Challenges in early detection and prognostication of sepsis: new approaches from the emergency department and intensive care unit.

Challenges in early detection and prognostication of sepsis: new approaches from the emergency department and intensive care unit - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice eClinicalMedicine . 2026 Apr 6;94:103864. doi: 10.1016/j.eclinm.2026.103864 Search in PMC Search in PubMed View in NLM Catalog Add to search Challenges in early detection and prognostication of sepsis: new approaches from the emergency department and intensive care unit Thomas Lafon Thomas Lafon a Emergency Department, CHU Limoges, Limoges, France b Inserm CIC 1435, CHU Limoges, Limoges, France Find articles by Thomas Lafon a, b , Melanie Weingart Melanie Weingart c Division of Pulmonary, Critical Care, Allergy and Sleep Medicine, University of California, San Francisco, San Francisco, CA, USA Find articles by Melanie Weingart c , Julien Vaidie Julien Vaidie b Inserm CIC 1435, CHU Limoges, Limoges, France d Medical-Surgical Intensive Care Unit, CHU Limoges, Limoges, France Find articles by Julien Vaidie b, d , Carolyn S Calfee Carolyn S Calfee c Division of Pulmonary, Critical Care, Allergy and Sleep Medicine, University of California, San Francisco, San Francisco, CA, USA Find articles by Carolyn S Calfee c , Shevin T Jacob Shevin T Jacob e Department of Clinical Sciences, Liverpool School of Tropical Medicine, Liverpool, United Kingdom f Walimu Kampala, Uganda Find articles by Shevin T Jacob e, f , Yonathan Freund Yonathan Freund g Sorbonne Université, IMProving Emergency Care FHU, Paris, France h Emergency Department, Hôpital Pitié-Salpêtrière, Assistance Publique-Hôpitaux de Paris (AP-HP), Paris, France Find articles by Yonathan Freund g, h , Nathan I Shapiro Nathan I Shapiro i Department of Emergency Medicine, Beth Israel Deaconess Medical Center, Boston, MA, USA Find articles by Nathan I Shapiro i , Olivier Barraud Olivier Barraud b Inserm CIC 1435, CHU Limoges, Limoges, France j Limoges University, INSERM, CHU Limoges, UMR 1092, Limoges, France Find articles by Olivier Barraud b, j , Guillaume Monneret Guillaume Monneret k Immunology Laboratory, Edouard Herriot Hospital - Hospices Civils de Lyon, Lyon, France l EA 7426 Pathophysiology of Injury-Induced Immunosuppression, Claude Bernard Lyon 1 University, Lyon, France Find articles by Guillaume Monneret k, l , Tom van der Poll Tom van der Poll m Center of Infection and Molecular Medicine, Division of Infectious Diseases, Amsterdam University Medical Center, University of Amsterdam, Amsterdam, the Netherlands Find articles by Tom van der Poll m , Yeleen Fromage Yeleen Fromage b Inserm CIC 1435, CHU Limoges, Limoges, France Find articles by Yeleen Fromage b , Bruno François Bruno François b Inserm CIC 1435, CHU Limoges, Limoges, France d Medical-Surgical Intensive Care Unit, CHU Limoges, Limoges, France j Limoges University, INSERM, CHU Limoges, UMR 1092, Limoges, France Find articles by Bruno François b, d, j, ∗ Author information Article notes Copyright and License information a Emergency Department, CHU Limoges, Limoges, France b Inserm CIC 1435, CHU Limoges, Limoges, France c Division of Pulmonary, Critical Care, Allergy and Sleep Medicine, University of California, San Francisco, San Francisco, CA, USA d Medical-Surgical Intensive Care Unit, CHU Limoges, Limoges, France e Department of Clinical Sciences, Liverpool School of Tropical Medicine, Liverpool, United Kingdom f Walimu Kampala, Uganda g Sorbonne Université, IMProving Emergency Care FHU, Paris, France h Emergency Department, Hôpital Pitié-Salpêtrière, Assistance Publique-Hôpitaux de Paris (AP-HP), Paris, France i Department of Emergency Medicine, Beth Israel Deaconess Medical Center, Boston, MA, USA j Limoges University, INSERM, CHU Limoges, UMR 1092, Limoges, France k Immunology Laboratory, Edouard Herriot Hospital - Hospices Civils de Lyon, Lyon, France l EA 7426 Pathophysiology of Injury-Induced Immunosuppression, Claude Bernard Lyon 1 University, Lyon, France m Center of Infection and Molecular Medicine, Division of Infectious Diseases, Amsterdam University Medical Center, University of Amsterdam, Amsterdam, the Netherlands ∗ Corresponding author. Réanimation Polyvalente, CHU Limoges, 87042 Limoges Cedex, France. [email protected] Received 2025 Dec 19; Revised 2026 Mar 17; Accepted 2026 Mar 19; Collection date 2026 Apr. © 2026 The Author(s) This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). PMC Copyright notice PMCID: PMC13090662  PMID: 42005923 Summary In this narrative review, we aimed to provide a comprehensive overview of emerging diagnostic strategies and precision medicine approaches in sepsis, while explicitly acknowledging the heterogeneity of clinical contexts. In the Emergency Department (ED), timely recognition of infection and sepsis represents one of the most frequent and challenging tasks, which may delay management directly increasing morbidity and mortality. Even if very popular and widely used, traditional scores and routine biomarkers remain of limited interest to confirm diagnosis and predict deterioration. Nevertheless, emerging point-of-care tools hold promise such as “real-time microbiology”, bedside immune profiling, and echocardiography for on-time hemodynamic phenotyping. More advanced strategies, such as omics technologies and transcriptomic signatures, offer deeper biological precision, while machine learning and artificial intelligence can integrate high-dimensional ED data to anticipate deterioration and capture the dynamic evolution of sepsis subphenotypes. Many of these tools are already feasible at the bedside and only await integration into routine ED workflows. Embedding them within dedicated sepsis pathways and multidisciplinary teams could optimize global patient care and accelerate the transition toward precision medicine in acute sepsis. Sustainable improvements in sepsis outcomes will most likely not come from isolated devices but from their integration into coordinated and sepsis-specific pathways. Keywords: Sepsis, Diagnosis, Artificial intelligence, Phenotypes, Emergency department Search strategy and selection criteria. This review was designed as a narrative, clinician-oriented position review rather than a systematic or scoping review; therefore, no formal study selection flow, risk-of-bias assessment, or quantitative synthesis were performed. References were identified through searches of PubMed with the search terms “sepsis”, “prediction”, “biomarkers”, “management”, “prognosis”, “scores”, “emergency department”, “screening”, “diagnosis”, “artificial intelligence”, “machine learning”, “transcriptomic”, “and phenotypes” from databased inception until October, 2025. Articles were also identified through searches of the authors’ own files. Only papers published in English were reviewed. The final reference list was generated on the basis of originality and relevance to the broad scope of this Review. Introduction Sepsis is defined by a dysregulated host response to an infection together with organ dysfunction. 1 Recent global estimates indicate that sepsis represents a far greater burden than previously recognized, notably driven by the inclusion of viral sepsis in contemporary epidemiological analysis. 2 Distinguishing true infection at presentation and anticipating clinical course remain major challenges. Microbiological confirmation is obtained in only about 40% of patients with diagnosed sepsis or septic shock, 3 and clinical signs, symptoms, and laboratory findings of sepsis are nonspecific. Meanwhile, late recognition and treatment delay are directly associated with poor outcome. 4 The current diagnostic tools, including conventional microbiologic cultures, biomarkers, pathogen detection tests, and imaging, have substantial limitations. Conventional severity scores and individual biomarkers insufficiently capture the heterogeneity of sepsis pathophysiology, resulting in suboptimal risk stratification. 5 , 6 Emerging approaches integrate clinical variables, biomarkers and scores to define sub-phenotypes 7 , 8 and incorporate artificial intelligence (AI) and machine learning (ML) to optimize patient management, including appropriate use of antibiotics or adjuvant treatment. Sepsis presentations and trajectories differ substantially between emergency department (ED), intensive care unit (ICU), and low- and middle-income country settings, influencing recognition, prognostication, and applicability of diagnostic tools. Accordingly, this review focuses on strategies to enhance sepsis management including early identification, prediction of clinical trajectories, and precision medicine approaches, while considering the unique operational constraints and time-sensitive decision-making inherent to ED. This review adopts a translational and clinician-centered perspective, aiming to incorporate emerging technologies in real-world decision-making. Sepsis management organization and patient journey Sepsis remains the only major emergency condition lacking a universally standardized and structured response pathway. The proposed “chain of survival and rehabilitation for sepsis” 9 provides a structured and systematic approach that has been successfully adapted to trauma, stroke, and myocardial infarction pathways, highlighting the complexity of coordinating prehospital emergency medical services with ED and subsequent in-hospital management of patients with sepsis ( Fig. 1 ). Importantly, the organization of sepsis care and the feasibility of early recognition strategies vary substantially across healthcare systems, as differences in prehospital resources, ED staffing, diagnostic capacity, and ICU availability directly influence the timeliness and structure of sepsis pathways. In addition, retrospective multicenter data show markedly lower compliance with initial resuscitation bundles and higher mortality among hospital-onset sepsis compared with those presenting through the ED. 10 These descriptions emphasize that patient trajectories within the hospital, even if heterogeneous, is an important step toward structured sepsis pathways and timely interventions. Fig. 1. Open in a new tab Comparative emergency care pathways for trauma, STEMI, stroke, and sepsis . STEMI: ST-elevation myocardial infarction; ED: Emergency department; EMS: Emergency medical services; ECG: Electrocardiogram; FAST: Face Arm Speech Time test; CCU: Critical care unit; ICU: intensive care unit; PCI: Percutaneous coronary intervention. Due to the non-specificity of sepsis symptoms, most patients initially contact either their general practitioner (GP) or the emergency services, making these first-line actors pivotal in early disease detection. 11 However, Mulders et al. reported that the majority of GPs do not consistently use sepsis identification guidelines nor prognosis scores (such as systemic inflammatory response syndrome [SIRS], quick sepsis-related organ failure assessment [qSOFA], national [NEWS] or modified [MEWS] early warning scores, SOFA), relying instead on clinical judgement. 12 This may reflect the limited diagnostic performance of these scores: qSOFA exhibits relatively low sensitivity (Se, 40–50%) but moderate specificity (Sp, 70–80%) for mortality prediction, SIRS shows high Se (80–90%) but low Sp (20–30%), and NEWS demonstrates intermediate performance (Se 60–70%, Sp 50–60%), with area-under-the-receiver-operating-curve (AUROC) values ranging from 0.65 (SIRS) to 0.77 (NEWS). 13 Nevertheless, GP referral based on suspected infection or sepsis is associated with reduced short-term mortality (OR = 0.56; 95% CI 0.32–0.97, p = 0.038). 14 While evidence supports prehospital sepsis “bundle strategies” in severely ill patients, the impact of organizational models of emergency medical services remains insufficiently defined. ED-based multi-disciplinary sepsis response teams 15 have shown promise in their ability to improve adherence to guideline-recommended care and clinical outcomes. 16 , 17 National organizations such as the UK Sepsis Trust and the US Sepsis Alliance play pivotal roles in structuring national sepsis care pathways through standardized protocols, continuous clinician education, public awareness campaigns, and quality improvement initiatives. In addition to national initiatives, specific in-hospital and interprofessional education models have been described and systematic reviews show that structured sepsis education for healthcare professionals—including active learning, simulation, and team-based modules—improves sepsis knowledge and care processes. 18 For instance, a pre-post intervention study demonstrated that implementing a dedicated sepsis response team increased bundle adherence from 4.6 to 32%, improved appropriateness of the initial antibiotic therapy from 30 to 79%, and was associated with a hazard ratio of 0.64 (95% CI 0.43–0.94) for 14-day all-cause mortality. 19 Similarly, Lafon et al. observed significant improvements in adherence to the SSC 3-h bundle, including lactate measurement (87% vs. 96%, p = 0.006), initiation of fluid resuscitation (36% vs. 65%, p < 0.001), blood culture collection (83% vs. 93%, p = 0.014), and timely antibiotic administration (18% vs. 46%, p < 0.001), following the establishment of a sepsis unit in the ED. 20 However, recent high-level evidence highlights substantial heterogeneity in the effectiveness of both sepsis bundles and organizational interventions. Systematic reviews report inconsistent associations with mortality for sepsis alerts and standardized bundle programs, with effects strongly influenced by local context, implementation strategies, and baseline care. 15 , 21 , 22 Despite growing consensus on the value of sepsis performance improvement programs, their operational design remains insufficiently defined—a gap that hampers scalability and reproducibility across health systems—and the difficulty in defining ‘trigger time zero’ impede the establishment of universally accepted standards. Despite the continued challenges of implementing a dedicated sepsis pathway upon ED admission—driven by heterogeneity in hospital resources, infrastructure, and organizational models—there is a compelling rationale to pursue this objective. Beyond isolated protocols or educational actions, sepsis care increasingly relies on structured Sepsis Performance Improvement Programs (SPIPs), which integrate early recognition, standardized pathways, and continuous education, as recently showed by Schinkel et al. 23 Embedding multidisciplinary Sepsis Teams or dedicated Sepsis Units as standardized components of emergency care pathways could improve early sepsis diagnosis and optimize guideline adherence, as well as provide the structured framework necessary to implement precision medicine ( Fig. 2 ). The adoption of dedicated sepsis pathways and new diagnostic tools will need to consider both their costs and the resource constraints of different healthcare settings ( Supplementary Figure S1 ). Fig. 2. Open in a new tab Sepsis trajectory in the Emergency department . ED: Emergency department; ICU: intensive care unit. Improving early diagnosis of sepsis Clinical presentation Clinical presentation of sepsis is highly heterogeneous, reflecting variability in the site of infection, causative pathogens, symptom duration, degree of organ dysfunction, and patients' baseline health status. 24 The Sepsis-3 international consensus acknowledges that sepsis can present as “a constellation of clinical signs and symptoms” that make “diagnosis difficult, even for experienced clinicians,” yet no specific recommendations guide clinicians' early sepsis recognition. The absence of a universally accepted reference standard for sepsis, together with the frequent uncertainty regarding the onset of infection and organ dysfunction, makes the definition of a reliable ‘time zero’ inherently challenging and limits the interpretation of time-based diagnostic and therapeutic strategies. In fact, when considering pathophysiology “early sepsis” should be seen as “early onset of sepsis” meaning the turning point when patients move from acute infection to global immune-inflammatory process together with organ dysfunctions, namely sepsis. Recognizing sepsis as a dynamic condition is crucial, as early changes in vital signs, laboratory markers, and organ function over time often provide critical diagnostic and prognostic indicators. Commonly used screening tools like the quick sepsis-related organ failure assessment (qSOFA) or systemic inflammatory response syndrome (SIRS) criteria have shown limited Se and Sp, particularly in prehospital and ED settings. 25 SOFA score remains the reference to define sepsis, though recent updates, such as SOFA-2, does not incorporate additional organ dysfunction yet. 26 According to the Global Burden of Disease 2021 analysis, sepsis affected an estimated 166 million people globally in 2021, resulting in approximately 21.4 million deaths, a substantial increase from previous estimates. This rise is partly explained by improved capture of viral infections, post-COVID-19 data, and recognition of sepsis as a complication of chronic diseases or trauma. 2 Since the COVID-19 pandemic, the routine use of molecular diagnostics, including multiplex polymerase chain reaction (PCR) panels in ED has expanded substantially. Recent studies in adults show that season-specific and multiplex PCR strategies accelerate pathogen identification. Importantly, viral infections are now detected in approximately 30% of adult sepsis cases and may be followed by secondary bacterial infections, particularly during pneumonia. Additionally, the concept of viral sepsis has gained renewed attention: studies highlight that viral infections—including influenza, coronaviruses and other respiratory viruses—contribute substantially to sepsis syndromes in adults. 27 , 28 Pediatric and maternal populations present additional diagnostic challenges due to atypical or non-specific clinical signs. Neonatal and maternal sepsis remain major causes of morbidity and mortality, particularly in low- and middle-income countries (LMIC), where limited access to rapid diagnostics and high-quality care further complicates early recognition. Recent reviews highlight the burden of neonatal sepsis and the constraints of implementing diagnostic solutions. 29 , 30 , 31 Whitfield et al. recently proposed a standardized clinical adjudication protocol to address the lack of a diagnostic gold standard in suspected sepsis. 32 This approach relies on independent expert review of all available clinical, biological, microbiological, and imaging data to classify cases as bacterial, viral, or non-infectious, rather than relying solely on microbiological confirmation. Integrating these strategies into real-life emergency care, alongside emerging machine learning tools and real-time decision support using artificial intelligence, could significantly enhance clinicians’ ability to identify sepsis. Sepsis screening: scores and biomarkers Early recognition of sepsis remains a cornerstone of patient management, but the choice of screening tool has been debated over the past decade. 33 The Sepsis-3 task force introduced the qSOFA as a simple bedside predictor of poor outcome in infected patients. 25 However, subsequent studies consistently demonstrated that qSOFA lacks Se for sepsis identification, particularly in ED and ward populations. Large meta-analyses and cohort studies have shown that early warning scores such as NEWS, NEWS2, MEWS, or even the long-standing SIRS criteria outperform qSOFA in terms of Se for detecting sepsis or identifying patients at risk of deterioration. 34 However, these tools were initially developed to capture risk of poor outcomes rather than sepsis specifically; therefore, their added value in early screening for sepsis and to guide early treatment is not proven. No single tool achieves both high Se and Sp, and bedside clinicians must recognize the limitations of each. The 2021 Surviving Sepsis Campaign (SSC) guidelines therefore recommended against the use of qSOFA as a sole screening tool, compared to SIRS, NEWS, NEWS2, or MEWS. 35 Noteworthy, different scores have been evaluated for different outcomes, i.e; different definitions of sepsis, which limit our ability to compare their performances. However, it remains a useful marker of disease severity, with good specificity and predictive performance comparable to SOFA for organ dysfunction. 36 The simplicity and ease of implementation of clinical scores make them the only readily available tools to support clinical judgment in certain settings. In contrast to biomarkers or even the SOFA score, which require laboratory testing, time and additional resources, these scores can be applied immediately at the bedside, often without delay and extra cost. Inflammatory biomarkers have long been investigated as adjuncts for sepsis recognition. C-reactive protein (CRP) and procalcitonin (PCT) are the most widely available. Both rise in response to bacterial infection but perform modestly for early sepsis detection. 37 , 38 CRP lacks specificity, while PCT is more specific for bacterial infection but rises only several hours after onset, limiting its utility for very early detection. Importantly, PCT reflects more a downstream systemic inflammatory response rather than an early host-response activation phase. While not specific for infection, lactate remains a key prognostic marker reflecting hypoperfusion and mortality risk. 39 Ultimately, sepsis remains a clinical diagnosis, but PCT and CRP may assist in guiding antimicrobial stewardship and de-escalation. Novel host response diagnostics have emerged, including tests such as monocyte distribution width (MDW), SeptiCyte Rapid®, IntelliSep®, or TriVerity®. These assays integrate cellular, transcriptomic, or protein signatures to stratify patients according to sepsis risk. Large prospective studies 40 , 41 confirmed that most prognostic biomarkers still offer only moderate discriminative ability. Meta-analyses also indicate that MDW offers moderate diagnostic performance with pooled Se (84%) and Sp (68%). 42 , 43 Cellular host-response tests like IntelliSep® have demonstrated diagnostic areas under the curve (AUCs) in the 0.74–0.92 range in ED cohorts, stratifying sepsis risk across adjudicated cases. 44 Consequently, there is currently insufficient evidence to recommend any of these assays for routine sepsis screening. Novel approach to pathogen detection and rapid laboratory point-of-care Rapid pathogen detection is key for appropriate management of patients with sepsis but remains a challenge since only 15–40% of patients with sepsis have microbiologically confirmed bloodstream infections, and turnaround time is lengthy. 3 Time-to-positivity of the blood cultures is 9–21 h for most common bacteria 45 and additional time is required to perform conventional pathogen identification and antimicrobial susceptibility testing (AST). Most novel rapid approaches are performed on positive blood cultures or directly on whole blood samples ( Fig. 3 ). From positive blood cultures, many phenotypic and genotypic tools are available. 46 Matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS) 47 directly applied to young or positive blood cultures enables rapid identification of bacteria and fungi within minutes and at low cost. Multiplex syndromic PCR-based approaches can also detect major determinants of bloodstream infections in around 1 h, as well as common extended-spectrum beta-lactamase (ESBL)-, carbapenemase- and methicillin-resistant encoding genes. 48 Recent advances in real-time microscopy, microfluidics, volatile organic compounds or even vibrational technologies resulted in the commercialization of rapid and accurate CE-IVD/FDA-cleared phenotypic tools to allow complete AST in a few hours for a few dozen euros. Innovation is still ongoing with many phenotypic platforms 49 that are not on the market yet. Fig. 3. Open in a new tab Evolution of microbiological diagnosis of sepsis: today and beyond . ID: identification; AST: antimicrobial susceptibility testing; PCR: polymerase chain reaction; MALDI-TOF MS: Matrix-assisted laser desorption ionization time-of-flight mass spectrometry; ARG: antibiotic resistance genes; mNGS: Metagenomic next-generation sequencing. Starting from raw whole blood sample is even more challenging given the low inoculum of the pathogen, but these technologies hold high potential in terms of reducing turnaround times. Technologies in the field are mostly based on genotypic tools, including multiplex real-time PCR, droplet digital PCR or unbiased molecular approaches like metagenomics-based assays. 50 , 51 Metagenomic next generation sequencing (mNGS) is very promising due to fast turnaround time 52 ability to identify both microorganisms, antibiotic resistance genes and applications for other biological samples like respiratory samples. 53 , 54 Unfortunately, mNGS is only available at a few specialized laboratories and still relies on bioinformatics skills. Other mNGS-based techniques are based on detection of microbial cell-free DNA from plasma. 55 Quantities of the detected pathogens can vary from dozens to hundreds with rapid turnaround time (2.5–10 h). A recent meta-analysis indicated that such rapid molecular assays may have value as add-on tests by increasing pathogen detection rates but without replacing blood cultures. 56 Real-time performance assessment of such tests is lacking, and well-designed studies are awaited for assessing their clinical impact. 57 Many innovative genotypic-based and even phenotypic-based technologies, some of them coupled with AI, are in the pipeline, 58 , 59 with prospects of identification of microorganisms and their associated antibiotic resistance profile within hours, in a few years. These innovative tools do not provide sepsis diagnosis but constitute complementary downstream strategies to refine therapy and allow optimized antimicrobial stewardship. Outside their performance, it is important to keep in mind the numerous direct and indirect limits of their use: integration within the ED and laboratory flows is not easy and not transposable from one hospital site to another. False positive results can occur, as already observed for blood cultures positive with coagulase-negative staphylococci, which can bias diagnosis. Lastly, a key question remains as to which patients, at which moment, at which frequency and at which cost such innovative tools should be applied. Special challenges of sepsis diagnosis and clinical course in low and middle-income countries While sepsis remains a leading cause of morbidity and mortality worldwide, its impact is disproportionately concentrated in low- and middle-income countries (LMICs), which account for ∼85% of global cases and deaths. 60 In these settings, particularly when resources are constrained, the clinical course of sepsis is frequently marked by delayed presentation, severe illness on admission and early mortality, reflecting barriers to timely care-seeking, recognition and intervention. Furthermore, comorbid conditions such as HIV, tuberculosis, malaria, and malnutrition increase susceptibility and complicate diagnosis and triage. As well, maternal and neonatal sepsis contributes substantially to preventable mortality, exacerbated by deficits in infection prevention and control and water, sanitation and hygiene infrastructure. 61 Consequently, diagnostic and triage strategies developed in high-income settings cannot be directly transposed to low- and middle-income countries, and must be adapted to local epidemiology, resource availability, and health system capacity in order to enable timely recognition and management of sepsis. Unlike high-income countries (HICs), sepsis in LMICs arises from a broader spectrum of pathogens which add complexity to determining appropriate empiric treatment. For example, severe malaria, typhoid and invasive non-typhoidal Salmonella overlap clinically with bacterial sepsis in Africa and Asia, but variable capacity to obtain blood cultures or other definitive tests alongside rising antimicrobial resistance complicates administration of effective antimicrobial treatment. 62 Furthermore, although HIV-related opportunistic infections like disseminated tuberculosis frequently present as sepsis-like syndromes in settings with high HIV prevalence, 63 standard empiric treatment does not typically include anti-tuberculosis treatment, and availability of accurate diagnostics to identify disseminated tuberculosis is limited. Other important ‘atypical’ sepsis presentations include melioidosis in Southeast Asia and viral hemorrhagic fevers (e.g., Ebola disease, Lassa Fever, Crimean-Congo hemorrhagic fever), which are clinically indistinguishable from bacterial sepsis in early stages and place health workers at risk without rapid diagnostic capacity. At the facility level, the absence of standardized triage, limited laboratory support, and shortages of trained staff hinder early diagnosis. Moreover, basic monitoring, including pulse oximetry or reliable blood pressure measurement, may be inconsistent, and existing sepsis tools derived in HICs, such as the sepsis-3 definition and qSOFA, perform inconsistently in LMIC settings. 64 To help bridge the diagnostic gap in LMICs, context-relevant strategies should integrate structured triage algorithms with affordable point-of-care tools and physiologic bedside assessments. Point-of-care lactate, increasingly available in African hospitals, provides rapid prognostic information and may improve early risk stratification where laboratory capacity is limited. 65 Simple bedside measures such as capillary refill time (CRT) also offer pragmatic guidance and need further evaluation in LMICs; the ANDROMEDA-SHOCK trials have shown the benefit of CRT-targeted hemodynamic resuscitation protocols in multicenter settings. 66 , 67 Affordable inflammatory markers such as CRP may further support infection assessment, although context-specific thresholds and validation remain essential. 68 Emerging host-response biomarkers and digital clinical decision-support tools may further standardize early identification, though context-specific validation and system integration remain essential. 30 , 69 While biomarkers such as PCT may support timing of antimicrobial treatment, cost and context-specific cut-offs limit their use 68 While not ubiquitous, point-of-care lactate testing is increasingly available in African hospitals and strongly predicts mortality 65 Despite recent innovations, a fundamental diagnostics gap persists, with nearly half the world's population lacking access to essential tests. 70 Cross-country facility surveys confirm shortfalls in culture, biomarker, and quality systems outside referral centers. 71 The World Health Organization Essential Diagnostics List provides a policy scaffold, but adoption and financing are uneven. 72 Closing this gap is critical to narrow the global inequity in sepsis outcomes. In resource-constrained LMIC settings, early sepsis recognition and stabilization are predominantly delivered outside ICUs by general medical officers, clinical officers, and nurses. Task shifting approaches that empower non-physician clinicians to initiate time-critical processes, like early sepsis screening, oxygen therapy, and first-dose antimicrobial administration, may reduce delays related to workforce shortages and align with WHO guidance on task-shifting. 73 Quality improvement strategies incorporating low-fidelity simulation and structured audit-and-feedback have been associated with improved adherence to time-sensitive emergency care processes in LMIC contexts. 74 Several brief, adaptable training models provide scalable platforms for strengthening health worker identification and management. The WHO/ICRC Basic Emergency Care (BEC) course teaches a structured ABCDE approach to the recognition and initial management of life-threatening sepsis-related conditions including shock, respiratory distress, and altered mental status. 75 For pediatric populations, the WHO Emergency Triage Assessment and Treatment (ETAT) framework has demonstrated improved early identification and management of severe illness, like sepsis, through protocolized triage and immediate interventions. 76 , 77 Case-based tele-mentoring models such as Project ECHO have also been associated with improved provider confidence and protocol adherence across diverse LMIC settings. 78 Contribution of artificial intelligence AI approaches introduce a fundamentally different analytical paradigm from traditional statistical methods in sepsis research. Classical models are typically hypothesis-driven and rely on predefined physiological or mechanistic assumptions. In contrast, ML methods are data-driven and optimized to learn complex, potentially non-linear relationships directly from the data. This paradigm shift is particularly relevant for sepsis, a clinically heterogeneous syndrome characterized by high dimensionality, correlated predictors, and dynamic trajectories. Importantly, AI should not be viewed as a replacement for conventional statistical approaches, which remain essential benchmarks due to their robustness and interpretability. Rather, its value lies in the complementary ability to leverage large-scale and complex clinical datasets to support individual-level prediction and pattern recognition. In this context, AI-based tools should be conceptualized as supportive instruments for decision-making, designed to enrich clinical assessment and adjudication rather than operate as independent diagnostic platforms. The choice of AI methodology, including conventional ML vs. deep learning, should therefore be guided by the clinical question, data structure, and implementation constraints rather than by assumptions of intrinsic methodological superiority. Accordingly, the clinical relevance of ML in sepsis lies in its integration within existing clinical workflows, where it may support screening and early diagnosis as well as inform assessment of clinical course and prognosis—domains discussed in the following sections. Screening and early diagnosis Artificial intelligence has become an increasingly important focus in the effort to improve early recognition and management of sepsis, however, their clinical impact remains dependent on implementation context, validation quality, and integration within established care pathways 79 The task of synthesizing these diverse data inputs in real time is a challenge to clinicians as well as traditional rule-based systems. AI approaches are well-suited to integrate such high-dimensional data, learning complex nonlinear interactions and generating patient-specific predictions that may enhance clinical decision-making. Recent studies of AI in sepsis emphasize that AI-based models have the potential to outperform these conventional approaches by analyzing continuous streams of electronic health record (EHR) data, physiologic monitoring, and laboratory values. The Targeted Real-time Early Warning System (TREWS), deployed across multiple hospitals, demonstrated that timely provider confirmation of its alerts was associated with reduced mortality, shorter hospital stays, and improved organ failure scores, underscoring both the potential and challenges of implementing AI-based sepsis EWS in real-world practice. 80 , 81 Another example of a successful AI-based algorithm is the Sepsis ImmunoScore, which consistently achieved AUROC values of 0.80–0.85 for sepsis-3 within 24 h. 82 Furthermore, its risk categories correlated with critical illness (e.g., ICU admission, vasopressor use, ventilator use, and mortality), leading to FDA de novo authorization. Another widely used sepsis tool is the EHR Epic's widely deployed sepsis prediction model; however, the first model was unsuccessfully validated in external trials due to a high rate of false alarms, underscoring the dangers of adopting algorithms without transparent development and rigorous external validation. Because real sepsis onset is rarely observable, most diagnostic and prognostic models—including AI-based tools—must rely on proxy labels, retrospective definitions, or expert adjudication, which complicates training, validation, and real-world implementation. A revised EPIC sepsis score is now implemented, and a number of other AI-based algorithms are in development or released, with the need for rigorous validation prior to implementation of these promising new approaches. While studies demonstrate that sepsis AI-based algorithms achieve good predictive accuracy, a diagnosis alone cannot alter outcomes. It must be linked to therapeutic actions to impact outcomes. Thus, the effect of an AI algorithm on patient-oriented outcomes such as mortality depends on implementation, response pathways leading to intervention, and adequate staffing to operationalize those interventions rather than algorithm performance alone. Real-world effectiveness also requires addressing human factors—including alert fatigue, clinician trust in algorithm outputs, seamless workflow integration, and barriers to sustained adoption in high-acuity emergency settings. Clinical course and prognosis Over the past few years, there has been a notable shift towards developing predictive models using data captured at the time of ED admission ( Supplementary Table S1 ). The input features used in these models typically aggregate multimodal data: demographic characteristics, clinical parameters, hemodynamic measures, laboratory biomarkers, and occasionally, free-text from clinical notes. Some models also incorporate dynamic physiological signals such as heart rate variability 83 or ECG. 84 Importantly, interpretability techniques have become essential for clinical adoption. These algorithms—most notably SHapley Additive exPlanations (SHAP) 85 —are specifically designed to clarify how predictions are generated, thereby reducing the “black box” effect and strengthening clinician trust. Despite often reporting promising performance metrics (with AUROC ranging from 0.69 to 0.99), 86 comparing these models remains difficult due to substantial heterogeneity in study designs. Key challenges include a lack of standardization in sepsis definitions, inconsistent prediction windows, and variable outcome definitions. Moreover, methodological standards such as TRIPOD + AI 87 and PROBAST + AI 88 are insufficiently implemented in practice. Models are developed using heterogeneous pipelines, with differing preprocessing strategies, performance metrics, thresholds, and validation approaches (e.g., internal cross-validation vs. external validation), hindering model performance comparisons. A common limitation across studies is the exclusive reliance on AUROC as a performance metric. Although it provides an overall measure of discrimination, AUROC can be misleading in highly imbalanced datasets—typical in sepsis research—where positive events are rare. 89 In such contexts, high AUROC values may coexist with limited clinical utility, reflecting the well-described “accuracy paradox,” whereby seemingly strong performance can be achieved despite poor identification of true positive cases. Because ROC curves are based on sensitivity and specificity, which are not directly affected by disease prevalence, this metric may yield overly optimistic estimates when the primary clinical objective is early case detection. In contrast, precision–recall analysis explicitly incorporates precision (positive predictive value), a metric that is directly influenced by disease prevalence and therefore more closely aligned with clinical decision-making in low-prevalence settings. As a result, the area under the precision–recall curve (AUPRC) provides a more clinically relevant assessment of a model's ability to correctly identify true positives. 90 One approach to mitigate on AUROC and AUPRC is to include calibration metrics—ensuring predicted probabilities reflect observed risk—and clinical utility metrics such as decision-curve analysis and net benefit. These measures clarify whether an algorithm meaningfully improves decision-making beyond existing care pathways, yet they remain inconsistently reported in sepsis AI research. Current implementation barriers First, the majority of models have been developed using retrospective datasets, potentially introducing bias due to poor data quality and missing information. 91 This limitation also applies to the definition of the outcomes to be predicted, which are frequently derived from claims-based or International Classification of Diseases-coded definitions that represent imperfect surrogates of the underlying clinical reality. As ML models are inherently data-driven, their performance is directly constrained by ground-truth uncertainty. In this regard, standardized clinical adjudication frameworks 32 provide a more reliable reference rather than administrative codes alone. Despite this, the baseline ‘time zero’ of sepsis onset is often unknown or unknowable, creating a major hurdle for ML-based tools. Furthermore, the evolution of clinical criteria—such as the transition from Sepsis-2 to Sepsis-3—illustrates the risk of temporal drift; a model trained on historical definitions may suffer from performance degradation as diagnostic standards and clinical practices shift over time. Second, many studies do not share code or make their algorithms publicly available, undermining reproducibility and transparency. Third, although the use of large numbers of variables allows for a truly multidimensional and personalized approach, it complicates integration across centers and limits scalability for routine clinical use. Of note, the study by Park et al. 92 is innovative in that it relies not on EHR data from a single institution but on the National Inpatient Sample, a nationally representative US dataset, enhancing potential generalizability. Moreover, most models are trained on data from the US or Asia, raising concerns about geographic and population-specific generalizability. Limited representation of certain demographic or clinical subgroups may lead to differential model performance and raise concerns regarding fairness and equity, which are critical considerations for real-world implementation. Given center-specific data structures, it may be unrealistic to expect a single universal model. Each institution may ultimately need to train its own, a challenge that underscores the inherent transportability limits of current ML tools. Importantly, only a small number of models have been validated in prospective, real-world settings—critical steps for assessing clinical relevance. 93 , 94 , 95 Finally, the lack of accessible and user-friendly tools impedes clinical uptake. Collectively, these constraints explain why few ML-based models have translated into bedside tools that are ready for use in real-time clinical scenarios. Future strategies must move beyond static validation and incorporate frameworks for continuous monitoring to ensure long-term algorithmic safety after deployment into clinical workflows. One promising direction involves stratification-based approaches 82 , 96 rather than complex numerical predictions, which may be difficult to interpret and apply in fast-paced ED environments. Predictors capturing host immune response patterns are increasingly explored, as they may better reflect clinical trajectories. ML has been central to leveraging omics data and advancing sepsis subphenotyping. Sepsis phenotypes and precision medicine in ED Sepsis phenotypes in EDs and in ICUs Within the heterogenous syndrome of sepsis, studies have increasingly emphasized clinical subphenotypes that predict outcomes and are associated with differential response to treatments due to shared pathobiological mechanisms. 97 This ongoing paradigm shift is driven by a growing body of evidence that leverages large observational sepsis cohorts and randomized controlled trials (RCTs) to identify subphenotypes using a combination of clinical, biomarker and molecular inputs early in the course of illness. Prospective subphenotype classification has the potential to improve early prognostication and identify which patients will benefit from specific therapies. However, optimal timing and data inputs for subphenotype classification have varied significantly by study ( Supplementary Table S2 ). In the ED, sepsis subphenotyping studies are largely limited to clinical data obtained from the EHR on patient arrival, in part due to the time required to enroll and collect biologic samples. For example, ML analysis applied to 29 EHR-derived clinical variables collected within the first 6 h after ED arrival identified four clinical subphenotypes associated with mortality. 98 A separate study using longitudinal data over the first 72 h after hospital arrival identified four temperature trajectories that similarly correlated with mortality. 99 Subphenotyping efforts for sepsis patients in the ICU have frequently leveraged the inclusion of protein biomarkers and transcriptomics ( Supplementary Table S2 ). Several independent studies conducted within the first 24 h of ICU admission, in either the ED or ICU, have identified 2 subphenotypes distinguished in large part by their plasma inflammatory biomarkers. 100 , 101 The “hyperinflammatory” and “Phenotype 1” are characterized by higher inflammatory biomarkers, greater mortality and differential response to activated protein C and early goal-directed therapy in secondary analysis of PROWESS-SHOCK and ProCESS trials, respectively. 100 , 101 The plasma protein biomarkers required for accurate classification of these phenotypes (e.g. IL-6, sTNFr1) can be measured in real time 102 and are being used at present to stratify enrolment in a global adaptive platform clinical trial ( https://panthertrial.org ). Additionally, analyses of whole blood mRNA transcriptomics early in ICU admission have identified 2–4 transcriptional endotypes that are associated with mortality and differential response to corticosteroids, suggesting this as a promising approach. 103 , 104 Overall, early ICU-based subphenotypes better distinguish prognostically relevant subgroups with variable treatment responses, although timely subphenotyping is also important, as patients become less hyperinflammatory over time. 105 Prospective studies with inclusion of targeted biomarkers in the ED are needed to determine optimal timing of subphenotype classification and advance precision medicine in sepsis. Emerging tools to rapidly identify immune status associated with poor outcome in EDs Cellular immunology provides complementary approaches to study the immune response that are highly relevant in sepsis, including absolute cell counting, detailed phenotyping (via clusters of differentiation [CD]), and functional testing. Recent advances suggest promising clinical applications in the ED. 106 A first application is rapid differentiation between bacterial and viral sepsis, using distinct CD marker expression. Neutrophils overexpress CD64 (IgG receptor) in bacterial infections, while monocytes upregulate CD169 (Siglec-1) in viral infections. Both show high specificity (>90%), with CD64 long established and CD169 highlighted during COVID-19. 107 Current flow cytometry can stain whole blood in minutes and deliver results in under 15 min—even from a fingerstick sample—making this approach feasible for emergency use. 108 While automated prototypes are under evaluation, limited ED access to flow cytometry currently remains the main barrier. A second application concerns prediction of short-term deterioration. Two flow cytometry markers are of particular interest: immature neutrophils (IN, CD10 - /CD16 - ) and decreased HLA-DR expression on monocytes (mHLA-DR). Elevated IN % reflects “emergency granulopoiesis” and correlates with bacterial load and severity. 109 , 110 Early mHLA-DR loss indicates compensatory immunosuppression and is considered a marker of poor outcome when measured at patients’ admission. 111 While not yet used routinely in the ED, both biomarkers could be integrated into multiparameter assays. With multicolor flow cytometry, CD64, CD169, IN, and mHLA-DR can be measured simultaneously. 107 , 108 , 112 Early pneumonia studies combining these markers with simple clinical variables (heart rate, respiratory rate, oxygen saturation) improved short-term risk prediction. 113 Severe lymphopenia also added prognostic value in this study. Traditional metrics such as absolute lymphocyte count or neutrophil-to-lymphocyte ratio, although limited alone, may be useful within such multiparameter strategies. Importantly, a massive increase in mHLA-DR may reveal IFN-γ–driven macrophage activation syndrome (MALS), an extreme hyperinflammatory state of sepsis. 114 , 115 The PROVIDE randomized trial recently demonstrated that ferritin levels, combined with mHLA-DR assessment, can help identify patients with this hyperinflammatory immunophenotype. 116 Altogether, these approaches could refine risk stratification, identify low-risk patients suitable for early discharge, and better allocate resources. Further validation in ED cohorts remains essential. A third emerging approach aims to bypass flow cytometry through functional assays. Though historically impractical in the ED, recent work shows global lymphocyte function can be assessed automatically via interferon-gamma release assay (IGRA) testing (response to a mitogen, independent of antigen specificity). This method was first validated in ICU sepsis 117 and has now been tested in the ED (# NCT06155266 ), with results expected in 2026. Like mHLA-DR, a profound lymphocyte functional defect is hypothesized to mark severe immunosuppression and could signal the need for closer monitoring. Omics technologies to early profile sepsis severity Systematic reviews and meta-analyses have reinforced the diagnostic potential of transcriptomic biomarkers. A review covering 117 studies and over 17,000 patients reported a pooled AUROC of 0.86 for transcriptomic signatures distinguishing sepsis from controls—including SIRS—significantly outperforming conventional biomarkers like CRP or PCT. 118 ML has been instrumental in unlocking complex omics datasets. A recent publication applied ML to transcriptomic data, identifying key sepsis biomarkers (e.g., BMX, GRB10, GADD45A) and revealing immune-related pathways and therapeutic targets. 119 Early detection and outcome prediction using omics technologies are particularly promising. 120 A network meta-analysis demonstrated that ML models outperform traditional scoring systems (e.g., SOFA and simplified acute physiology score [SAPS] II), with neural networks and decision trees, particularly for short-term predictions. 121 Yet, a scoping review of feature engineering in ML for sepsis underscored the critical value of vital signs and laboratory values, especially when integrated into algorithms like random forests and XGBoost. 122 Conceivably, integration of data derived from advanced omics technologies with data that are readily available from clinical practice can push this field further. Despite these advances, challenges remain. Many studies are retrospective or rely on discovery cohorts, risking overfitting and limited external validity. Many analyses used samples obtained on admission to the ICU, while the host response to sepsis is highly dynamic in nature and most omics techniques are not routinely available or rapid enough for clinical use in the ED. 123 Early echocardiography and hemodynamics profile: phenotypes and prognosis Early and accurate identification of the leading mechanism of cardiovascular failure in sepsis is critical to subsequently improve prognosis. 124 Echocardiography provides unparalleled information on central hemodynamics and is currently recommended as the first-line approach for the diagnostic work-up of circulatory failure. 125 It promptly identifies various cardiovascular phenotypes in septic patients. 126 Five distinct hemodynamic profiles have been described using early echocardiography assessment of patients presenting with septic shock: left ventricular (LV) systolic dysfunction (symptomatic septic cardiomyopathy), hyperkinetic state (sustained vasoplegia), fluid responsiveness (persistent hypovolemia), right ventricular (RV) failure (frequently associated with acute respiratory distress syndrome), and “normalized” hemodynamic profiles (i.e., none of the preceding abnormalities after initial resuscitation). 127 These cardiovascular phenotypes can be observed early in the course of sepsis including, LV or RV systolic dysfunction in one-third of patients. 128 The impact of echocardiography-guided management of sepsis-induced circulatory failure on the course of organ failure is currently evaluated in both the ED and ICU settings ( NCT04580888 and NCT04166331 ). Due to the inability of clinical assessment to accurately predict the underlying hemodynamic profile of septic patients, early and repeated echocardiography assessment best allows to identity cardiovascular phenotypes, their evolution over time, and both the efficacy and tolerance of therapeutic interventions. 127 The growing education of emergency physicians and intensivists in critical care echocardiography promises to best guide initial management. Biomarker-guided early interventions A cutting-edge frontier in sepsis management is the use of biomarker-guided targeted therapeutics in the ED, namely theranostic approaches, that aim to stratify patients for novel drugs. Instead of non-specific biomarker-guided interventions, mechanism-based approaches including specific biomarker should be considered for septic shock treatment. An expert workshop recently underscored early-phase clinical promises for agents such as nangibotide, a TREM-1 pathway inhibitor tailored to patients with elevated soluble TREM-1 (sTREM-1), 129 and enibarcimab, an anti-adrenomedullin antibody for those with elevated bio-Adrenomedullin (bioADM) levels. 6 , 130 At the very early stage of septic shock, procizumab 131 is a monoclonal antibody (mAb) targeting dipeptidyl peptidase 3 (DPP3), a ubiquitous cytosolic enzyme released into the bloodstream after tissue injury, that can degrade angiotensin II. This mAb is also moving into clinical development and could represent another promising theranostic approach in the ED, knowing that DPP3 can also be measured at the bedside in a timely fashion. 132 These interventions exemplify the paradigm in which biomarker measurement at ED presentation could identify the patients most likely to benefit from pathway-specific therapies. Current work also emphasizes time-resolved measurements of biomarkers to monitor evolving host response, which may guide dosing or treatment duration of these targeted agents. 6 For example, repeating bioADM levels at 6 or 12 h post-intervention might inform whether continuation of adrenomedullin blockade is indicated. However, several challenges remain before widespread adoption of biomarker-guided drugs in the ED: assay availability and feasibility in ED workflows, turnaround time, standardization of cut-offs, and demonstrating impact on hard outcomes such as mortality or organ failure. Moreover, regulatory pathways for biomarker-guided new drugs require robust trial data showing efficacy specifically in biomarker-positive patients. 6 Biomarker guided administration of new sepsis drugs represent a promising evolution in early sepsis therapy, especially when deployed in the ED setting, to match immunomodulatory agents to patients with specific host response profiles. Current frontrunners exemplify this precision medicine approach, but their successful translation into the ED requires mature point-of-care assays, standardized protocols matching biomarker thresholds to treatment arms, prospective trials demonstrating outcome benefit, and integration within emergency care workflows. Outstanding questions Unlike stroke or myocardial infarction, sepsis lacks a clear and standardized “chain of survival” with adapted structures and specific procedures, contributing to delays and variability in recognition and management. This could be developed in the future by creating specific sepsis pathways in the ED based on dedicated multidisciplinary teams and dedicated tools such as comprehensive immune profiling at the point of care, multi-omic subphenotyping and timely biomarker-guided interventions ( Supplementary Table S3 ). Large-scale prospective RCTS in real-world settings are also needed to demonstrate that AI-assisted interventions could improve clinical management, beyond predictive accuracy. Establishing a standardized, ILCOR-style reporting framework or registry for sepsis could help reduce heterogeneity in data collection, facilitate benchmarking, and support the development of structured clinical pathways. Conclusion Sepsis remains one of the most challenging and heterogeneous syndromes in critical care medicine and early recognition with accurate prediction of its clinical course continues to be an unmet need. While ICU have historically driven advances in understanding and management, the ED has emerged as a decisive setting where rapid actions can determine outcomes. Current diagnostic and prognostic strategies do not adequately capture the clinical and biological complexity of sepsis, underscoring the need for precision medicine adapted to the ED. This should integrate emerging biomarkers, real-time microbiological point-of-care testing, and artificial intelligence tools able to take into consideration multidimensional data and identify patients at high risk of deterioration. Future advances in sepsis care are unlikely to rely on a single breakthrough tool; rather, they will depend on the intelligent integration of multiple data streams - clinical, biological, and physiological. Supported by dedicated sepsis teams or specialized units participating of organizational framework for implementation, embedding such innovations into ED workflows could transform early sepsis management and improve trajectories. Contributors TL, JV, YeF and BF oversaw the writing of the manuscript. All authors wrote parts of the original draft and all reviewed and edited the manuscript. All authors approved the final version. Data sharing statement As this is a narrative review, no original data were collected. Declaration of interests BF declares a grant from the French government (PHRC-23-0002) to his institution, consulting fees from Enlivex, PPD, and Eagle, and participation in Advisory boards for Aurobac and Inotrem. CC declares a grant from NIH to her institution, grants from Roche-Genetech, Quantum Leap Healthcare Collaborative, and DOD to her institution, consulting fees from Vasomune, Gen1e Life Sciences, NGM Bio, Cellenkos, Calcimedica, Arrowhead, EnliTISA, Novartis, Aerogen, Boehringer, Merck, Healios, and Matisse, having spoken at a symposium sponsored by Fisher Paykel, being co-recipient of a patent on metagenomic sequencing for sepsis diagnosis and being an unpaid council member of the International Sepsis Forum. MW declares a NIH Training grant (T32HL007185-48). OB declares consulting fees from bioMérieux, personal fees from VIDAL France, PFIZER SAS, MSD France and SHIONOGI, meeting or travel support from EUMEDICA SA, PFIZER SAS and MSD France, being co-inventor on a pending patent (WO2021205330A1), and being cofounder and CSO in DAMOCLES Diagnostics. TVDP declares grants from the Ministry of Economic Affairs & Health Holland, the Dutch Thrombosis Foundation and EU Horizon 2020 (FAIR, No 847786) to his institution and unpaid participation in REMAP-CAP advisory board. STJ declares a grant from the German Federal Ministry of Research, Technology and Space (BMFTR) for the Sub-Saharan African Consortium for the Advancement of Innovative Research and Care in Sepsis (STAIRS) to his institution, and being Secretary General and Executive Committee Member of the African Sepsis Alliance, Executive Committee Member of the Global Sepsis Alliance, and Committee Member of the Surviving Sepsis Campaign Guidelines Committee. NS declares grants from Bleujay Diagnostics and Lumos Diagnostics to his institution, consulting fees from AccUrine, Prenosis, and Cambridge Medical Technologies and equity options from AccUrine, Prenosis and Cambridge Medical Technologies. The other authors have no conflicts of interest to declare. Acknowledgements As some authors are non-native English speakers, generative AI, namely ChatGPT (version GPT-4.1, free, OpenAI) was used occasionally to check on the grammar and syntax of specific sentences during the writing of the manuscript. However, all the authors, including native English speakers, ultimately proofread the final text. Footnotes Appendix A Supplementary data related to this article can be found at https://doi.org/10.1016/j.eclinm.2026.103864 . Appendix A. Supplementary data Supplementary Figure and Tables mmc1.docx (62.6KB, docx) References 1. Singer M., Deutschman C.S., Seymour C.W., et al. The third international consensus definitions for sepsis and septic shock (Sepsis-3) JAMA. 2016;315:801–810. doi: 10.1001/jama.2016.0287. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Gray A.P., Chung E., Hsu R.L., et al. Global, regional, and national sepsis incidence and mortality, 1990–2021: a systematic analysis. Lancet Glob Health. 2025 doi: 10.1016/S2214-109X(25)00356-0. [ DOI ] [ PubMed ] [ Google Scholar ] 3. Li Y., Guo J., Yang H., Li H., Shen Y., Zhang D. Comparison of culture-negative and culture-positive sepsis or septic shock: a systematic review and meta-analysis. Crit Care. 2021;25:167. doi: 10.1186/s13054-021-03592-8. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Seymour C.W., Gesten F., Prescott H.C., et al. Time to treatment and mortality during mandated emergency care for sepsis. N Engl J Med. 2017;376:2235–2244. doi: 10.1056/NEJMoa1703058. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Antcliffe D.B., Burrell A., Boyle A.J., Gordon A.C., McAuley D.F., Silversides J. Sepsis subphenotypes, theragnostics and personalized sepsis care. Intensive Care Med. 2025;51:756–768. doi: 10.1007/s00134-025-07873-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Llitjos J.-F., Carrol E.D., Osuchowski M.F., et al. Enhancing sepsis biomarker development: key considerations from public and private perspectives. Crit Care. 2024;28:238. doi: 10.1186/s13054-024-05032-9. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Castela Forte J., Perner A., van der Horst I.C.C. The use of clustering algorithms in critical care research to unravel patient heterogeneity. Intensive Care Med. 2019;45:1025–1028. doi: 10.1007/s00134-019-05631-z. [ DOI ] [ PubMed ] [ Google Scholar ] 8. DeMerle K.M., Angus D.C., Baillie J.K., et al. Sepsis subclasses: a framework for development and interpretation. Crit Care Med. 2021;49:748–759. doi: 10.1097/CCM.0000000000004842. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Hidalgo J.L., Kumar V.K., Akech S.O., et al. The sepsis chain of survival: a comprehensive framework for improving sepsis outcomes. Crit Care Med. 2025;53:e1886–e1892. doi: 10.1097/CCM.0000000000006796. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Leisman D.E., Angel C., Schneider S.M., D'Amore J.A., D'Angelo J.K., Doerfler M.E. Sepsis presenting in hospitals versus emergency departments: demographic, resuscitation, and outcome patterns in a multicenter retrospective cohort. J Hosp Med. 2019;14:340–348. doi: 10.12788/jhm.3188. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Loots F.J., Arpots R., van den Berg R., Hopstaken R.M., Giesen P., Smits M. Recognition of sepsis in primary care: a survey among GPs. BJGP Open. 2017;1 doi: 10.3399/bjgpopen17X100965. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Mulders M.C.F., Loots F.J., van Nieuwenhoven J., Ter Maaten J.C., Bouma H.R. Use of sepsis-related diagnostic criteria in primary care: a survey among general practitioners. Fam Pract. 2021;38:617–622. doi: 10.1093/fampra/cmab020. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Wang C., Xu R., Zeng Y., Zhao Y., Hu X. A comparison of qSOFA, SIRS and NEWS in predicting the accuracy of mortality in patients with suspected sepsis: a meta-analysis. PLoS One. 2022;17 doi: 10.1371/journal.pone.0266755. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 14. Nardot A., Lacorre C., Lanneluc A., et al. A program to improve sepsis management in the Emergency Department: a multicenter prospective study in France. Intern Emerg Med. 2025;20:2523–2530. doi: 10.1007/s11739-025-03877-8. [ DOI ] [ PubMed ] [ Google Scholar ] 15. Uffen J.W., Oosterheert J.J., Schweitzer V.A., Thursky K., Kaasjager H.A.H., Ekkelenkamp M.B. Interventions for rapid recognition and treatment of sepsis in the emergency department: a narrative review. Clin Microbiol Infect. 2021;27:192–203. doi: 10.1016/j.cmi.2020.02.022. [ DOI ] [ PubMed ] [ Google Scholar ] 16. Ju T., Al-Mashat M., Rivas L., Sarani B. Sepsis rapid response teams. Crit Care Clin. 2018;34:253–258. doi: 10.1016/j.ccc.2017.12.004. [ DOI ] [ PubMed ] [ Google Scholar ] 17. Ludikhuize J., Marshall D., Devchand M., et al. Improving the management of medical emergency team calls due to suspected infections: a before-after study. Crit Care Resusc. 2023;25:136–139. doi: 10.1016/j.ccrj.2023.06.004. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Choy C.L., Liaw S.Y., Goh E.L., See K.C., Chua W.L. Impact of sepsis education for healthcare professionals and students on learning and patient outcomes: a systematic review. J Hosp Infect. 2022;122:84–95. doi: 10.1016/j.jhin.2022.01.004. [ DOI ] [ PubMed ] [ Google Scholar ] 19. Viale P., Tedeschi S., Scudeller L., et al. Infectious diseases team for the early management of severe sepsis and septic shock in the emergency department. Clin Infect Dis. 2017;65:1253–1259. doi: 10.1093/cid/cix548. [ DOI ] [ PubMed ] [ Google Scholar ] 20. Lafon T., Baisse A., Karam H.H., et al. Sepsis unit in the emergency department: impact on management and outcome of septic patients. Shock. 2023;60:157–162. doi: 10.1097/SHK.0000000000002155. [ DOI ] [ PubMed ] [ Google Scholar ] 21. Kim H.-J., Ko R.-E., Lim S.Y., Park S., Suh G.Y., Lee Y.J. Sepsis alert systems, mortality, and adherence in emergency departments: a systematic review and meta-analysis. JAMA Netw Open. 2024;7 doi: 10.1001/jamanetworkopen.2024.22823. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Ford J.S., Morrison J.C., Kyaw M., et al. The effect of severe sepsis and septic shock management bundle (SEP-1) compliance and implementation on mortality among patients with sepsis: a systematic review. Ann Intern Med. 2025;178:543–557. doi: 10.7326/ANNALS-24-02426. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Schinkel M., Nanayakkara P.W.B., Wiersinga W.J. Sepsis performance improvement programs: from evidence toward clinical implementation. Crit Care. 2022;26:77. doi: 10.1186/s13054-022-03917-1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Meyer N.J., Prescott H.C. Sepsis and septic shock. N Engl J Med. 2024;391:2133–2146. doi: 10.1056/NEJMra2403213. [ DOI ] [ PubMed ] [ Google Scholar ] 25. Seymour C.W., Liu V.X., Iwashyna T.J., et al. Assessment of clinical criteria for sepsis: for the third international consensus definitions for sepsis and septic shock (Sepsis-3) JAMA. 2016;315:762–774. doi: 10.1001/jama.2016.0288. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Ranzani O.T., Singer M., Salluh J.I.F., et al. Development and validation of the Sequential Organ Failure Assessment (SOFA)-2 score. JAMA. 2025;334:2090–2103. doi: 10.1001/jama.2025.20516. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. G Gürtler L., Schramm W., Seitz R. Viral sepsis – pathophysiology and disease manifestation. Infection. 2025;53:775–784. doi: 10.1007/s15010-025-02486-z. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Li H., Liu L., Zhang D., et al. SARS-CoV-2 and viral sepsis: observations and hypotheses. Lancet. 2020;395:1517–1520. doi: 10.1016/S0140-6736(20)30920-X. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 29. Dramowski A., Bolton L., Fitzgerald F., Bekker A., NeoNET AFRICA Partnership Neonatal sepsis in low- and middle-income countries: where are we now? Pediatr Infect Dis J. 2025;44:e207–e210. doi: 10.1097/INF.0000000000004815. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 30. Gleeson B., Ferreyra C., Palamountain K., et al. A call to bridge the diagnostic gap: diagnostic solutions for neonatal sepsis in low- and middle-income countries. BMJ Glob Health. 2024;9 doi: 10.1136/bmjgh-2024-015862. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Rees C.A., Lim J., Westbrook A.L., et al. Systematic review and meta-analysis of the diagnostic value of four biomarkers in detecting neonatal sepsis in low- and middle-income countries. BMJ Paediatr Open. 2023;7 doi: 10.1136/bmjpo-2022-001627. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Whitfield N.N., Hogan C.A., Chenoweth J., et al. A standardized protocol using clinical adjudication to define true infection status in patients presenting to the emergency department with suspected infections and/or sepsis. Diagn Microbiol Infect Dis. 2024;110 doi: 10.1016/j.diagmicrobio.2024.116382. [ DOI ] [ PubMed ] [ Google Scholar ] 33. Bolanaki M., Kurland L., Brabrand M., et al. Current sepsis management practices in European emergency departments: the ISG-emergency department European Survey. Eur J Emerg Med. 2025;32:368–376. doi: 10.1097/MEJ.0000000000001255. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Serafim R., Gomes J.A., Salluh J., Póvoa P. A comparison of the quick-SOFA and systemic inflammatory response syndrome criteria for the diagnosis of sepsis and prediction of mortality: a systematic review and meta-analysis. Chest. 2018;153:646–655. doi: 10.1016/j.chest.2017.12.015. [ DOI ] [ PubMed ] [ Google Scholar ] 35. Evans L., Rhodes A., Alhazzani W., et al. Surviving sepsis campaign: international guidelines for management of sepsis and septic shock 2021. Intensive Care Med. 2021;47:1181–1247. doi: 10.1007/s00134-021-06506-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 36. Freund Y., Lemachatti N., Krastinova E., et al. Prognostic accuracy of Sepsis-3 criteria for In-Hospital mortality among patients with suspected infection presenting to the emergency department. JAMA. 2017;317:301–308. doi: 10.1001/jama.2016.20329. [ DOI ] [ PubMed ] [ Google Scholar ] 37. Ghatak T., Pal A., Rochwerg B., et al. The diagnostic accuracy of serum procalcitonin for sepsis in critically ill adults. A systematic review and diagnostic meta-analysis. J Crit Care. 2026;92 doi: 10.1016/j.jcrc.2025.155402. [ DOI ] [ PubMed ] [ Google Scholar ] 38. Chuang C.-L., Yeh H.-T., Niu K.-Y., Chen C.-B., Seak C.-J., Yen C.-C. Diagnostic performances of procalcitonin and C-reactive protein for sepsis: a systematic review and meta-analysis. Eur J Emerg Med. 2025;32:248–258. doi: 10.1097/MEJ.0000000000001235. [ DOI ] [ PubMed ] [ Google Scholar ] 39. Lu J., Dong Z., Ye L., Gao Y., Zheng Z. Predictive value of SOFA, PCT, Lactate, qSOFA and their combinations for mortality in patients with sepsis: a systematic review and meta-analysis. PLoS One. 2025;20 doi: 10.1371/journal.pone.0332525. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Lafon T., Cazalis M.-A., Vallejo C., et al. Prognostic performance of endothelial biomarkers to early predict clinical deterioration of patients with suspected bacterial infection and sepsis admitted to the emergency department. Ann Intensive Care. 2020;10:113. doi: 10.1186/s13613-020-00729-w. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Lafon T., Cazalis M.-A., Hart K.W., et al. SEPSIGN: early identification of sepsis signs in emergency department. Intern Emerg Med. 2025;20:1575–1587. doi: 10.1007/s11739-024-03802-5. [ DOI ] [ PubMed ] [ Google Scholar ] 42. Huang Y.-H., Chen C.-J., Shao S.-C., et al. Comparison of the diagnostic accuracies of monocyte distribution width, procalcitonin, and C-reactive protein for sepsis: a systematic review and meta-analysis. Crit Care Med. 2023;51:e106–e114. doi: 10.1097/CCM.0000000000005820. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Eisinger G.J., Hosler Q., Crouser E.D., Herman D.D. Diagnostic performance of monocyte distribution width for the detection of sepsis: a systematic review and meta-analysis. JACEP Open. 2025;6 doi: 10.1016/j.acepjo.2025.100073. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. O'Neal H.R., Sheybani R., Janz D.R., et al. Validation of a novel, rapid sepsis diagnostic for emergency department use. Crit Care Explor. 2024;6 doi: 10.1097/CCE.0000000000001026. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 45. Qin Y., Liao Y., Zhou J., et al. Comparative evaluation of BacT/ALERT VIRTUO and BACTEC FX400 blood culture systems for the detection of bloodstream infections. Microbiol Spectr. 2025;13 doi: 10.1128/spectrum.01850-24. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Wolk D.M., Parrott J.S., Babady N.E., et al. The American society for Microbiology's evidence-based laboratory medicine practice guidelines for the diagnosis of bloodstream infections using rapid tests: a systematic review and meta-analysis. Clin Microbiol Rev. 2025;38 doi: 10.1128/cmr.00137-24. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Calderaro A., Chezzi C. MALDI-TOF MS: a reliable tool in the real life of the clinical microbiology laboratory. Microorganisms. 2024;12:322. doi: 10.3390/microorganisms12020322. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Wang Y., Lindsley K., Bleak T.C., et al. Performance of molecular tests for diagnosis of bloodstream infections in the clinical setting: a systematic literature review and meta-analysis. Clin Microbiol Infect. 2025;31:360–372. doi: 10.1016/j.cmi.2024.12.007. [ DOI ] [ PubMed ] [ Google Scholar ] 49. Reszetnik G., Hammond K., Mahshid S., et al. Next-generation rapid phenotypic antimicrobial susceptibility testing. Nat Commun. 2024;15:9719. doi: 10.1038/s41467-024-53930-x. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Peri A.M., Harris P.N.A., Paterson D.L. Culture-independent detection systems for bloodstream infection. Clin Microbiol Infect. 2022;28:195–201. doi: 10.1016/j.cmi.2021.09.039. [ DOI ] [ PubMed ] [ Google Scholar ] 51. Weng S., Lin L., Xie J., et al. Performance of ddPCR-GNB for microbial diagnosis of suspected bloodstream infection due to the four most common gram-negative bacteria: a prospective, multicenter study. Microbiol Spectr. 2025;13:e0101524. doi: 10.1128/spectrum.01015-24. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 52. Gu W., Wang J., Qin X., et al. The performance of nanopore sequencing in rapid detection of pathogens and antimicrobial resistance genes in blood cultures. Diagn Microbiol Infect Dis. 2025;111 doi: 10.1016/j.diagmicrobio.2025.116720. [ DOI ] [ PubMed ] [ Google Scholar ] 53. Charalampous T., Alcolea-Medina A., Snell L.B., et al. Routine metagenomics service for ICU patients with respiratory infection. Am J Respir Crit Care Med. 2024;209:164–174. doi: 10.1164/rccm.202305-0901OC. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 54. Heitz M., Levrat A., Lazarevic V., et al. Metagenomics for the microbiological diagnosis of hospital-acquired pneumonia and ventilator-associated pneumonia (HAP/VAP) in intensive care unit (ICU): a proof-of-concept study. Respir Res. 2023;24:285. doi: 10.1186/s12931-023-02597-x. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 55. Nielsen M.E., Søgaard K.K., Karst S.M., Krarup A.L., Albertsen M., Nielsen H.L. Application of rapid nanopore metagenomic cell-free DNA sequencing to diagnose bloodstream infections: a prospective observational study. Microbiol Spectr. 2025;13:e0329524. doi: 10.1128/spectrum.03295-24. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 56. Rapszky G.A., Do To U.N., Kiss V.E., et al. Rapid molecular assays versus blood culture for bloodstream infections: a systematic review and meta-analysis. eClinicalMedicine. 2025;79 doi: 10.1016/j.eclinm.2024.103028. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 57. Chen Y.-C., Liao P.-H., Chen Y.-W., Yen D.H.-T., How C.-K., Chang C.-M. Optimization of metagenomic next-generation sequencing workflow with a novel host depletion method for enhanced pathogen detection. Mol Diagn Ther. 2025;29:689–699. doi: 10.1007/s40291-025-00797-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 58. Das S., Barman P., Chakraborty R., et al. Optical biosensor for bacteremia detection from human blood samples at a label-free liquid crystal-aqueous interface: a rapid and point-of-care approach. J Colloid Interface Sci. 2025;683:79–89. doi: 10.1016/j.jcis.2024.12.030. [ DOI ] [ PubMed ] [ Google Scholar ] 59. Lim J., Koprowski K., Wester M., Valera E., Bashir R. Review on biphasic blood drying method for rapid pathogen detection in bloodstream infections. SLAS Technol. 2025;32 doi: 10.1016/j.slast.2025.100276. [ DOI ] [ PubMed ] [ Google Scholar ] 60. Rudd K.E., Johnson S.C., Agesa K.M., et al. Global, regional, and national sepsis incidence and mortality, 1990-2017: analysis for the Global Burden of Disease Study. Lancet. 2020;395:200–211. doi: 10.1016/S0140-6736(19)32989-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 61. World Health Organization . World Health Organization; Geneva: 2020. Global Report on the Epidemiology and Burden of Sepsis: Current Evidence, Identifying Gaps and Future Directions. https://iris.who.int/bitstream/handle/10665/334216/9789240010789-eng.pdf?sequence=1 [ Google Scholar ] 62. Naghavi M., Vollset S.E., Ikuta K.S., et al. Global burden of bacterial antimicrobial resistance 1990–2021: a systematic analysis with forecasts to 2050. Lancet. 2024;404:1199–1226. doi: 10.1016/S0140-6736(24)01867-1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 63. Barr D.A., Lewis J.M., Feasey N., et al. Mycobacterium tuberculosis bloodstream infection prevalence, diagnosis, and mortality risk in seriously ill adults with HIV: a systematic review and meta-analysis of individual patient data. Lancet Infect Dis. 2020;20:742–752. doi: 10.1016/S1473-3099(19)30695-4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 64. Adegbite B.R., Edoa J.R., Ndzebe Ndoumba W.F., et al. A comparison of different scores for diagnosis and mortality prediction of adults with sepsis in low-and-middle-income countries: a systematic review and meta-analysis. eClinicalMedicine. 2021;42 doi: 10.1016/j.eclinm.2021.101184. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 65. Moore C.C., Jacob S.T., Pinkerton R., et al. Point-of-care lactate testing predicts mortality of severe sepsis in a predominantly HIV type 1- infected patient population in Uganda. Clin Infect Dis. 2008;46:215–222. doi: 10.1086/524665. [ DOI ] [ PubMed ] [ Google Scholar ] 66. Investigators for the ANDROMEDA Research Network Spanish Society of Anesthesiology Reanimation and Pain Therapy SEDAR and Latin American Intensive Care Network LIVEN, Hernandez G., Ospina-Tascón G.A., et al. Personalized hemodynamic resuscitation targeting capillary refill time in early septic shock: the ANDROMEDA-SHOCK-2 randomized clinical trial. JAMA. 2025;334:1988–1999. doi: 10.1001/jama.2025.20402. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 67. Hernández G., Ospina-Tascón G.A., Damiani L.P., et al. Effect of a resuscitation strategy targeting peripheral perfusion status vs serum lactate levels on 28-day mortality among patients with septic shock: the ANDROMEDA-SHOCK randomized clinical trial. JAMA. 2019;321:654–664. doi: 10.1001/jama.2019.0071. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 68. Escadafal C., Incardona S., Fernandez-Carballo B.L., Dittrich S. The good and the bad: using C reactive protein to distinguish bacterial from non-bacterial infection among febrile patients in low-resource settings. BMJ Glob Health. 2020;5 doi: 10.1136/bmjgh-2020-002396. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 69. Cummings M.J., Jacob S.T. Equitable endotyping is essential to achieve a global standard of precise, effective, and locally-relevant sepsis care. eBioMedicine. 2022;86 doi: 10.1016/j.ebiom.2022.104348. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 70. Fleming K.A., Horton S., Wilson M.L., et al. The lancet commission on diagnostics: transforming access to diagnostics. Lancet Lond Engl. 2021;398:1997–2050. doi: 10.1016/S0140-6736(21)00673-5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 71. Yadav H., Shah D., Sayed S., Horton S., Schroeder L.F. Availability of essential diagnostics in ten low-income and middle-income countries: results from national health facility surveys. Lancet Glob Health. 2021;9:e1553–e1560. doi: 10.1016/S2214-109X(21)00442-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 72. World Health Organization . World Health Organization; Geneva: 2023. The Selection and Use of Essential In Vitro Diagnostics: Report of the Fourth Meeting of the WHO Strategic Advisory Group of Experts on In Vitro Diagnostics, 2022 (Including the Fourth WHO Model List of Essential In Vitro Diagnostics) [ Google Scholar ] 73. World Health Organization. PEPFAR, UNAIDS . World Health Organization; Geneva: 2007. Task Shifting: Rational Redistribution of Tasks Among Health Workforce Teams : Global Recommendations and Guidelines. [ Google Scholar ] 74. Rowe A.K., Rowe S.Y., Peters D.H., Holloway K.A., Chalker J., Ross-Degnan D. Effectiveness of strategies to improve health-care provider practices in low-income and middle-income countries: a systematic review. Lancet Glob Health. 2018;6:e1163–e1175. doi: 10.1016/S2214-109X(18)30398-X. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 75. Basic Emergency Care: Approach to the Acutely Ill and Injured. World Health Organization and the International Committee of the Red Cross; Geneva: 2018. [ Google Scholar ] 76. World Health Organization . World Health Organization, Dept. of Child and Adolescent Health and Development; Geneva, Switzerland: 2005. Emergency Triage Assessment and Treatment (ETAT): Facilitator Guide. [ Google Scholar ] 77. Molyneux E. Improving triage and emergency care for children reduces inpatient mortality in a resource-constrained setting. Bull World Health Organ. 2006;2006:314–319. doi: 10.2471/blt.04.019505. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 78. Zhou C., Crawford A., Serhal E., Kurdyak P., Sockalingam S. The impact of project ECHO on participant and patient outcomes: a systematic review. Acad Med. 2016;91:1439–1461. doi: 10.1097/ACM.0000000000001328. [ DOI ] [ PubMed ] [ Google Scholar ] 79. Bignami E.G., Berdini M., Panizzi M., et al. Artificial intelligence in sepsis management: an overview for clinicians. J Clin Med. 2025;14:286. doi: 10.3390/jcm14010286. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 80. Adams R., Henry K.E., Sridharan A., et al. Prospective, multi-site study of patient outcomes after implementation of the TREWS machine learning-based early warning system for sepsis. Nat Med. 2022;28:1455–1460. doi: 10.1038/s41591-022-01894-0. [ DOI ] [ PubMed ] [ Google Scholar ] 81. Arabi Y.M., Alsaawi A., Alzahrani M., et al. Electronic sepsis screening among patients admitted to hospital wards: a stepped-wedge cluster randomized trial. JAMA. 2025;333:763–773. doi: 10.1001/jama.2024.25982. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 82. Bhargava A., López-Espina C., Schmalz L., et al. FDA-authorized AI/ML tool for sepsis prediction: development and validation. NEJM AI. 2024;1 [ Google Scholar ] 83. Chiew C.J., Liu N., Tagami T., Wong T.H., Koh Z.X., Ong M.E.H. Heart rate variability based machine learning models for risk prediction of suspected sepsis patients in the emergency department. Medicine (Baltim) 2019;98 doi: 10.1097/MD.0000000000014197. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 84. Kwon J., Lee Y.R., Jung M.-S., et al. Deep-learning model for screening sepsis using electrocardiography. Scand J Trauma Resusc Emerg Med. 2021;29:145. doi: 10.1186/s13049-021-00953-8. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 85. Lundberg S.M., Lee S.-I. Proceedings of the 31st International Conference on Neural Information Processing Systems. Curran Associates Inc.; Red Hook, NY, USA: 2017. A unified approach to interpreting model predictions; pp. 4768–4777. [ Google Scholar ] 86. Chua M.T., Boon Y., Lee Z.Y., et al. The role of artificial intelligence in sepsis in the Emergency Department: a narrative review. Ann Transl Med. 2025;13:4. doi: 10.21037/atm-24-150. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 87. Collins G.S., Moons K.G.M., Dhiman P., et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385 doi: 10.1136/bmj-2023-078378. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 88. Moons K.G.M., Damen J.A.A., Kaul T., et al. PROBAST+AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods. BMJ. 2025;388 doi: 10.1136/bmj-2024-082505. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 89. Tharwat A. Classification assessment methods. Appl Comput Inform. 2021;17:168–192. [ Google Scholar ] 90. Saito T., Rehmsmeier M. The precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets. PLoS One. 2015;10 doi: 10.1371/journal.pone.0118432. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 91. Schwabe D., Becker K., Seyferth M., Klaß A., Schaeffter T. The METRIC-framework for assessing data quality for trustworthy AI in medicine: a systematic review. NPJ Digit Med. 2024;7:203. doi: 10.1038/s41746-024-01196-4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 92. Park J.Y., Hsu T.-C., Hu J.-R., et al. Predicting sepsis mortality in a population-based national database: machine learning approach. J Med Internet Res. 2022;24 doi: 10.2196/29982. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 93. Giannini H.M., Ginestra J.C., Chivers C., et al. A machine learning algorithm to predict severe sepsis and septic shock: development, implementation, and impact on clinical practice. Crit Care Med. 2019;47:1485–1492. doi: 10.1097/CCM.0000000000003891. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 94. Burdick H., Pino E., Gabel-Comeau D., et al. Effect of a sepsis prediction algorithm on patient mortality, length of stay and readmission: a prospective multicentre clinical outcomes evaluation of real-world patient data from US hospitals. BMJ Health Care Inform. 2020;27 doi: 10.1136/bmjhci-2019-100109. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 95. Brown S.M., Jones J., Kuttler K.G., Keddington R.K., Allen T.L., Haug P. Prospective evaluation of an automated method to identify patients with severe sepsis or septic shock in the emergency department. BMC Emerg Med. 2016;16:31. doi: 10.1186/s12873-016-0095-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 96. Taneja I., Damhorst G.L., Lopez-Espina C., et al. Diagnostic and prognostic capabilities of a biomarker and EMR-based machine learning algorithm for sepsis. Clin Transl Sci. 2021;14:1578–1589. doi: 10.1111/cts.13030. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 97. Gordon A.C., Alipanah-Lechner N., Bos L.D., et al. From ICU syndromes to ICU subphenotypes: consensus report and recommendations for developing precision medicine in the ICU. Am J Respir Crit Care Med. 2024;210:155–166. doi: 10.1164/rccm.202311-2086SO. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 98. Seymour C.W., Kennedy J.N., Wang S., et al. Derivation, validation, and potential treatment implications of novel clinical phenotypes for sepsis. JAMA. 2019;321:2003–2017. doi: 10.1001/jama.2019.5791. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 99. Bhavani S.V., Carey K.A., Gilbert E.R., Afshar M., Verhoef P.A., Churpek M.M. Identifying novel sepsis subphenotypes using temperature trajectories. Am J Respir Crit Care Med. 2019;200:327–335. doi: 10.1164/rccm.201806-1197OC. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 100. Sinha P., Kerchberger V.E., Willmore A., et al. Identifying molecular phenotypes in sepsis: an analysis of two prospective observational cohorts and secondary analysis of two randomised controlled trials. Lancet Respir Med. 2023;11:965–974. doi: 10.1016/S2213-2600(23)00237-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 101. DeMerle K.M., Kennedy J.N., Chang C.-C.H., et al. Identification of a hyperinflammatory sepsis phenotype using protein biomarker and clinical data in the ProCESS randomized trial. Sci Rep. 2024;14:6234. doi: 10.1038/s41598-024-55667-5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 102. Sinha P., Calfee C.S., Cherian S., et al. Prevalence of phenotypes of acute respiratory distress syndrome in critically ill patients with COVID-19: a prospective observational study. Lancet Respir Med. 2020;8:1209–1218. doi: 10.1016/S2213-2600(20)30366-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 103. Antcliffe D.B., Burnham K.L., Al-Beidh F., et al. Transcriptomic signatures in sepsis and a differential response to steroids. From the VANISH randomized trial. Am J Respir Crit Care Med. 2019;199:980–986. doi: 10.1164/rccm.201807-1419OC. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 104. Scicluna B.P., van Vught L.A., Zwinderman A.H., et al. Classification of patients with sepsis according to blood genomic endotype: a prospective cohort study. Lancet Respir Med. 2017;5:816–826. doi: 10.1016/S2213-2600(17)30294-1. [ DOI ] [ PubMed ] [ Google Scholar ] 105. van Amstel R.B.E., Bartek B., Vlaar A.P.J., et al. Temporal transitions of the hyperinflammatory and hypoinflammatory phenotypes in critical illness. Am J Respir Crit Care Med. 2025;211:347–356. doi: 10.1164/rccm.202406-1241OC. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 106. Cajander S., Kox M., Scicluna B.P., et al. Profiling the dysregulated immune response in sepsis: overcoming challenges to achieve the goal of precision medicine. Lancet Respir Med. 2024;12:305–322. doi: 10.1016/S2213-2600(23)00330-2. [ DOI ] [ PubMed ] [ Google Scholar ] 107. Bourgoin P., Soliveres T., Barbaresi A., et al. CD169 and CD64 could help differentiate bacterial from CoVID-19 or other viral infections in the Emergency Department. Cytometry A. 2021;99:435–445. doi: 10.1002/cyto.a.24314. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 108. Ait Belkacem I., Mossadegh-Keller N., Bourgoin P., et al. Cell analysis from dried blood spots: new opportunities in immunology, hematology, and infectious diseases. Adv Sci Weinh Baden-Wurtt Ger. 2021;8 doi: 10.1002/advs.202100323. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 109. Daix T., Guerin E., Tavernier E., et al. Multicentric standardized flow cytometry routine assessment of patients with sepsis to predict clinical worsening. Chest. 2018;154:617–627. doi: 10.1016/j.chest.2018.03.058. [ DOI ] [ PubMed ] [ Google Scholar ] 110. Venet F., Textoris J., Blein S., et al. Immune profiling demonstrates a common immune signature of delayed acquired immunodeficiency in patients with various etiologies of severe injury∗. Crit Care Med. 2022;50:565–575. doi: 10.1097/CCM.0000000000005270. [ DOI ] [ PubMed ] [ Google Scholar ] 111. Simon P.M., Delude R.L., Lee M., et al. Duration and magnitude of hypotension and monocyte deactivation in patients with community-acquired pneumonia. Shock. 2011;36:553–559. doi: 10.1097/SHK.0b013e318235331e. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 112. Minutolo A., Petrone V., Fanelli M., et al. High CD169 Monocyte/Lymphocyte ratio reflects immunophenotype disruption and oxygen need in COVID-19 patients. Pathogens. 2021;10:1639. doi: 10.3390/pathogens10121639. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 113. Lafon T., Chapuis N., Guerin E., et al. Along with PaO2/FiO2 ratio and lymphopenia, low HLA-DR monocytes are the only additional parameter that independently predicts the clinical course of undifferentiated SARS-CoV-2 patients in emergency departments. J Leukoc Biol. 2024;115:1131–1142. doi: 10.1093/jleuko/qiae022. [ DOI ] [ PubMed ] [ Google Scholar ] 114. Raimbault S., Monneret G., Gossez M., et al. Elevated monocyte HLA-DR in pediatric secondary hemophagocytic lymphohistiocytosis: a retrospective study. Front Immunol. 2023;14 doi: 10.3389/fimmu.2023.1286749. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 115. Remy S., Gossez M., Belot A., et al. Massive increase in monocyte HLA-DR expression can be used to discriminate between septic shock and hemophagocytic lymphohistiocytosis-induced shock. Crit Care. 2018;22:213. doi: 10.1186/s13054-018-2146-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 116. Leventogiannis K., Kyriazopoulou E., Antonakos N., et al. Toward personalized immunotherapy in sepsis: the PROVIDE randomized clinical trial. Cell Rep Med. 2022;3 doi: 10.1016/j.xcrm.2022.100817. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 117. Haem Rahimi M., Daniel S., Venet F., et al. Fully automated interferon-γ release assay to monitor antigen-independent T cell functionality: a proof of concept study in septic shock. Cytokine. 2023;169 doi: 10.1016/j.cyto.2023.156263. [ DOI ] [ PubMed ] [ Google Scholar ] 118. Loi M.V., Sultana R., Nguyen T.M., Tia S.T., Lee J.H., O'Connor D. The diagnostic utility of host RNA biosignatures in adult patients with sepsis: a systematic review and meta-analysis. Crit Care Explor. 2025;7 doi: 10.1097/CCE.0000000000001212. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 119. Cheng Y., Peng H., Chen Q., Xu L., Qin L. Machine learning-based transcriptmics analysis reveals BMX, GRB10, and GADD45A as crucial biomarkers and therapeutic targets in sepsis. Front Pharmacol. 2025;16 doi: 10.3389/fphar.2025.1576467. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 120. Galtung N., Diehl-Wiesenecker E., Lehmann D., et al. Prospective validation of a transcriptomic severity classifier among patients with suspected acute infection and sepsis in the emergency department. Eur J Emerg Med. 2022;29:357–365. doi: 10.1097/MEJ.0000000000000931. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 121. Yadgarov M.Y., Landoni G., Berikashvili L.B., et al. Early detection of sepsis using machine learning algorithms: a systematic review and network meta-analysis. Front Med. 2024;11 doi: 10.3389/fmed.2024.1491358. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 122. Bomrah S., Uddin M., Upadhyay U., et al. A scoping review of machine learning for sepsis prediction- feature engineering strategies and model performance: a step towards explainability. Crit Care. 2024;28:180. doi: 10.1186/s13054-024-04948-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 123. Tsakiroglou M., Evans A., Pirmohamed M. Leveraging transcriptomics for precision diagnosis: lessons learned from cancer and sepsis. Front Genet. 2023;14 doi: 10.3389/fgene.2023.1100352. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 124. Aneman A., Vieillard-Baron A. Cardiac dysfunction in sepsis. Intensive Care Med. 2016;42:2073–2076. doi: 10.1007/s00134-016-4503-4. [ DOI ] [ PubMed ] [ Google Scholar ] 125. Cecconi M., De Backer D., Antonelli M., et al. Consensus on circulatory shock and hemodynamic monitoring. Task force of the European Society of Intensive Care Medicine. Intensive Care Med. 2014;40:1795–1815. doi: 10.1007/s00134-014-3525-z. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 126. Volpicelli G., Lamorte A., Tullio M., et al. Point-of-care multiorgan ultrasonography for the evaluation of undifferentiated hypotension in the emergency department. Intensive Care Med. 2013;39:1290–1298. doi: 10.1007/s00134-013-2919-7. [ DOI ] [ PubMed ] [ Google Scholar ] 127. Geri G., Vignon P., Aubry A., et al. Cardiovascular clusters in septic shock combining clinical and echocardiographic parameters: a post hoc analysis. Intensive Care Med. 2019;45:657–667. doi: 10.1007/s00134-019-05596-z. [ DOI ] [ PubMed ] [ Google Scholar ] 128. Lafon T., Appert A., Hadj M., et al. Comparative early hemodynamic profiles in patients presenting to the emergency department with septic and nonseptic acute circulatory failure using focused echocardiography. Shock. 2020;53:695–700. doi: 10.1097/SHK.0000000000001449. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 129. Francois B., Lambden S., Fivez T., et al. Prospective evaluation of the efficacy, safety, and optimal biomarker enrichment strategy for nangibotide, a TREM-1 inhibitor, in patients with septic shock (ASTONISH): a double-blind, randomised, controlled, phase 2b trial. Lancet Respir Med. 2023;11:894–904. doi: 10.1016/S2213-2600(23)00158-3. [ DOI ] [ PubMed ] [ Google Scholar ] 130. Laterre P.F., Pickkers P., Marx G., et al. Safety and tolerability of non-neutralizing adrenomedullin antibody adrecizumab (HAM8101) in septic shock patients: the AdrenOSS-2 phase 2a biomarker-guided trial. Intensive Care Med. 2021;47:1284–1294. doi: 10.1007/s00134-021-06537-5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 131. Deniau B., Blet A., Santos K., et al. Inhibition of circulating dipeptidyl-peptidase 3 restores cardiac function in a sepsis-induced model in rats: a proof of concept study. PLoS One. 2020;15 doi: 10.1371/journal.pone.0238039. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 132. Garcia B., Ter Schiphorst B., Santos K., et al. Inhibition of circulating dipeptidyl-peptidase 3 by procizumab in experimental septic shock reduces catecholamine exposure and myocardial injury. Intensive Care Med Exp. 2024;12:53. doi: 10.1186/s40635-024-00638-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Figure and Tables mmc1.docx (62.6KB, docx) Articles from eClinicalMedicine are provided here courtesy of Elsevier ACTIONS View on publisher site PDF (1.5 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top