7) when examining various score cutoffs and time intervals during which diagnosis criteria were met. Comparison of false positives/negatives with various score thresholds showed a low rate of false positives with a NEWS of 7. Score trends in the hours leading up to sepsis criteria being met showed a marked increase for the sepsis group while for the cases there was a decrease during a comparable period.ConclusionsTemporal analyses of scores for patients coded as having sepsis provides novel insights into patterns of deterioration. The methods and results provide practical details demonstrating how general deterioration algorithms can be used to alert trained responders to potential cases of sepsis to improve sepsis recognition and treatment opportunities.Copyright © 2021 Elsevier Inc. All rights reserved.

20 keywords...

hide…