Recent advances in screening, diagnosis, prognosis and personalized treatment of sepsis

Abstract

Sepsis is a multi-system physiological syndrome defined as a dysregulated host response to infection. Despite recent advances clinicians continue to depend on clinically or physiologically defined syndromes as poor classifiers and therefore fail to accurately predict outcome or responses to treatment. Recent studies suggest that advances in screening, diagnosis, prognosis, and treatment of sepsis can be used to tailor immunomodulatory therapies. In the present narrative review, we critically analyze important academic literature published between January 2025 and June 2026 with a focus on epidemiology, screening, diagnosis, prognosis and personalized management. We believe that sepsis care will soon be defined by the convergence of rapid diagnostics, multi-omics profiling, and artificial intelligence, enabling a shift from generalized treatment approaches toward truly personalized, dynamic patient management. While the scientific and technological foundation for this transformation is emerging rapidly, clinical validation using innovative (adaptive) clinical trial design is required. Successful implementation within routine care will require multidisciplinary collaboration, robust infrastructure, clinician and patient engagement, and healthcare systems capable of integrating innovation into practice. The coming years may see a paradigm shift in sepsis management comparable to advances observed in personalized management of malignancies.

1. Introduction

Sepsis is a multi-system physiological syndrome defined as a dysregulated host response to infection [1]. For clinical use, at least two newly increased points on the Sequential Organ Failure Assessment (SOFA) score, over and above the usual baseline of a patient, represent deterioration in organ function that distinguishes sepsis from other infections. This score incorporates six organ systems (brain, cardiovascular, respiratory, hepatic, renal, and coagulation) and was updated in 2025. The current sepsis definition allows identification of heterogeneous patient populations varying widely in clinical syndrome, triggering pathogen, and prognosis. Nonetheless, most of these patients with sepsis are immediately, aggressively, and uniformly treated, most often including the use of intravenous fluids and broad-spectrum antibiotics. Evidence continues to mount, however, that immediate antibiotics are only critical for patients with septic shock or multiorgan dysfunction, while patients with single organ dysfunction without shock can safely tolerate short delays until antibiotics, allowing for diagnostic testing.

To improve sepsis management, large-scale sepsis quality improvement initiatives such as the United Kingdom Sepsis Six programme (https://sepsistrust.org/wp-content/uploads/2024/07/Sepsis-Manual-7th-Edition-2024-V1.0.pdf), Surviving Sepsis Campaign bundles (https://www.sccm.org/survivingsepsiscampaign/guidelines-and-resources), Germany's sepsis quality initiative of the Federal Joint Committee, (G-BA [2]), and New York State's “Rory's Regulations” (https://www.endsepsis.org/sepsis-protocols/) have promoted bundle-based sepsis management to achieve benefits with regard to early recognition, timeliness of treatment, and survival outcomes. The observed outcome gains are characterized by coordinated, system-level activities that integrate training, real-time monitoring, and accountability. Nevertheless, clinicians are left with tools that focus on clinical signs and symptoms and laboratory-based diagnostic solutions that are imperfect.

Despite recent advances in immunobiology, identification of specific immune responses in sepsis (Fig. 1) is not yet broadly accessible to clinicians who continue to depend on clinically or physiologically defined syndromes as poor classifiers and therefore fail to accurately predict outcome or responses to treatment. Importantly, several recent studies suggest that personalized approaches can be used to tailor adjunctive immunomodulatory therapies in patients with sepsis (either by repurposing existing drugs or by developing new drugs targeting the restoration of immune homeostasis). A precision medicine approach therefore appears within reach.

Fig. 1.

The spectrum of immune responses to sepsis

While a recent seminar article elegantly unraveled the complexity of sepsis, the focus was put on patient management [1]. In the present narrative review, we therefore critically analyze important academic literature published in 2025 and 2026 with a focus on epidemiology, screening, diagnosis, prognosis and personalized management, as summarized in Fig. 2. We strongly believe that major advancements have occurred along the patient flow in the healthcare system during this time window; these deserve to be presented to a broad audience ranging from clinicians to laboratorians and clinical researchers.

Fig. 2.

Advances in the field of sepsis: From screening/diagnosis to prognosis and personalized management along the patient flow in the healthcare system. The present narrative review covers significant advances published in 2025 and 2026 (cut-off: June 2026)

*including a collection of patient data, among others vital signs, laboratory values, and also artificial intelligence tools such as Sepsis ImmunoScore (some of which may be used for prognosis as well).

2. Epidemiology

The most recent report on the global epidemiology of sepsis was published, assessing sepsis cases for the year 2021 [3]. Using Sepsis-3 definitions, the authors estimated a staggering 166 million sepsis cases and 21.4 million all-cause sepsis-related deaths globally, representing 31.5% of total global deaths. Sepsis-related deaths decreased between 1990 and 2019, followed by a surge in 2020 and 2021. Those aged 70 years and older had the highest sepsis-related mortality in 2021, with 9.3 million deaths. Sepsis-related deaths from infectious underlying causes decreased from 11.8 million in 1990 to 8.3 million in 2019, then increased markedly to 15.5 million in 2021 (led by bloodstream infections inclusive of HIV and malaria and lower respiratory infections inclusive of COVID-19); mortality due to non-infectious underlying causes, i.e. stroke, chronic obstructive pulmonary disease, and cirrhosis, of death increased from 4.7 million in 1990 to 5.8 million in 2021. The report identified a number of risk groups incl. 1) children under five with almost 2.9 million annual sepsis deaths and maternal sepsis with approx. 10.7% of global maternal deaths, and 2) low- and middle-income countries, i.e., sub-Saharan Africa, South Asia and Oceania, where poor health systems and limited access to early observation and treatment contribute to increased risks.

La Via [4] recently reported results of an extensive meta-analysis and Global Burden of Disease (GBD) data showing that the pooled global incidence per 100,000 persons ranged from 437 to 678 cases, with the corresponding mortality rates varying between 15% in high-income to 34% in low-income regions. Regional differences are pronounced, with incidence rates (cases/100,000 population) ranging from 38.5 in Australia to 42–48 in Europe and the USA, 85 in South America, and 170–190 in Asia and Africa. Although age-standardized incidence declined by 37% and mortality by 53% between 1990 and 2017, population growth and aging have partially offset these improvements.

Thus, the global burden of sepsis increased in 2020 and 2021, reversing progress from 1990, and sepsis mortality continues to increase as a complication of non-infectious causes of death. Population dynamics will continue to have a major impact on the epidemiology of sepsis.

3. The clinical conundrum: uncertainty, need for preventing deaths vs. antimicrobial stewardship

The Sepsis-3 definition (“life-threatening organ dysfunction caused by a dysregulated host response to infection”) provided a valuable new concept but implementation at the bedside remains a challenge [1]. To determine potential underlying infection and to determine the quality of host responses is not feasible easily. The combination of suspected infection plus changes in clinical scores (i.e. increase in SOFA score by 2 points or more) is used for identifying sepsis. Using this approach, a very heterogeneous group of patients that vary widely in their presenting syndromes, severities, prognoses, triggering pathogens, probability of true infection, and likelihood that organ dysfunction is due to infection versus some other cause can be identified. Recently, a revised SOFA score, termed SOFA-2 to include contemporary organ support treatments and new score thresholds, was found to describe organ dysfunction and be predictive of mortality in a large population of critically ill adults across multiple countries [5]. The quick SOFA (qSOFA) score, developed with the Sepsis-3 consensus was intended to help identify sepsis quickly, but demonstrated poor accuracy when used as the sole screening tool. Recall that SOFA/qSOFA are tools to risk stratify outcomes but do not help identify early or evolving illness. NEWS2 appears to provide an overall higher accuracy for patient screening in general wards or community hospitals [6].

While screening tools identify many patients as potentially septic, a sizable fraction of these is not infected or do not require very aggressive management such as high-volume fluid resuscitation. The very wide spectrum of illnesses captured by current sepsis screening mechanisms are also problematic because clinicians and regulators have been taught to treat all sepsis patients in a common fashion with immediate broad-spectrum antibiotics, and for those with hypotension and/or elevated serum lactate concentrations to get high volume intravenous fluids. Regulators and quality improvement advocates reinforce and accelerate this perception through public reporting and pay-for-performance metrics that penalize clinicians and hospitals for failing to administer these interventions within time limits of patients' qualifying as possible sepsis. Differentiating bacterial sepsis from a host of mimicking syndromes continues to be very difficult, particularly in the early hours of initial presentation (approx. ⅓ of patients treated with intravenous antibiotics are diagnosed with viral infections or non-infectious conditions, e.g. acute heart failure, obstructive lung disease exacerbations, cancer, thromboembolic disease, drug reactions, drug withdrawal, toxidromes, etc.). These patients are exposed to the potential risks of antibacterials without benefit. What clinicians need in these circumstances is the ability to sort out infected from uninfected patients and bacterial from non-bacterial infections quickly. Currently, this means that clinicians have to take time to work out these differences, potentially putting patients at risk with delays in treatment as the subset of patients needing rapid care is not always clear with insufficient diagnostic tools.

The diagnostic conundrum described above has been elegantly discussed by two clinical experts in a NEJM Clinical Decision Interactive Article [7] (Fig. 3).

Fig. 3.

Case vignette demonstrating the clinical conundrum and diagnostic uncertainty in patients presenting with potential sepsis, adopted from [7]

ED: emergency department.

Protocolized sepsis bundles such as the Severe Sepsis and Septic Shock Early Management (SEP-1) bundle have been developed to enforce improved sepsis care delivery. Of the individual sepsis bundle components, timely antibiotic administration has the most robust effect on mortality; the effect of fluid resuscitation on mortality is discussed controversially as inappropriate fluid administration can cause harm to many patients. The overall effect of sepsis bundle compliance on mortality remains uncertain, as most of the existing evidence is observational and disentangling confounding from the true effect of SEP-1 has proved difficult. Ford [8] recently reviewed studies of adults with sepsis that included 3- or 6-h sepsis bundles defined by SEP-1 specifications and found no high- or moderate-level evidence to suggest that SEP-1 compliance was associated with improved mortality (except for Medicare beneficiaries and patients with septic shock). However, the rigid all-or-nothing mandate has been criticized for disempowering clinicians and preventing them from providing personalized care. Thus, the potential benefits of mandated standardization of sepsis treatment must be weighed against the potential harm to patients and the greater health care system. For example, pressure to comply with SEP-1's time requirements in patients with noninfectious sepsis mimics, such as methamphetamine intoxication or pancreatitis, may lead to increased broad-spectrum antimicrobial use and increased risk for adverse events, such as acute kidney injury, delirium, or contracting a nosocomial or multidrug-resistant organism [9]. Furthermore, overuse of the sepsis bundle may lead to increased health system resource utilization and antimicrobial resistance [10].

4. Screening

The recently published sepsis guidelines from the Surviving Sepsis Campaign [11] continue to recommend standard operating procedures for sepsis screening. While not formally updated, recent new randomized controlled trials were referenced as further support for recommendations. Among 872 patients, implementation of the 1-h bundle in the emergency department (ED) was associated with shorter time to antibiotics and an uncertain effect on mortality [12]. Furthermore, the implementation of an electronic alert system, staff education, and feedback resulted in lower 90-day in-hospital mortality among patients with and without sepsis in a large randomized clinical trial involving 60,055 patients [13].

As early detection and management of sepsis are crucial for patient survival, i.e., in the ED, numerous sepsis alert systems have been developed. In brief, these include electronic clinical decision support tools and cell morphology-based laboratory assays aimed at identifying patients at increased risk of sepsis.

• a)Electronic health record and artificial intelligence-driven tools

Clinical decision support tools and Sepsis ImmunoScore (Prenosis, IL. USA)

In a systematic review and meta-analysis including 22 studies with 19,580 patients [14], sepsis alert systems were associated with better outcomes and adherence with sepsis management in EDs; of note, lower mortality, shorter hospital stay, and improved sepsis-bundle adherence (shorter time to fluid administration, blood culture, antibiotic administration, and lactate measurement) were reported. These findings suggest that sepsis alert systems in EDs were associated with improved patient outcomes and greater adherence to sepsis management protocols. Recently, Ostermayer [15] compared the clinical outcomes of sepsis patients when both an augmented systemic inflammatory response syndrome (SIRS+) and the Epic Sepsis Predictive Model version 1 (ESPMv1; Epic Systems, USA) alert were activated among more than 800,000 patients in the ED at two county hospitals. SIRS+ alerts had a sensitivity of 14.25%, specificity of 96.1%, positive predictive value of 7.8%, and negative predictive value of 98%. Similarly, ESPMv1 had a sensitivity of 15.6%, specificity of 95.4%, positive predictive value of 8.1%, and negative predictive value of 98% for diagnosing sepsis. No statistical differences in time to antibiotics, time to blood culture draws, or time to intensive care unit (ICU) admission were observed. The authors concluded that both alerting systems had similarly poor diagnostic characteristics. Thus, while leveraging this type of clinical decision support could dramatically improve the quality and consistency of sepsis management, reality has thus far fallen far short of expectation [16].

One recently launched innovative solution is the Sepsis ImmunoScore (Prenosis, IL, USA). The Sepsis ImmunoScore is an FDA-cleared risk stratification software tool that uses artificial intelligence (AI) to aid in the identification of patients likely to have or progress to sepsis within 24 h of assessment [17]. ImmunoScore analyzes 22 different parameters, combining biological data including biomarkers, clinical lab test results, vital signs, and demographics to assist providers in anticipating and diagnosing sepsis using the electronic health record (EHR). The scores (low, medium, high, and very high sepsis risk) showed overall diagnostic accuracy with an area under the curve (AUC) of 0.81 for predicting adjudicated sepsis. Most recently, it was investigated whether added transparency to the AI tool would improve decision making in a clinical setting by using a structured assessment among providers and nurses [18]. Individualized feature importance scores were generated to differentiate between ImmunoScore predictions for specific clinical inputs and the overall average predictions. Participants correctly interpreted importance scores in 98% of real clinical scenarios and unanimously reported that the use improved their understanding of ImmunoScore. Thus, health care providers appear to value transparency into AI algorithms to improve tool adoption and clinical utility. However, future work is warranted to establish the generalizability of ImmunoScore to other settings and to assess its impact on clinical decision-making, sepsis care, and associated resource utilization and costs.

• b)Cell morphology-based diagnostics

Monocyte Distribution Width (MDW, Beckman Coulter, CA, USA) and IntelliSep (Cytovale, CA, USA)

Advanced host-response tests for the prognosis of the patient's condition have been developed based on the biophysical properties of blood cells. Whereas MDW [19] relies on changes in shape of monocytes following activation, which increases variability in volume and size, IntelliSep analyzes the shape of white blood cells as they are stretched through a microfluidic channel [20]. MDW can be generated automatically during a routine Complete Blood Count (CBC) with differential test, provided the lab's instrument is equipped with the specific software from the manufacturer. In contrast, IntelliSep requires specific instrumentation and a specific blood draw. Most recently, among adult ED patients with suspected sepsis, MDW was measured along with presepsin. MDW, presepsin, and their combination achieved AUCs ranging from 0.52 to 0.65 for sepsis diagnosis. MDW and presepsin had sensitivities of 90.6 and 89.8% for the diagnosis of sepsis, whereas the combination improved sensitivity to 98.4% [21]. Thomas [22] recently reported results of the first year of experience following the implementation of IntelliSep as part of a sepsis health initiative. Post-implementation, sepsis-associated mortality dropped significantly from 10.9 to 6.6%, and there was also a 0.76-day reduction in mean hospital length of stay among sepsis diagnosis-related group (DRG) patients (P < 0.05); blood culture utilization also fell from 50.8 to 45.7%, driven by a reduction among patients receiving a Band 1 IntelliSep score. Similarly, Sheybani [23] evaluated IntelliSep's ability to affect decision-making, risk stratification, and diagnostic efficacy in the ED. In a case-vignette-based, randomized, multisite decision-impact study involving 52 attending physicians at 3 health care institutions, 100 real-world case vignettes were analyzed with or without IntelliSep results. The test was found to support or change/augment clinician decisions in 86% of cases. While paper-based, this study demonstrates that IntelliSep could improve clinician decisions in the ED. However, interventional trials are warranted to demonstrate the true impact as a single intervention for early diagnosis of sepsis. Of note, IntelliSep only predicts the risk of sepsis but not of severe illness other than sepsis, which may cause clinicians to overlook non-sepsis at-risk patients.

Overall, it appears that the high sensitivity of tests based on cell morphology comes with low specificity, which must be considered when designing optional patient flows, i.e., using these as screening tools. Both the electronic record and cell morphology-based systems hold promise for enhancing emergency department responses to sepsis, potentially leading to better patient outcomes. However, consideration must be given to the potential for false-positive alarms. AI for sepsis management in the ED may assist in the development and validation of these tools (recently reviewed by de Vries [24]). As most of these tools have been validated only in observational studies, limiting their clinical utility, future research should focus on better understanding the real-world impacts of sepsis alert systems, particularly focusing on the rate of false positives and their associated consequences, such as resource overutilization and the overprescription of antibiotics.

5. Diagnosis

The diagnosis of sepsis is complex, as under the current Sepsis-3 definition organ dysfunction and an underlying infectious cause must be determined. In clinical practice, clinical scores, biomarkers (host response), and pathogen identification (ID) tests are most commonly used but are far from accurate (see above Clinical Conundrum chapter).

Advances have been reported in four key areas for rapid diagnostics of bloodstream infections and sepsis. These include four different approaches ranging from pathogen ID directly from blood, pathogen ID from positive blood culture to host response tests and the combination of pathogen ID plus host response as summarized in Table 1. Importantly, while solutions 1–3 are tools for the diagnosis of bloodstream infections, only pathogen identification combined with host-response tests for detecting organ dysfunction (i.e., severity markers) should be considered true sepsis diagnostic tools. Figure 4 provides an overview of the differences between sepsis and bloodstream infection, as well as the clinical microbiology workflow in the laboratory.

Table 1.

Key characteristics of advanced solutions for the diagnosis of sepsis

Solution	Technologies*	Target(s)/processing step	Time to result	Impact	Comments
Pathogen ID/AST direct from blood	PCR, NGS, Magnetic resonance	Bacteria, viruses, fungi*, many (any) resistance marker	Fast to moderate	High but still experimental	Most commonly without antimicrobial susceptibility testing
Pathogen identification/antimicrobial susceptibility testing from positive blood culture	PCR, MALDI-ToF mass spectrometry	Bacteria, fungi, few resistance markers	Culture-based: Slow (growth)/very slow (AST) (positive growth needed) Molecular: Moderately slow (growth required)	Moderate as often too slow	With or without antimicrobial susceptibility testing
Host response	Direct from blood: Protein-based, mRNA based (PCR, LAMP, RNA-sequencing)	Host response to classes of pathogens can be detected, e.g. bacteria, viruses, non-infectious etiologies, plus severity of the condition	Fast	High
Pathogen identification and host response combined	DNA- and RNA-sequencing	As above	Moderately slow	High but experimental
Technological advances	Multiple	Focus on sample preparation	Speeding up time to result	Minor as only supportive

*PCR: polymerase-chain reaction, NGS: next-generation sequencing, LAMP: loop-amplified isothermal amplification, MALDI-ToF: matrix-assisted laser desorption/ionization-time of flight.

Fig. 4.

Sepsis definition and the microbiology workflow in the laboratory (arrow color indicates speed: red = slow, green = fast; PCR, polymerase-chain reaction; NGS, next-generation sequencing)

*time to result dependent on further sub-culturing.

The diagnostic test most commonly associated with bacterial sepsis is the blood culture. Bacterial bloodstream infections are common in sepsis, but bacteremia is not the same as sepsis (Fig. 4). In fact, only 38–69% of septic patients have bacterial bloodstream infections [25].

• a)Pathogen identification starting from positive blood culture bottles

Many tests can be used to establish a diagnosis of acute infection, but the diagnostic test that is most often associated with bacterial sepsis is blood culture. While the long turnaround times (24–72+ hours) for pathogen identification remain a problem for clinical management, major improvements have been achieved through molecular diagnostics. Recently, advances have been reported that focus on reducing time to result and increasing sensitivity using multiple Food and Drug Administration (FDA)-cleared molecular platforms. The 2025 American Society of Microbiology evidence-based laboratory medicine practice guidelines for the diagnosis of bloodstream infections using rapid tests provide an excellent summary of these technologies [26]. The expert committee recommended using rapid diagnostic tests, importantly, combined with active communication to decrease the time to targeted therapy and patient length of stay; the panel also supported using rapid tests to impact mortality (despite low evidence). While costs of molecular detection methods are significant, only a minority of (positive) samples require the downstream use of these technologies. One of the first clinical implementation studies of Oxford Nanopore metagenomic sequencing directly from routine blood cultures was recently published [27]. The method achieved 97% sensitivity and 94% specificity for species identification compared with that of routine culture and MALDI-ToF-based diagnostics; both sensitivity and specificity increased to 100% after adjudication of plausible additional infections. Species identification results were delivered approximately 10 h earlier than routine diagnostic methods, and antimicrobial resistance testing results were delivered up to 20 h earlier than current antimicrobial susceptibility testing. Thus, metagenomic sequencing has the potential to rapidly and comprehensively detect pathogens and antimicrobial resistance (AMR) in bloodstream infections from blood cultures. Nonetheless, these improvements do not shorten the time required for initial pathogen growth in blood culture bottles.

• b)Pathogen identification direct from blood

An ideal strategy remains direct from blood or other direct-from-specimen techniques to provide rapid, more clinically actionable results. Blood is a complex specimen type, and pathogens tend to be present at extremely low concentrations, very few per milliliter, while present in a milieu of human cells that outnumber pathogens a million to one. Rapid diagnostics directly from whole blood, rather than from positive blood cultures, were reported a decade or more ago (SeptiFast and IRIDICA [28]), with the promise of earlier actionable insight without waiting for culture results. While demonstrating increased sensitivity, the economic reality proved difficult as every patient sample must be tested even though most will ultimately be negative. In this regard, several technically innovative solutions, SeptiFast, IRIDICA, and T2 Biosystems [29], demonstrated promise but failed to achieve sustained adoption due to technical as well as economic and operational hurdles. Traditional molecular approaches, consisting of lysing samples and searching for microbial nucleic acids, have been described as “finding a needle in a haystack”. In the past year, we have seen some of the most promising technologies disappear while others have shown promise for reviving the field. Here, we provide an overview of the commercially available and emerging culture-independent technologies that hold promise for improving the early diagnosis of sepsis (summarized in Table 2).

Table 2.

Direct from blood pathogen-identification platforms for the diagnosis of bloodstream infections

Test	Targets	Technology	Turnaround time	Antimicrobial resistance detection	Instrument	Regulatory status
Sepsis-Q Real Time PCR (Genes 2 Me)	32 bacteria + viruses + fungi (20 in point-of-care version)	Multiplex RT-PCR**	∼2h	No	Proprietary OnePCR (POC*) Rapi-Q (high-throughput)	Commercially available
InfectID BSI (MicroBio)	20 bacteria +6 yeasts	SNP-qPCR*** + melting curve analysis	<3h	No	QIAGEN Rotor-Gene Q	Commercially available
UllCORE (Deepull)	50 bacteria, fungi + AMR* genes	Multiplex RT-PCR**	∼1h	Phenotypic	Proprietary UllCORE system	In development
RaPID/BSI (HelixBind)	21 bacteria, + fungi	Probes for dsDNA + PCR (16S/18S)	∼4h	Genotypic	Proprietary RaPID diagnostic platform	In development
Melio (Melio Tech)	Broad bacterial + fungal + viruses incl. emerging pathogens	DNA melting + acoustic isolation + machine learning- analysis	<6h	No	Standard quantitative PCR instrument?	In development
Vivid Dx	Broad panel (Raman spectral database)	Raman spectroscopy + machine learning-driven analysis	30 min (pathogen identification) 3h (AMR*)	Phenotypic	Proprietary Raman instrument	In development

*AMR: antimicrobial resistance testing.

**RT-PCR: real time-polymerase-chain reaction.

***SNP: single nucleotide polymorphism.

5.1. Commercially available platforms

In brief, the Sepsis-Q RT-Kit (Genes 2 Me) detects 32 pathogens, including bacteria, viruses, and fungi, using multiplex real time-polymerase chain reaction (RT-PCR) with a full turnaround time of 2 h. The sample-prep chemistry includes specialized reagents for lysis and host DNA depletion, as well as pathogen enrichment beads that enable efficient nucleic acid extraction. The test is available on a point-of-care and a high-throughput platform. Preliminary analytical data have been presented but peer-reviewed clinical validation studies are not yet available. InfectID BSI (MicroBio) identifies 26 pathogens in about 3 h using single-nucleotide polymorphisms from conserved regions of bacterial and fungal genomes via quantitative PCR (qPCR) and high-resolution melt analysis and is validated for use on the Qiagen Rotor Gene Q platform. Preliminary results showed superior accuracy compared to blood culture in 368 samples from emergency room patients; the clinical concordance was 100% [30].

5.2. Pre-commercial platforms

The UllCORE benchtop system (Deepull) is a cartridge-based test providing results for over 50 targets and selected antimicrobial resistance genes in about one hour. The underlying technology is multiplex RT-PCR, following sample prep and nucleic acid extraction of total microbial DNA from ∼8 mL of whole blood. Extraction of total microbial DNA enables detection of fastidious, non-viable, and cell-free microbial DNA, all of which are non-culturable by traditional blood culture methods. Peer-reviewed clinical validation data are not yet available.

RaPID/BSI (HelixBind) is a cartridge-based test performed on the sample-to-answer RaPID platform. The targets cover 21 of the most common bacteria and fungi with species-level identification. Gamma-modified peptide nucleic acids bind to the helical structure of double-stranded DNA rather than hybridizing to single-stranded DNA which promises to improve accuracy. Sample preparation involves lysis of human cells, followed by electrostatic removal of human genetic material, cell-free microbial DNA, and downstream inhibitors. Preliminary clinical validation studies demonstrated >95% sensitivity and >90% concordance with blood cultures [31].

5.3. In development platforms

Universal Digital High-Resolution Melting (U-dHRM), combined with acoustic cell isolation and AI classification, is used by Melio for direct-from-blood isolation and pathogen ID. DNA is melted at temperatures between 50 and 90 °C, creating unique sequence-dependent changes that can be detected with special dyes. Early data suggest the database can detect a broad range of bacteria, viruses, and emerging pathogens. Compared with blood culture, U-dHRM demonstrated 100% concordance and 88% concordance with clinical adjudication in a pilot study of 17 whole blood samples from pediatric patients. More recently, U-dHRM was reported to detect pathogenic molds in 73% of 30 samples classified as invasive mold infections, including mixed infections [32].

Raman spectroscopy, rather than PCR, was proposed by Vivid Dx for the diagnosis of bloodstream infection. By capturing changes induced by the interaction of light with cellular molecules a distinct molecular fingerprint of proteins, DNA, and other molecules is generated. Each Raman spectrum can be used for organism identification and antimicrobial resistance based on AI algorithms. Vivid Dx's platform is designed to provide identification results within 30 min and antimicrobial susceptibility (AST) results within 3 h. We are not aware of published reports regarding the accuracy of the technology.

5.4. Rapid AST

The jury is still out on whether rapid diagnostic tests (with or without antimicrobial susceptibility testing) and antimicrobial stewardship improve management compared to the standard of care. Most recently, the FAST randomized multicenter clinical trial reported negative results. The use of rapid antimicrobial susceptibility testing directly from blood culture bottles that were positive for Gram-negative bacteria did not improve clinical outcomes at day 30 compared with standard care [33]. In contrast, in the ADAPT-Sepsis trial [34] that randomized patients to either standard care, daily C-reactive protein (CRP), or daily procalcitonin (PCT) to promote antibiotic discontinuation, the PCT-guided group was found to have a significant reduction in total antibiotic duration and was noninferior with regard to 28-day all-cause mortality; the CRP-guided group did not show a reduction in antibiotic duration and was inferior to the control group with regard to mortality.

5.5. Next-generation sequencing

Metagenomic next-generation sequencing (NGS) enables simultaneous, hypothesis-free detection of a broad array of pathogens including bacteria, viruses, fungi, and parasites. Clinical applications and implementation challenges in infectious diseases, both overall and specifically in bloodstream infections, have recently been reviewed [35–37]. Unlike culture and (targeted) molecular solutions, NGS can identify novel, fastidious, polymicrobial infections and antimicrobial resistance genes. However, interpretation of results (true pathogen, contaminant, or colonization) is complex. Lastly, the implementation of NGS remains slow, and hurdles remain, e.g., time to results, identification of patient groups/clinical settings most benefiting from NGS, and cost.

Recently, a prospective, observational, multicenter study (Next GeneSiS) in Germany compared the positivity rates of NGS-based identification of causative pathogens with those of blood culture in almost 500 patients with sepsis or septic shock [38]. Within the first three days after sepsis onset, the positivity rate of NGS-based diagnostics was 70% compared to 19% for blood culture. NGS results were evaluated by an expert panel as plausible in 98% of cases; additional knowledge from NGS results would have led to adaptations in anti-infective treatment in 32% of patients. Potentially inadequately treated NGS positive/blood culture-negative patients showed worse outcomes.

Importantly, the first randomized controlled trial in sepsis, DigiSep, investigated the impact of NGS in addition to the standard of care across 24 intensive care units in Germany [39]. While the primary endpoint, improved Desirability of Outcome Ranking/Response Adjusted for Duration of Antibiotic Risk (DOOR/RADAR) scores, was not achieved, secondary endpoints were improved, including a reduced duration of mechanical ventilation, faster shock resolution, and improved 90-day health-related quality of life.

At this time, we are unable to make conclusive statements about the impact of NGS integration on sepsis workflows. NGS will not solve the problem of sepsis identification, but, together with other solutions (host response, see below), may improve overall identification and management. Moving forward, a framework for successful implementation should include, at a minimum, advances in technology (automation, bioinformatics), translation (standardized protocols and integration into clinical and laboratory workflows), training of genomics staff (clinicians, microbiologists, and data scientists) and overall funding [37, 40].

• c)Host response tests

The clinical conundrum described above underlines the need for rapid diagnostic tests that – early in the clinical course of sepsis – allow a) differentiation of sterile inflammation and sepsis, b) differentiation of bacterial from viral infection, and c) predict the severity of the condition. Answers to these questions, often discussed as the two axes of sepsis (presence of an infection and its severity), have clinically actionable consequences, including initiation or withdrawal of antimicrobial therapy, follow-up patient management decisions (e.g. ordering of imaging, syndromic pathogen ID panels, other laboratory tests) and decisions regarding patient disposition. Until recently, no such diagnostic tests were available but rapid host response signatures showed promise in potentially improving treatment decisions in the early phase of sepsis and improving the overall outcome of patients. The recently updated sepsis guidelines from the Surviving Sepsis Campaign mention host response testing but the authors concluded that “there is insufficient evidence to make a recommendation regarding use of novel rapid host response test” [11]. Two manufacturers have launched molecular host response tests for sepsis based on multi-markers mRNA signatures (transcriptomics). These claim to aid in separating infectious from non-infectious inflammation (SeptiCyte RAPID, Immunexpress, WA, USA) or aid in differentiating infection types (bacterial, viral, non-infectious) and predicting disease severity (including sepsis) (TriVerity, Inflammatix, CA, USA) [41]. Of interest, both tests generate results that are presented in different risk bands with each band having its own performance characteristics expressed in likelihood ratios. This innovative presentation allows to demonstrate risk characteristics from very low to very high rather than single dichotomous (abnormal/normal, positive/negative) presentations with inferior accuracy. Table 3 provides an overview of host response tests for the screening and diagnosis of sepsis.

Table 3.

Key characteristics of host response tests for the screening and diagnosis of sepsis

Test	Intended use/(targets)	Rule-out accuracy	Rule-in accuracy	Sample type and vial	Time to result (min)	Instrument
Screening
MDW (Beckman Coulter)	Risk of sepsis using monocyte distribution width	LR* 0.4 for non-sepsis	LR* 2.6 for sepsis	EDTA blood vacutainer inserted into instrument	∼60	Unicel DxH 800/900 Coulter System (large)
IntelliSep (Cytovale)	Probability of sepsis (multi-cell morphology)	LR* 0.1 for sepsis in lowest band	LR* 2.8 for sepsis in highest band	Whole blood pipetted into instrument	∼30	Sample preparation and image visualization modules (large)
Likelihood of sepsis
SeptiCyte RAPID (Immun-express)	Probability of sepsis (2 host RNAs)	10% probability of sepsis for lowest band	80.6% probability of sepsis for highest band	PAXgene RNA blood vacutainer inserted into cartridge	60	BioCartis Idylla (large)
Infectious (bacterial vs. viral) vs. non-infectious etiology plus severity
TriVerity (Inflammatix)	Presence, type, and severity of infection (29 host RNAs)	LR* 0.1/0.1 for lowest bacterial/viral band; LR 0.2 for lowest illness severity band	LR* 8.0/40.9 for highest bacterial/viral band; LR 11.3 for illness severity band	PAXgene RNA blood vacutainer inserted into cartridge	∼30	Inflammatix Myrna (small)

*LR: likelihood ratio.

5.6. SeptiCyte RAPID (Immunexpress)

SeptiCyte® RAPID is a RNA-based host response assay to differentiate sterile inflammation from sepsis in patients with suspected sepsis within the first 24 h of their ICU stay. SeptiCyte RAPID was launched with a 1h turnaround time measuring two host mRNAs on BioCartis' Idylla platform generating quantitative SeptiScores (separated into four bands indicating either sterile inflammation or sepsis). AUCs for the discrimination of sepsis from non-infectious systemic inflammation in retrospective studies using Sepsis-2 criteria ranged between 0.7 and 0.9 [42]. More recently, von der Forst [43] investigated the use of SeptiCyte RAPID to differentiate sepsis from inflammation in patients after abdominal surgery in a single center. Using clinical adjudication as the gold standard (42% patients were categorized as “inflammation”, 58% as “sepsis”) septic patients showed significantly higher mean SeptiScores with AUCs between 0.71 and 0.80 for the discrimination of inflammation and sepsis. SeptiScore results were significantly higher in the blood culture positive compared to the blood culture negative group. The authors concluded that clinical implementation of SeptiCyte RAPID for this patient cohort would require integration into a comprehensive diagnostic algorithm and positive findings in large cohorts. A multicenter study in septic ICU patients found an AUC of 0.84 for differentiating sepsis from sterile systemic inflammation, significantly better than CRP but not different from PCT; positive predictive values for the identification of sepsis were 68% (band 3) and 92% (band 4) while band 1 predicted sterile inflammation in 100% of patients [44].

The rather narrow claim (ICU patients with suspected sepsis within 24h of admission), need for expensive capital equipment (BioCartis Idylla, not commonly available in the ICU/central laboratory) and the lack of findings from an interventional trial may make broad adoption of SeptiCyte RAPID difficult.

5.7. TriVerity (Inflammatix)

TriVerity, performed on the point-of-care Myrna instrument, uses isothermal amplification of 29 mRNAs in the blood of patients with suspected infection or suspected sepsis in the ED. The RNAs are read and interpreted by machine learning algorithms to determine the likelihoods of bacterial and viral infections, as well as the need for critical care interventions, within 7 days. Each of the three scores falls into one of five interpretation bands; band-specific likelihood ratios, sensitivities (for ruling out) and specificities (for ruling in) can be calculated. Results of the biological role of the 29 RNAs [45] as well as results of analytical validation were published [46]. In a recently published large registrational trial (SEPSIS-SHIELD, [47]) TriVerity distinguished bacterial infections from viral ones or non-infectious conditions that do not need antibiotic treatment and identified patients with incipient sepsis with an AUC of 0.91 for the diagnosis of bacterial infections; AUCs were significantly lower for CRP, PCT, or white blood cell count (WBC). The AUC for the detection of viral infection was 0.91. The severity score showed an AUC of 0.78 for predicting the need for ICU care within 7 days and allowed reclassification of risk for critical care interventions compared to clinical assessment (qSOFA) alone. Each of the three scores had rule-in specificity >92% and rule-out sensitivity >95%. Of interest, the combination of high severity and high bacterial scores allowed to rule-in but also to rule-out bacterial sepsis. In a small proof-of-concept trial among adult patients presenting to the ED with suspicion of infection, TriVerity demonstrated 95% rule-in specificity and 95% rule-out sensitivity for the diagnosis of bacterial infection (100% rule-in specificity and 92% rule-out sensitivity for the diagnosis of viral infection) [48].

Large interventional trials will need to determine where in the acute care patient flow the test should be implemented (early triage or later in the work-up?) and how it impacts current standard-of-care protocols (e.g. add-on or replacement of other testing?). Results of the interventional TIMED (pre/post design, NCT06637904) trial using SEP-1 bundle compliance and time to final ED disposition order as outcomes are anticipated.

• d)Clinical adjudication as the reference standard

One important feature in evaluating test performance for all the above tests is the selection of a validated reference standard; most importantly, as the novel test may show performance characteristics superior to the typical comparator, e.g., blood culture or single target biomarkers. Clinical adjudication has thus been proposed and even mandated by the FDA for host response tests for the diagnosis of infection [49–51]. Importantly, guidance must be provided to adjudicators who should be experts in the setting investigated, including infectious diseases and/or clinical microbiology, and must remain blinded to results of the test under investigation. Clinical information provided must be comprehensive throughout the patient journey through the healthcare setting.

In conclusion, results from pathogen detection tests and from multi-target host response signatures will likely need to be interpreted in combination (as exemplified in an early trial in ICU patients [52]). These results may not be stand-alone findings but could be entered into an all-encompassing data platform that also incorporates conventional laboratory values, imaging findings, results from continuous monitoring of vital parameters, and patient history (i.e. comorbidities and contextual factors). This could empower clinicians to rapidly infer the type, severity, and trajectory of sepsis for improved management and outcomes. Given the ever-increasing number of variables and results like these informing patient care, there will also likely be a significant role for AI to play in monitoring and concatenating these results in the context of all the patients' clinical variables.

6. Prognosis

Lactate, a metabolic byproduct of glycolysis, indicates tissue hypoperfusion and thus is closely associated with the severity and prognosis of sepsis [53]. For adults with possible, probable, or definite sepsis or septic shock, the Surviving Sepsis Guidelines “suggest” measuring blood lactate to guide resuscitation with values ≥4 mmol L⁻¹ triggering specific, aggressive intervention bundles specifically around maintaining adequate tissue perfusion [11]. Important advances in the prognosis of sepsis have been made using multi-marker RNA-based signatures, combined with machine learning-supported result generation. These can be divided into commercially available, FDA-cleared solutions such as the 29 RNA-signature TriVerity test (measuring the likelihood of need for ICU-level care within 7-days as described above) or advances in research awaiting translation into clinical use. Malic [54] recently reported a six-gene RNA signature of immune cell reprogramming termed Sepset (in development by Sepset Biosciences (https://www.sepset.ca/)). In conjunction with machine learning, high accuracy was reported for the prediction of clinical deterioration (higher in ICU compared to ED patients). In a small real-world evaluation of a point-of-care prototype instrument, the signature was able to predict clinical deterioration in patients with suspected sepsis at presentation (worsening of SOFA score or need for ICU admission) with 92% sensitivity and 89% specificity. While still in development, this report further supports the value of transcriptomic approaches for the prediction of sepsis severity. In this regard, Spottiswoode [55] investigated host and microbial factors associated with in-hospital sepsis mortality to build prognostic classifiers based on RNA sequencing. Among 321 critically ill adults with adjudicated sepsis, transcriptional profiling as well as proteomic and metagenomic analyses were conducted. Mortality was associated with increased expression of genes related to neutrophil degranulation, lower expression of genes related to T-cell signaling, and higher interleukin (IL)-8 levels. Mortality was also associated with greater microbial mass and greater bacterial relative dominance. Similar findings were observed in a broader group that also included patients with culture-negative sepsis or indeterminate sepsis status. An integrated host-microbe metagenomic classifier predicted sepsis mortality with an AUC of 0.79, similar to a host transcriptomic classifier (AUC 0.75); both classifiers were found to perform significantly better than a clinical score (Acute Physiology, Age, Chronic Health Evaluation III [APACHE] score). Combining host and microbial factors associated with mortality in critical illness may be a promising new approach to mortality prediction in sepsis.

7. Towards personalized treatment of sepsis

Various specific treatments of septic patients remain controversial despite extensive study, including basic concepts such as fluid-liberal, fluid restrictive/early vasopressor, and steroid containing treatment regimens. Additionally, hundreds of millions of dollars have been expended enrolling well over 30,000 patients in clinical trials to test and develop new immunomodulating agents, anti-inflammatory agents, and anti-endotoxin agents. Yet, aside from improvements in infection control and organ support, not a single agent has convincingly demonstrated consistent efficacy in clinical trials. In 2014, Opal [56] proposed 12 specific recommendations that, if implemented, could improve the outlook for developing new drugs for sepsis treatment. Among these, two recommendations involved biomarkers: a) develop biomarkers or response indicators to guide patient selection early in development (biomarkers should fit the mechanism of action of the drug or device to select responsive patients and improve success), and b) improved biomarkers or surrogate endpoint measures are greatly needed to define optimal patient cohorts (real time, dynamic genomics methods are needed to immunophenotype patients likely to benefit). It is now appreciated that the heterogeneous nature of sepsis likely contributed to the lack of progress on the proposed recommendations. In this regard, corticosteroids are currently recommended with moderate evidence only as adjunctive therapy for adults with septic shock and ongoing requirement for vasopressor therapy [11]; this recommendation is further supported by a recent Cochrane review that found only moderate-certainty evidence indicating that corticosteroids probably reduce 28‐day, 90‐day and in-hospital mortality [57].

7.1. Host directed therapies and patient subgroups (endotypes)

Advances in our understanding of the immunopathology of sepsis have kicked-off a new chapter of adjuvant immune (host-directed) therapies that opened a second front in addition to the battle against pathogens [58]. These immunological discoveries also enabled the identification of patient subgroups, thereby enabling tailored treatments for individual patients (sometimes termed “treatable traits”) and enriching clinical trials with participants most likely to benefit from specific interventions [59, 60]. So-called “endotypes” are subgroups based on molecular classification. Initial endotyping results for sepsis patients were published in the late 2010s. In 2016, Davenport [61] applied an integrated genomics approach to understand the heterogeneity in sepsis; transcriptomic analysis of peripheral blood leucocytes defined two distinct sepsis response signatures (SRS1 and SRS2) associated with different immune response states and prognoses. Antcliffe [62] studied the role of vasopressors and corticosteroids in sepsis using these sepsis endotypes. SRS1 was associated with “immunosuppression” whereas SRS2 was characterized by “immunocompetence”. Using 7 genes and data from the VANISH trial, they showed that hydrocortisone use led to increased mortality in patients with the SRS2 phenotype whereas vasopressor use was not significantly associated with any SRSs. In 2017, Scicluna [63] reported four molecular endotypes for sepsis using a 140-gene expression signature. Mars1 endotype was associated with the highest 28-day mortality. Soon thereafter, P. Khatri's group used 33 mRNA transcripts discovered across 14 discovery datasets to report three robust clusters of ICU patients with bacterial sepsis they termed inflammopathic, adaptive, and coagulopathic [64]. In both the discovery and validation data (9 studies), the adaptive subtype was associated with a lower clinical severity and lower mortality rate, whereas the coagulopathic subtype was associated with higher mortality and coagulopathy. The inflammopathic group had greater bandemia (high number of immature band neutrophils) and a lower lymphocyte proportion on white blood cell differential. This work describes a unifying framework for understanding the molecular heterogeneity of the sepsis syndrome and paves the way for further studies enabling a precision medicine approach of matching immunomodulatory therapies with septic patients most likely to benefit. These early studies provide tools for the molecular classification of patients with sepsis and pave the way for selection of patients in clinical trials and personalized management.

7.2. Stratification of patients for treatment prediction and personalized medicine

In December of 2025, two seminal papers were published along with an editorial in Nature Medicine further advancing the field towards stratification for treatment prediction/personalized medicine [65–67]. Leveraging gene expression data from individuals with bacterial and viral infections enrolled in 19 cohorts (n = 1,460), Moore [66] applied hierarchical clustering and network analysis to identify four conserved transcriptional subtypes, which they then reproduced in ten independent cohorts (n = 3,013) from a separate consortium. Single-cell RNA sequencing analysis revealed that these subtypes differed markedly along two cardinal dimensions of the host response: the myeloid line of white blood cells central to innate immune responses, and the lymphoid line that determines the adaptive ones. Quantitative indices of dysregulation along these two axes form the basis of a proposed Human immune Dysregulation Evaluation Framework (Hi-DEF). The Hi-DEF score predicts specific outcomes in subsets of individuals with sepsis and other critical illnesses (acute respiratory distress syndrome, trauma, and burns) and, notably, it identifies subgroups with differential treatment responses to anakinra (an IL-1 receptor antagonist) in COVID-19 and to corticosteroids in sepsis. Scicluna [67] merged gene expression data from two prospective sepsis cohorts (n = 444) and applied three previously published transcription classifying algorithms. Network analysis suggested three ‘consensus transcriptomic subtypes’ (CTS) defined by distinct biological pathway alterations – inflammation and immature neutrophils (CTS1); heme metabolism and coagulopathy (CTS2); and adaptive immunity (CTS3). Each CTS subtype is associated with a distinct level of clinical severity score and outcome. Lastly, in an indication of real-world validity, a post hoc analysis of observational and clinical trial data showed that corticosteroid treatment was linked to a strong signal of harm in patients within the CTS2 subtype. Thus, single-cell atlases and perturbation screens have clarified that innate and adaptive immunity move along reproducible axes in acute illnesses. The interpretation and implementation of these emerging paradigms is not without caveats. The immune dysregulation framework proposed by Moore and colleagues, although conceptually appealing, is an oversimplified representation of the complex, multidimensional nature of the host response. The Hi-DEF paradigm seems to generalize across critical illness syndromes in existing datasets, but pediatric populations, low-resource settings, and new pathogens require dedicated validation. Treatment–subtype signals from retrospective analyses such as these are hypothesis-generating and require prospective validation. To reduce miscalibration, prospective deployments should include Bayesian or likelihood-ratio decision support that makes the dependence on pretest probability explicit and adjustable - for example, to account for geographic, demographic and temporal variation in disease-relevant variables. The value proposition is speed and simplicity; assay workflows must be single-tube, single-run and robust to ED and ICU variability. Health economic evaluation should be built into implementation trials, with co-primary endpoints including antibiotic days, pathogen resistance patterns, ICU admission, and length of stay.

7.3. Enrichment approaches based on biomarkers

Prospective clinical immunotherapy trials in sepsis employing enrichment approaches based on immunopathological pathways have been reviewed recently by Alevizou [68]. Among other biomarkers, ferritin, monocytic histocompatibility complex DR (HLA-DR), soluble triggering receptor expressed on myeloid cell-1 (sTREM-1), dipeptidyl peptidase 3 (cDPP3) and biologically active adrenomedullin (bioADM) as well as cytokines incl. IL-6, IL-7, IL-15 were employed. Previous work had shown that early initiation of anakinra treatment, guided by soluble urokinase-type plasminogen activator receptor (suPAR) concentrations, significantly reduced the risk of worse clinical outcomes at day 28 in patients hospitalized with moderate and severe COVID-19. Most recently, the ImmunoSep network investigated whether precision immunotherapy targeting macrophage activation-like syndrome (MALS) and sepsis-induced immunoparalysis improved organ dysfunction compared with placebo [69, 70]. Those in the precision therapy group received a 15-day course of either intravenous anakinra if they had MALS or subcutaneous recombinant interferon (IFN-γ) if they had immune paralysis. A greater proportion of those in the intervention groups had at least a 1.4-point improvement in their SOFA score (primary endpoint) compared with the control group and therapy was effective in both the immunosuppressive and immunostimulatory signatures. Monitoring of HLA-DR was used to stop interventions and in a subset of patients with sequential biomarkers, host immune signatures appeared to normalize more rapidly in the intervention group. Secondary analyses of organ dysfunction broadly supported the primary findings, but there was no significant improvement in mortality. It is remarkable that a multicenter precision therapy trial for sepsis has now been fully executed with fidelity across multiple sites and countries, and that the approach, despite testing only a limited set of biomarkers and therapeutic strategies, met the primary outcome.

7.4. Interferon-γ-driven sepsis and additional endotypes

Another sepsis endotype (approx. 20% of patients with suspected sepsis) was recently reported, termed “interferon-γ-driven sepsis” (IDS), with a hallmark of interferon-γ action on tissue macrophages that stimulates the release of the cytotoxic chemokine C-X-C motif chemokine ligand 9 (CXCL9). IDS was an independent risk factor for death in the presence of other endotypes, severity scores, and organ dysfunctions; lower CXCL9 blood levels within the first 72h were associated with better outcomes. Preliminary results of the EMBRACE2a trial were presented in March 2026 (https://www.sobi.com/sites/sobi/files/pr/202603186625-1.pdf) and showed an improvement in organ function (primary endpoint) in 60% of patients receiving high-dose anti-IFN-γ vs. 40% of placebo-treated control patients. Mortality was 40% in the high-dose anti-IFN-γ group and 52% in the control arm. The drug was generally well tolerated, and adverse events (i.e., infections) were not associated with mortality. The EMBRACE2b/3 trial is currently in preparation.

The topic of endotyping is of particular importance for the administration of glucocorticoids commonly used in sepsis treatment. Li [71] recently reviewed glucocorticoid use and underlying pathomechanisms in sepsis, and clinical trials had been extensively reviewed in 2024 [72]. The treatment continues to be controversial as efficacy and safety vary significantly among patients receiving glucocorticoid treatment. The Surviving Sepsis Campaign panel felt that potential adverse events (e.g., increases in hyperglycemia and hypernatremia and the uncertain effect on neuromuscular weakness) would be outweighed by the potential benefits on mortality and shock reversal, in certain patients, and thus made a conditional recommendation (with low evidence) favoring IV corticosteroids in patients with septic shock; this recommendation is in line with the recently published guidance from the Society of Critical Care Medicine [73]. However, the introduction of endotypes that allow stratification of patients based on differences in their inflammatory responses could also identify patients who benefit from corticosteroid therapy, thereby advancing personalized medicine in sepsis.

There also remain clinical phenotypes of sepsis that have not been molecularly defined as endotypes, which could certainly benefit clinical care. For example, shock states vary by individual and their disease state. Broadly speaking, hypotension and hypoperfusion can be phenotypically categorized as either ‘fluid responsive’ hypovolemic shock or ‘fluid refractory’ vasoplegic shock. However, differentiating these opposing shock states can be complicated. While both invasive monitoring, such as central venous pressure and stroke volume assessment, as well as non-invasive monitoring, such as point-of-care-ultrasound have been used to estimate fluid needs, these are imperfect and generally patients currently need a fluid ‘challenge’ with significant volume and/or rate of administration to determine how they will respond. This can lead to harm through fluid overload or from restricting volume or rate of administration until it is clearer how the patient will respond. These varying shock states are an important sepsis phenotype where real harm is happening to patients with both over or under-resuscitation based on currently limited tools.

Precision medicine for sepsis thus is on the horizon but several challenges remain before these insights can be implemented in clinical practice. Current methods often rely on advanced molecular technologies, which need to become more accessible and feasible for real-time use in the ED. Additionally, as shown above, there is no consensus on standardized subtyping systems, which complicates the integration of subtypes across studies. Another challenge is the dynamic nature of sepsis; patients often transition between subtypes during their illness, requiring adaptive models that can account for these changes. Recent research demonstrated that sepsis patients were frequently reclassified across immune profiles over short intervals, with approximately one-third of patients reassigned to different subgroup at each timepoint. This instability challenges the clinical utility of biomarker-derived endotypes [74].

In summary, the integration of enrichment strategies, the advancement of bedside diagnostics, and the adoption of dynamic treatment algorithms mark a new frontier in precision critical care. With corticosteroids as the only (yet not personalized) therapy currently endorsed in guidelines, the field is wide open for the validation and implementation of novel, biomarker-guided immunotherapies, offering hope for improved outcomes in this complex and deadly syndrome.

8. Future outlook

In this sepsis update article we present a few areas of major advances, i.e. diagnostics and personalized treatment. These will certainly improve patient management and outcomes once established in clinical routine. So what will the future bring in addition? Most recently, D. Angus elegantly laid out five big ideas for improving the care of critically ill patients [75]. Three of these ideas are covered in the present review, i.e. earlier detection, the integration of smart technologies, and evaluation of what is fit for purpose (the other topics centering on patients and staff). The integration of rapid point-of-care platforms and other smart technologies into critical care e.g., where diagnostic delays impact outcomes, will address the above-mentioned points but needs to be smartly embedded into patient flow and alerts for sepsis teams to impact diagnosis and the delivery of treatment decisions, including antimicrobial stewardship. Co-developing clinical algorithms aligned with local epidemiology, providing 24/7 clinical microbiology support for result interpretation and targeted therapies, structured training of users, and monitoring test utilization and clinical outcomes is critical to adjust and refine strategies. AI will drive further advances in early recognition, risk stratification, and personalized treatment of sepsis, recently reviewed here [23, 76]. Key applications include AI-driven early warning systems, sub-phenotyping based on clinical and biological data, and decision support tools that adapt to real-time patient information. However, most currently available AI solutions for sepsis have not been evaluated using randomized trials or real-world evidence to demonstrate clinical utility. Meaningful integration into clinical workflows is essential to ensure that these tools improve outcomes. As discussed above, there are additional barriers, including explainability, generalizability, patient involvement, and implementation (see below).

In regard to clinical trials, we envision advances through real-time investigation of immune state control (static or longitudinal) based on computational immunology and point-of-care (POC) diagnostics in trials applying adaptive and other innovative designs; innovation will likely lead to the development of high-multiplex multi-omics immune solutions involving AI interpretation and result readouts. Patient-specific immune responses in sepsis will be measured at defined clinical time points and longitudinally. Furthermore, the design of trials will change towards adaptive platform trials that allow for real-time modifications, e.g. changing from ineffective treatments to new treatments while trials are ongoing; multi-arm trials will allow testing multiple interventions simultaneously. Lastly, AI tools, including patient-specific mechanistic digital twins that model immune and organ trajectories, will likely be leveraged, creating opportunities for timely intervention independent of later outcomes.

The evolution of personalized treatment options in the oncology, e.g. non-small cell lung cancer field, provides an interesting future trajectory for the development of sepsis treatments with a delay of approximately two decades (Fig. 5). Initial findings in molecular pathology translate into the development of specific drug treatments and ultimately to an environment with broad therapeutic options based on the profiling of the tumor, or the immune response (endotypes or tailored traits) in sepsis. Just as immunotherapy transformed oncology by unleashing the patient's own immune system against cancer, precision immunotherapy will do the same against sepsis [58].

Fig. 5.

Evolution of individualized treatment paradigms in non-small cell lung cancer compared to sepsis

EGFR = epidermal growth factor receptor; ALK = anaplastic lymphoma kinase; KRAS = kirsten rat sarcoma viral oncogene homolog; PCR = polymerase chain reaction; IHC = immunohistochemistry; PD = programmed cell death protein; PD-L = programmed cell death protein ligand; CTLA = cytotoxic T-lymphocyte-associated protein; SIRS = systemic inflammatory response syndrome; IL = interleukin; IFN = interferon; AI = artificial intelligence

Implementation of any new solution into clinical practice continues to be a major hurdle. Key operational lessons should provide roadmaps for broader implementation, focusing on workflow integration, stewardship, financial planning, and the early identification of facilitators involved in institutional and clinical support, training, quality assurance, and questionnaires that capture perceived quick results and broad feedback.

In conclusion, moving sepsis care forward will be defined by the convergence of rapid diagnostics, multi-omics profiling, and AI, enabling a shift from generalized treatment approaches toward truly personalized, dynamic patient management. While the scientific and technological foundation for this transformation is emerging rapidly, clinical validation using innovative (adaptive) clinical trial design is required. Successful implementation within routine care will require multidisciplinary collaboration, robust infrastructure, clinician and patient engagement, and healthcare systems capable of integrating innovation into practice. The coming years may see a paradigm shift in sepsis management comparable to advances observed in personalized management of malignancies.

Funding Statement

Funding sources: None.