The Use of Artificial Intelligence for the Early Detection of Sepsis in the Emergency Department

Lessons learned about policymaking: Moving an emergency department-initiated screening protocol to systemwide input in the development and implementation process.

Most septic patients are initially encountered in the emergency department where sepsis recognition is often delayed, in part due to the lack of effective biomarkers. This study evaluated the diagnostic accuracy of peripheral blood monocyte distribution width alone and in combination with WBC count for early sepsis detection in the emergency department. An Institutional Review Board approved, blinded, observational, prospective cohort study conducted between April 2017 and January 2018. Subjects were enrolled from emergency departments at three U.S. academic centers. Adult patients, 18-89 years, with complete blood count performed upon presentation to the emergency department, and who remained hospitalized for at least 12 hours. A total of 2,212 patients were screened, of whom 2,158 subjects were enrolled and categorized per Sepsis-2 criteria, such as controls (n = 1,088), systemic inflammatory response syndrome (n = 441), infection (n = 244), and sepsis (n = 385), and Sepsis-3 criteria, such as control (n = 1,529), infection (n = 386), and sepsis (n = 243). The primary outcome determined whether an monocyte distribution width of greater than 20.0 U, alone or in combination with WBC, improves early sepsis detection by Sepsis-2 criteria. Secondary endpoints determined monocyte distribution width performance for Sepsis-3 detection. Monocyte distribution width greater than 20.0 U distinguished sepsis from all other conditions based on either Sepsis-2 criteria (area under the curve, 0.79; 95% CI, 0.76-0.82) or Sepsis-3 criteria (area under the curve, 0.73; 95% CI, 0.69-0.76). The negative predictive values for monocyte distribution width less than or equal to 20 U for Sepsis-2 and Sepsis-3 were 93% and 94%, respectively. Monocyte distribution width greater than 20.0 U combined with an abnormal WBC further improved Sepsis-2 detection (area under the curve, 0.85; 95% CI, 0.83-0.88) and as reflected by likelihood ratio and added value analyses. Normal WBC and monocyte distribution width inferred a six-fold lower sepsis probability. An monocyte distribution width value of greater than 20.0 U is effective for sepsis detection, based on either Sepsis-2 criteria or Sepsis-3 criteria, during the initial emergency department encounter. In tandem with WBC, monocyte distribution width is further predicted to enhance medical decision making during early sepsis management in the emergency department.

Sepsis is an organ dysfunction caused by a dysregulated host response to infection. Early detection is fundamental to improving the patient outcome. Laboratory medicine can play a crucial role by providing biomarkers whose alteration can be detected before the onset of clinical signs and symptoms. In particular, the relevance of monocyte distribution width (MDW) as a sepsis biomarker has emerged in the previous decade. However, despite encouraging results, MDW has poor sensitivity and positive predictive value when compared to other biomarkers. This study aims to investigate the use of machine learning (ML) to overcome the limitations mentioned earlier by combining different parameters and therefore improving sepsis detection. However, making ML models function in clinical practice may be problematic, as their performance may suffer when deployed in contexts other than the research environment. In fact, even widely used commercially available models have been demonstrated to generalize poorly in out-of-distribution scenarios. In this multicentric study, we developed ML models whose intended use is the early detection of sepsis on the basis of MDW and complete blood count parameters. In total, data from 6 patient cohorts (encompassing 5344 patients) collected at 5 different Italian hospitals were used to train and externally validate ML models. The models were trained on a patient cohort encompassing patients enrolled at the emergency department, and it was externally validated on 5 different cohorts encompassing patients enrolled at both the emergency department and the intensive care unit. The cohorts were selected to exhibit a variety of data distribution shifts compared to the training set, including label, covariate, and missing data shifts, enabling a conservative validation of the developed models. To improve generalizability and robustness to different types of distribution shifts, the developed ML models combine traditional methodologies with advanced techniques inspired by controllable artificial intelligence (AI), namely cautious classification, which gives the ML models the ability to abstain from making predictions, and explainable AI, which provides health operators with useful information about the models' functioning. The developed models achieved good performance on the internal validation (area under the receiver operating characteristic curve between 0.91 and 0.98), as well as consistent generalization performance across the external validation datasets (area under the receiver operating characteristic curve between 0.75 and 0.95), outperforming baseline biomarkers and state-of-the-art ML models for sepsis detection. Controllable AI techniques were further able to improve performance and were used to derive an interpretable set of diagnostic rules. Our findings demonstrate how controllable AI approaches based on complete blood count and MDW may be used for the early detection of sepsis while also demonstrating how the proposed methodology can be used to develop ML models that are more resistant to different types of data distribution shifts.

To create and evaluate a continuous automated alert system embedded in the electronic health record for the detection of severe sepsis among pediatric inpatient and emergency department patients. Retrospective cohort study. The main outcome was the algorithm's appropriate detection of severe sepsis. Episodes of severe sepsis were identified by chart review of encounters with clinical interventions consistent with sepsis treatment, use of a diagnosis code for sepsis, or deaths. The algorithm was initially tested based upon criteria of the International Pediatric Sepsis Consensus Conference; we present iterative changes which were made to increase the positive predictive value and generate an improved algorithm for clinical use. A quaternary care, freestanding children's hospital with 404 inpatient beds, 70 ICU beds, and approximately 60,000 emergency department visits per year PATIENTS:: All patients less than 18 years presenting to the emergency department or admitted to an inpatient floor or ICU (excluding neonatal intensive care) between August 1, 2016, and December 28, 2016. Creation of a pediatric sepsis screening algorithm. There were 288 (1.0%) episodes of severe sepsis among 29,010 encounters. The final version of the algorithm alerted in 9.0% (CI, 8.7-9.3%) of the encounters with sensitivity 72% (CI, 67-77%) for an episode of severe sepsis; specificity 91.8% (CI, 91.5-92.1%); positive predictive value 8.1% (CI, 7.0-9.2%); negative predictive value 99.7% (CI, 99.6-99.8%). Positive predictive value was highest in the ICUs (10.4%) and emergency department (9.6%). A continuous, automated electronic health record-based sepsis screening algorithm identified severe sepsis among children in the inpatient and emergency department settings and can be deployed to support early detection, although performance varied significantly by hospital location.

Early Detection and Treatment of Severe Sepsis in the Emergency Department: Identifying Barriers to Implementation of a Protocol-based Approach

Public Health Initiatives in the Emergency Department: Not So Good for the Public Health?

Association of ischemia modified albumin with mortality in qSOFA positive sepsis patients by sepsis-3 in the emergency department.

Public Health and Emergency Medicine

Early detection and management of sepsis are crucial for patient survival. Emergency departments (EDs) play a key role in sepsis management but face challenges in timely response due to high patient volumes. Sepsis alert systems are proposed to expedite diagnosis and treatment initiation per the Surviving Sepsis Campaign guidelines. To review and analyze the association of sepsis alert systems in EDs with patient outcomes. A thorough search was conducted in PubMed, EMBASE, Web of Science, and the Cochrane Library from January 1, 2004, to November 19, 2023. Studies that evaluated sepsis alert systems specifically designed for adult ED patients were evaluated. Inclusion criteria focused on peer-reviewed, full-text articles in English that reported on mortality, ICU admissions, hospital stay duration, and sepsis management adherence. Exclusion criteria included studies that lacked a control group or quantitative reports. The review followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline. Two independent reviewers conducted the data extraction using a standardized form. Any disagreements were resolved through discussion. The data were synthesized using a random-effects model due to the expected heterogeneity among the included studies. Key outcomes included mortality, intensive care unit admissions, hospital stay duration, and adherence to the sepsis bundle. Of 3281 initially identified studies, 22 (0.67%) met inclusion criteria, encompassing 19 580 patients. Sepsis alert systems were associated with reduced mortality risk (risk ratio [RR], 0.81; 95% CI, 0.71 to 0.91) and length of hospital stay (standardized mean difference [SMD], -0.15; 95% CI, -0.20 to -0.11). These systems were also associated with better adherence to sepsis bundle elements, notably in terms of shorter time to fluid administration (SMD, -0.42; 95% CI, -0.52 to -0.32), blood culture (SMD, -0.31; 95% CI, -0.40 to -0.21), antibiotic administration (SMD, -0.34; 95% CI, -0.39 to -0.29), and lactate measurement (SMD, -0.15; 95% CI, -0.22 to -0.08). Electronic alerts were particularly associated with reduced mortality (RR, 0.78; 95% CI, 0.67 to 0.92) and adherence with blood culture guidelines (RR, 1.14; 95% CI, 1.03 to 1.27). These findings suggest that sepsis alert systems in EDs were associated with better patient outcomes along with better adherence to sepsis management protocols. These systems hold promise for enhancing ED responses to sepsis, potentially leading to better patient outcomes.

Diagnosis of bacteraemia in well-appearing children who present to the paediatric emergency department for fever.

Exploring the Role of Artificial Intelligence in an Emergency and Trauma Radiology Department.

This study aimed to explore the perceived barriers and enablers influencing the adoption of artificial intelligence (AI)-based triage tools in emergency departments (EDs) from the perspective of frontline healthcare professionals. A qualitative research design was employed, utilizing semi-structured interviews with 19 participants—including emergency physicians, triage nurses, department managers, clinical administrators, and health informatics experts—working in emergency departments across Canada. Participants were selected using purposive sampling to ensure diversity in professional roles and institutional settings. Data collection continued until theoretical saturation was reached. Interviews were transcribed verbatim and analyzed using grounded theory methodology. Open, axial, and selective coding were conducted with the assistance of NVivo software to identify emerging themes and construct a conceptual model of AI adoption dynamics. The analysis revealed five core categories shaping AI-based triage adoption: (1) perceived risk and uncertainty, including lack of trust in AI outputs and concerns over legal liability; (2) institutional and organizational readiness, such as infrastructure limitations and workflow misalignment; (3) human capital and knowledge systems, including digital literacy gaps and lack of training; (4) system-level support and governance, highlighting the role of managerial commitment and national policy frameworks; and (5) value proposition and practical benefits, including efficiency gains, clinical decision support, and user-friendly integration. These categories reflected the interplay of technical, organizational, and human factors that either hindered or enabled AI integration in emergency care settings. Adopting AI-based triage tools in emergency departments requires addressing a complex ecosystem of trust, readiness, training, infrastructure, and systemic support. The findings underscore the importance of clinician engagement, targeted education, transparent design, and multi-level policy alignment to ensure effective and sustainable implementation.

BackgroundSepsis is a critical emergency condition characterized by life-threatening organ dysfunction due to a dysregulated response to infection. In the fast-paced emergency department (ED) setting, rapid identification and prompt initiation of treatment within the initial hours following sepsis onset are critical for reducing mortality and improving patient outcomes. However, a timely and accurate diagnosis remains a significant challenge in emergency medicine. Biomarkers such as procalcitonin (PCT) and presepsin (P-SEP) have been proposed as tools to distinguish sepsis from other non-infectious inflammatory conditions frequently encountered in the ED, though their diagnostic effectiveness remains controversial. This study aimed to evaluate the diagnostic performance of PCT and P-SEP for diagnosis patients with sepsis.MethodsA comprehensive systematic search was conducted across the Cochrane Central Register of Controlled Trials, PubMed, and Scopus databases up to April 1st, 2024 and updated on June 30th, 2025. Studies reporting sensitivity and specificity of PCT and P-SEP for sepsis detection among patients in acute and emergency settings were included. Hierarchical modeling techniques were utilized to pool data for sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC) along with their 95% confidence intervals (CIs).ResultsThirty-eight observational studies met inclusion criteria. The pooled sensitivities and specificities for detecting sepsis using PCT were 0.78 (95% CI: 0.74–0.81) and 0.77 (95% CI: 0.71–0.82), respectively. Similarly, for P-SEP, pooled sensitivity and specificity were 0.82 (95% CI: 0.77–0.86) and 0.78 (95% CI: 0.73–0.83), respectively. No statistically significant differences were identified between PCT and P-SEP regarding sensitivity (p = 0.169) or specificity (p = 0.792). The summary receiver operating characteristic analysis yielded an AUROC of 0.84 (95% CI: 0.81–0.87) for PCT and 0.87 (95% CI: 0.84–0.90) for P-SEP.ConclusionsBoth PCT and P-SEP represent reliable biomarkers for early and accurate sepsis detection in acute and ED settings, demonstrating comparable diagnostic performance. Their integration into routine ED assessment protocols may support timely clinical decision-making and prompt initiation of appropriate treatment strategies.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12873-025-01433-3.

Background: Sepsis is a life threating medical condition where infection leads to massive inflammatory response and eventually lead to organ dysfunction. It continues to pose a significant health threat despite remarkable developments in critical care medicine and extensive study of septic patients. Early recognition and treatment of sepsis in emergency department (ED) is important to reduce mortality, hospital length of stay and morbidity. Objective: This study was aimed to validate the performance of qSOFA scoring system and SIRS criteria in early sepsis diagnosis in the Emergency Department. Patient and method: This prospective observational clinical study was carried out in Emergency Department (ED) on 100 patients with suspected infection presented and admitted at o Mansoura University Emergency Hospital from February 2019 to February 2020. Patients were divided into two groups: infection group with qSOFA –ve criteria and sepsis group with qSOFA +ve group. We compare between qSOFA and SIRS scores in both groups. Results: The study demonstrated that organ dysfunction >2, 2ry to infection (according to sepsis definition by sepsis-3 task force) was more frequently reported among qSOFA + SIRS + group, compared to qSOFA - SIR – group. Moreover, both qSOFA and SIRS had comparable sensitivity (100%) in prediction of mortality within 1 week, while qSOFA demonstrated higher specificity (53.3%) in comparison with SIRS (20%). Both scores had comparable sensitivity for prediction ICU admission and of mechanical ventilation (86.67 and 88.9) whereas qSOFA demonstrated higher specificity than SIRS for ICU admission (94.29 versus 71.43) as well as mechanical ventilation (82.9 versus 63.41). Conclusion: It could be concluded that qSOFA is considered as specific not sensitive tool, while SIRS is more sensitive but not specific score for sepsis detection in emergency room.

BackgroundThe initial presentation of sepsis in the emergency department (ED) is difficult to distinguish from other acute illnesses based upon similar clinical presentations. A new blood parameter, a measurement of increased monocyte volume distribution width (MDW), may be used in combination with other clinical parameters to improve early sepsis detection. We sought to determine if MDW, when combined with other available clinical parameters at the time of ED presentation, improves the early detection of sepsis.MethodsA retrospective analysis of prospectively collected clinical data available during the initial ED encounter of 2158 adult patients who were enrolled from emergency departments of three major academic centers, of which 385 fulfilled Sepsis-2 criteria, and 243 fulfilled Sepsis-3 criteria within 12 h of admission. Sepsis probabilities were determined based on MDW values, alone or in combination with components of systemic inflammatory response syndrome (SIRS) or quick sepsis-related organ failure assessment (qSOFA) score obtained during the initial patient presentation (i.e., within 2 h of ED admission).ResultsAbnormal MDW (> 20.0) consistently increased sepsis probability, and normal MDW consistently reduced sepsis probability when used in combination with SIRS criteria (tachycardia, tachypnea, abnormal white blood count, or body temperature) or qSOFA criteria (tachypnea, altered mental status, but not hypotension). Overall, and regardless of other SIRS or qSOFA variables, MDW > 20.0 (vs. MDW ≤ 20.0) at the time of the initial ED encounter was associated with an approximately 6-fold increase in the odds of Sepsis-2, and an approximately 4-fold increase in the odds of Sepsis-3.ConclusionsMDW improves the early detection of sepsis during the initial ED encounter and is complementary to SIRS and qSOFA parameters that are currently used for this purpose. This study supports the incorporation of MDW with other readily available clinical parameters during the initial ED encounter for the early detection of sepsis.Trial registrationClinicalTrials.gov, NCT03145428. First posted May 9, 2017. The first subjects were enrolled June 19, 2017, and the study completion date was January 26, 2018.