Sepsis Alerts in Emergency Departments: A Systematic Review of Accuracy and Quality Measure Impact

Abstract

Introduction For early detection of sepsis, automated systems within the electronic health record have evolved to alert emergency department (ED) personnel to the possibility of sepsis, and in some cases link them to suggested care pathways. We conducted a systematic review of automated sepsis-alert detection systems in the ED. Methods We searched multiple health literature databases from the earliest available dates to August 2018. Articles were screened based on abstract, again via manuscript, and further narrowed with set inclusion criteria: 1) adult patients in the ED diagnosed with sepsis, severe sepsis, or septic shock; 2) an electronic system that alerts a healthcare provider of sepsis in real or near-real time; and 3) measures of diagnostic accuracy or quality of sepsis alerts. The final, detailed review was guided by QUADAS-2 and GRADE criteria. We tracked all articles using an online tool (Covidence), and the review was registered with PROSPERO registry of reviews. A two-author consensus was reached at the article choice stage and final review stage. Due to the variation in alert criteria and methods of sepsis diagnosis confirmation, the data were not combined for meta-analysis. Results We screened 693 articles by title and abstract and 20 by full text; we then selected 10 for the study. The articles were published between 2009–2018. Two studies had algorithm-based alert systems, while eight had rule-based alert systems. All systems used different criteria based on systemic inflammatory response syndrome (SIRS) to define sepsis. Sensitivities ranged from 10–100%, specificities from 78–99%, and positive predictive value from 5.8–54%. Negative predictive value was consistently high at 99–100%. Studies showed some evidence for improved process-of-care markers, including improved time to antibiotics. Length of stay improved in two studies. One low quality study showed improved mortality. Conclusion The limited evidence available suggests that sepsis alerts in the ED setting can be set to high sensitivity. No high-quality studies showed a difference in mortality, but evidence exists for improvements in process of care. Significant further work is needed to understand the consequences of alert fatigue and sensitivity set points.

Machine Learning Model to Identify Sepsis Patients in the Emergency Department: Algorithm Development and Validation

“…However, the in-hospital mortality for patients with sepsis present on admission, using the Sepsis-3 criteria for sepsis identification was similar to the previous epidemiologic result (13.5% in our study vs. 13.4% in previous study) [24]. Second, heterogeneity in study designs has been shown to hinder the comparison of performance among different models [18,27]. Therefore, we could hardly compare the performance of our model with previous models [50][51][52][53] because considerable heterogeneity was found among our study and previous studies, including the patient characteristics (e.g., all ED visits versus only ED admissions, with differences in sepsis prevalence), selecting predictors (e.g., vitals, lab test results versus text data from the EHR), the reference standard for sepsis (ICD coding versus the Sepsis-3 criteria), and diagnostic outcomes (e.g., sepsis versus severe sepsis and septic shock versus mortality).…”

“…The increasing availability of electronic health records (EHR) and advancing machine learning (ML) techniques has stimulated attempts to identify patient conditions through the automated analysis of medical records. Previous studies have shown that the ML approach can facilitate the detection of sepsis and septic shock [10,[16][17][18]. However, the clinical utility of these models in the emergency department (ED) setting remains uncertain.…”

“…However, the clinical utility of these models in the emergency department (ED) setting remains uncertain. The majority of previous studies developed and validated the ML models using clinical data only from the intensive care unit (ICU) [18][19][20][21][22][23].…”

Protective Effect of Arbidol Against Pulmonary Fibrosis and Sepsis in Mice

“…Sepsis, a complex disease, characterized by a host’s dysfunctional response to infection that leads to life-threatening multiorgan dysfunction ( Dugar et al, 2020 ). Studies have found that the release of inflammatory cytokines, such as tumor necrosis factor (TNF-α) and interleukins (IL-1β and IL-6) increases, and promotes many immunopathological processes in sepsis, which are often referred to as “cytokines storm” ( Ono et al, 2018 ; Hwang et al, 2020 ). As the most common source of infection in patients with sepsis comes from the abdominal cavity, in recent years, the abdominal infection model has been gradually used as a commonly used animal model of sepsis.…”

Recent Advances in Bedside Device-Based Early Detection of Sepsis

“…Neither automated nor rule-based sepsis alerts in a systematic review conducted in 2020 showed any difference in mortality, nor were potential harms assessed. 21 Another systematic review in 2015 highlighted the danger of alert fatigue as many ICU patients can meet the SIRS criteria without being septic. 22 Therefore, other researchers have turned to developing device-based diagnostics that can be performed at the bedside.…”