Background Identifying clinical deterioration is a global health priority. Sepsis is a leading cause of deterioration, responsible for around 46,000 deaths annually in the United Kingdom. Early warning scores based on patients’ vital signs can be embedded into electronic patient records to digitally alert clinicians to those at risk. Rapid identification and treatment – particularly with targeted intravenous antibiotics – are critical to improving outcomes in sepsis patients. Research question This study aimed to evaluate the effectiveness of digital alerts in improving outcomes for patients with sepsis. Using routine electronic patient record data from four United Kingdom National Health Service acute trusts, we investigated how digital alert systems influence patient outcomes and explored mechanisms and mediators of their effectiveness. Objectives Map the types of digital alerts currently in use across United Kingdom hospitals for identifying patients at risk of sepsis (Workstream 1). Evaluate the impact of digital alerts on patient outcomes (Workstream 2). Examine how the implementation process affects alert performance, guided by the consolidated framework for implementation research (Workstream 3). Provide recommendations on alert effectiveness and implementation strategies using systems modelling and mediation analysis (Workstream 4). Methods A mixed-methods approach was employed. A national survey assessed the use of digital sepsis alerts in English National Health Survey hospitals (Workstream 1). Qualitative interviews and focus groups explored the implementation process and its influence on alert performance (Workstream 3). A natural experiment with multilevel interrupted time series analysis examined the impact of sepsis screening tools and digital alerts on outcomes, primarily in-hospital mortality (Workstream 2). Routinely collected clinical data were processed following National Institute for Health Research-Health Information Collaborative standards. Combining quantitative and qualitative data enabled us to link implementation processes with patient outcomes. Results All four trusts experienced reduced mortality rates among patients with serious infections following the introduction of digital sepsis screening tools. After adjustment for patient case-mix, admission patterns and pre-existing trends, one trust showed a statistically significant decrease in mortality linked to digital alert implementation. In two trusts, older patients experienced greater mortality reduction than younger ones following alert introduction. Qualitative findings highlighted factors contributing to more effective use of digital alerts: deployment in general wards rather than intensive care units; use by clinicians familiar with similar technologies; availability of 24/7 emergency outreach teams; robust technological infrastructure and alerts that were user-friendly, non-intrusive and not part of multiple competing alert systems. Conclusions The effectiveness of digital sepsis screening tools varies and may depend on patient’s age and care setting. Our findings suggest that digital alerts should leverage a wider range of electronic patient record data and be tailored to specific patient groups. Different trusts and patient populations may require distinct indicators, thresholds and treatment protocols. These findings align with healthcare practitioners’ calls for more sophisticated, patient-centred sepsis screening tools targeted at relevant clinical teams. Future work and limitations The study involved four National Health Service Trusts with strong data collaboration, but noted limitations include reliance on simple algorithms and varied case-mix and implementation processes. Future research should focus on robust evaluation methods, leveraging granular electronic patient record data and establishing a public registry of digital alert tools. Funding This synopsis presents independent research funded by the National Institute for Health and Care Research (NIHR) Health and Social Care Delivery Research programme as award number NIHR129082. Plain language summary Identifying clinical deterioration is a major focus for health systems across the world. Sepsis is a specific cause of clinical deterioration and death, with an estimated 123,000 cases and 46,000 deaths in the United Kingdom annually. Early warning screening systems are used to identify clinical deterioration and prevent avoidable mortality. Many of these systems use early warning scores based on patients’ vital signs: heart rate, blood pressure, temperature and oxygen saturation. Predetermined levels in each vital sign are associated with scores; these are then added up. The Digital Alerts for Sepsis study aimed to investigate the impact of digital sepsis alerts on patient outcomes and staff activity in National Health Service hospital trusts across England and Wales. As United Kingdom hospitals move from paper-based to electronic health records, the integration of digital alerts to identify patients at risk of deterioration has also become common. The implementation of digital alerts in hospitals is a complex health intervention. Therefore, we used a mixed-methods approach to ensure understanding of the relationship between inherent aspects of the alerts, such as the underlying algorithm and the method of clinician notification. Using appropriate qualitative and quantitative methods, we evaluated the implementation of digital screening tools across four National Health Service Trusts. We examined the impact of these digital tools on mortality due to sepsis. We found that, in some hospitals, there was an important reduction in mortality following introduction of sepsis screening tools. We also showed that these tools may have a bigger impact on older patient groups. Overall, we found that none of the tools made the best use of the rich information contained in the electronic patient record. We recommend that, in future, digital screening tools and alerts should use more of the patients data and that tools should be designed specifically for different patient groups.

Importance Ethical, administrative, regulatory, and logistical (EARL) procedures can hamper clinical trial delivery. Quantification of these hurdles is rare, prohibiting identification of areas for improvement. Objective To identify and quantify EARL hurdles in trial delivery before and during the COVID-19 pandemic. Design, Setting, and Participants This cohort study used data from the ongoing Randomized Embedded Multifactorial Adaptive Platform Trial for Community-Acquired Pneumonia to enable comparison of EARL procedures for multiple protocols across 19 European countries in the pre-COVID-19 pandemic (February 19, 2016 to March 10, 2020) and COVID-19 pandemic (March 11, 2020, to May 4, 2023) periods. Data were analyzed from November 2024 to March 2025 with contracts and protocol submissions as the units of analysis. Main Outcome and Measures Time to (1) site contract completion, (2) regulatory and ethical approval (TTA), and (3) first patient in (FPI). The UK was compared with non-UK countries because of its distinct research infrastructure. Results There were 257 fully signed first contracts with study sites for analysis. In the UK, contract completion times decreased by 97% (95% CI, 95% to 98%), from a median (IQR) of 196 (154 to 250) days in the pre-COVID-19 pandemic period to 5 (1 to 11) days during the COVID-19 pandemic. In non-UK countries, median (IQR) contract completion times were 224 (119 to 412) days and 183 (62 to 291) days before and during the COVID-19 pandemic, respectively (relative difference, -18%; 95% CI, -43% to 52%). In total, 44 interventions in 16 domains were submitted, yielding 232 protocol approvals for analysis. During the COVID-19 pandemic, median (IQR) TTA was 8 (5 to 31) days in the UK and 115 (47 to 103) days in non-UK countries (median difference, 107 days; 95% CI, 76 to 123 days), with large variation across non-UK countries. Time between approval and FPI during the COVID-19 pandemic was, on average, 3 months faster in the UK compared with non-UK countries (median difference, 90 days; 95% CI, 42 to 141 days). Conclusions and Relevance This study found that EARL procedures were lengthy and variable between countries, reflecting different interpretations of trial regulations, with faster processes in the UK. These findings underscore the need to streamline processes across European countries to improve trial efficiency, in particular during future public health emergencies such as pandemics.

Classical approaches to subgroup analysis in randomised controlled trials (RCTs) to identify heterogeneous treatment effects (HTEs) involve testing the interaction between each pre-specified possible treatment effect modifier and the treatment effect. However, individual significant interactions may not always yield clinically actionable subgroups, particularly for continuous covariates. Non-parametric causal machine learning approaches are flexible alternatives for estimating HTEs across many possible treatment effect modifiers in a single analysis. We conducted a secondary analysis of the VANISH RCT, which compared the early use of vasopressin with norepinephrine on renal failure-free survival for patients with septic shock at 28 days. We used classical (separate tests for interaction with Bonferroni correction), data-adaptive (hierarchical lasso regression), and non-parametric causal machine learning (causal forest) methods to analyse HTEs for the primary outcome of being alive at 28 days. Causal forests comprise honest causal trees, which use sample splitting to determine tree splits and estimate treatment effects separately. The modal initial (root) splits of the causal forest were extracted, and the mean value was used as a threshold to partition the population into subgroups with different treatment effects. All three models found evidence of HTE with serum potassium levels. Univariable logistic regression OR 0.435 (95

Background Almost all large-scale trials of disease-modifying therapeutic agents in critical care have failed to show benefit for patients, which may be explained in part by the clinical and biological heterogeneity inherent in virtually all critical illness syndromes. Enrichment strategies have been developed to separate responders from non-responders and better target treatments. In patients with the acute respiratory distress syndrome, a critical illness syndrome involving severe lung inflammation, latent class analysis and other clustering approaches have led to the discovery of subgroups (phenotypes) that appear to respond differently to treatment based on retrospective analyses of published clinical trials and observational cohorts. The next step is to test these phenotypes in a prospective trial. Rapid, point-of-care analytical methods have now made such a trial possible. There is a need to advance treatment for patients with acute respiratory distress syndrome and other critical illness syndromes by incorporating a phenotype-based approach into prospective trial design. The hyperinflammatory and hypoinflammatory phenotypes, that have been identified in acute respiratory distress syndrome, will be the first to be included in such a trial, with scope for further phenotypes to be studied over time. Future work This Efficacy and Mechanism Evaluation report, through expert consensus, describes a new Phase II, multiarm, adaptive platform randomised controlled trial design that tests multiple pharmacological therapies in a population of patients with acute respiratory distress syndrome stratified by baseline inflammatory phenotype. This report also reviews issues to be considered in developing precision medicine trials in critical care, which are designed with newly developed clinical phenotypes in mind. This work has been used to develop the Precision medicine Adaptive Network platform Trial in Hypoxaemic acutE respiratory failuRe precision medicine trial in acute respiratory distress syndrome, which has been funded and will begin recruitment in June 2025. Limitations This report is the result of expert consensus review, rather than utilising strict review methodologies (e.g. Delphi consensus process). However, expert consensus has been found to generate similar results to consensus processes when a high degree of agreement is reached and > 70% agreement was reached for all included recommendations. Funding This article presents independent research funded by the (NIHR) Efficacy and Mechanism Evaluation programme as award number NIHR154493.

Importance Whether alpha(2)-adrenergic receptor agonist-based sedation, compared with propofol-based sedation, reduces time to extubation in patients receiving mechanical ventilation in the intensive care unit (ICU) is uncertain. Objective To evaluate whether dexmedetomidine- or clonidine-based sedation reduces duration of mechanical ventilation compared with propofol-based sedation (usual care). Design, Setting, and Participants Pragmatic, open-label randomized clinical trial conducted at 41 ICUs in the UK including adults who were within 48 hours of starting mechanical ventilation, were receiving propofol plus an opioid for sedation and analgesia, and were expected to require mechanical ventilation for 48 hours or longer. The median time from intubation to randomization was 21.0 (IQR, 13.2-31.3) hours. Recruitment occurred from December 2018 to October 2023; the last follow-up occurred on December 10, 2023. Interventions The bedside algorithms used targeted a Richmond Agitation-Sedation Scale score of -2 to 1 (unless clinicians requested deeper sedation). The algorithms supported uptitration in the dexmedetomidine- and clonidine-based sedation intervention groups and supported downtitration for propofol-based sedation followed by sedation primarily with the allocated sedation (dexmedetomidine or clonidine). If required, supplemental use of propofol was permitted. Main Outcomes and Measures The primary outcome was time from randomization to successful extubation. The secondary outcomes included mortality, sedation quality, rates of delirium, and cardiovascular adverse events. Results Among the 1404 patients in the analysis population (mean age, 59.2 [SD, 14.9] years; 901 [64%] were male; and the mean APACHE II score was 20.3 [SD, 8.2]), the subdistribution hazard ratio (HR) for time to successful extubation was 1.09 (95% CI, 0.96-1.25; P = .20) for dexmedetomidine (n = 457) vs propofol (n = 471) and was 1.05 (95% CI, 0.95-1.17; P = .34) for clonidine (n = 476) vs propofol (n = 471). The median time from randomization to successful extubation was 136 (95% CI, 117-150) hours for dexmedetomidine, 146 (95% CI, 124-168) hours for clonidine, and 162 (95% CI, 136-170) hours for propofol. In the predefined subgroup analyses, there were no interactions with age, sepsis status, median Sequential Organ Failure Assessment score, or median delirium risk score. Among the secondary outcomes, agitation occurred at a higher rate with dexmedetomidine vs propofol (risk ratio [RR], 1.54 [95% CI, 1.21-1.97]) and with clonidine vs propofol (RR, 1.55 [95% CI, 1.22-1.97]). Compared with propofol, the rates of severe bradycardia (heart rate <50/min) were higher with dexmedetomidine (RR, 1.62 [95% CI, 1.36-1.93]) and clonidine (RR, 1.58 [95% CI, 1.33-1.88]). Compared with propofol, mortality was similar over 180 days for dexmedetomidine (HR, 0.98 [95% CI, 0.77-1.24]) and clonidine (HR, 1.04 [95% CI, 0.82-1.31]). Conclusions and Relevance In critically ill patients, neither dexmedetomidine nor clonidine was superior to propofol in reducing time to successful extubation.

RATIONALE: In patients with acute respiratory distress syndrome (ARDS), the hyperinflammatory and hypoinflammatory phenotypes have been reported to have different outcomes and treatment effects in retrospective analyses of completed clinical trials. We tested the hypothesis that real-time identification of inflammatory phenotypes at the bedside with a point-of-care assay and a validated parsimonious classifier model was feasible. METHODS: Patients with ARDS (defined using identical criteria to the 2024 global definition), were recruited in 30 intensive care units in the United Kingdom and Ireland within 72 hours of ARDS onset. Clinical data were collected at baseline. Freshly collected plasma samples were quantitatively analyzed for interleukin-6 (IL-6) and soluble tumour necrosis factor receptor-1 (sTNFR-1) using an Evidence MultiSTAT point-of-care analyzer (Randox Laboratories Ltd). These values were used along with an arterial bicarbonate measurement in the validated parsimonious regression classifier model to calculate the probability of belonging to the hyperinflammatory phenotype, with a cut-off > 0.5 indicating allocation to the hyperinflammatory phenotype. The primary outcome was difference in mortality at 60 days between the hyperinflammatory and hypoinflammatory ARDS phenotypes (NCT04009330). RESULTS: 512 patients were recruited and consented to data usage. For 22 of these patients (4.2%), phenotype allocation was not possible due to assay failure. The prevalence of the hyperinflammatory phenotype was 18% (89/490). Despite the phenotypes having similar age ranges and pulmonary dysfunction as measured by the PaO2/FiO2 ratio and lung injury score, patients with the hyperinflammatory phenotype were more severely ill as measured by APACHE II and SOFA scores. In patients on high-flow nasal oxygen (HFNO) at enrolment, progression to intubation was more frequent in the hyperinflammatory phenotype (7/89 [63.6%] compared to 27/401 [31.0%]). In the hyperinflammatory phenotype, mortality was higher (51.1% vs. 27.9%; 23.2% difference [95% CI 11.9-34.6]; adjusted OR 2.6 [95% CI 1.6-4.4, p<0.001]) and successful extubation was lower (Figure 1). CONCLUSION: Our large multicenter study indicates that real-time classification of ARDS inflammatory phenotypes is feasible in a real-world setting and that the identified phenotypes have distinct clinical characteristics and outcomes. Our findings advance the field by enabling precision medicine clinical trials, such as the PANTHER trial, to test differential pharmacological treatment effects in these phenotypes.

International clinical practice guidelines addressing corticosteroid treatment for patients hospitalised with non-viral community-acquired pneumonia (CAP) are inconsistent. We conducted a systematic review of randomized controlled trials (RCTs) evaluating the use of corticosteroids in hospitalised adult patients with suspected or probable CAP. We performed random effects pairwise, Bayesian, and dose–response meta-analyses using the restricted maximum likelihood (REML) heterogeneity estimator. We assessed certainty of evidence using GRADE methodology. We identified 30 eligible RCTs, including a total of 7519 patients. The prednisone-equivalent doses ranged between 29 mg/day and 100 mg/day. Corticosteroids probably reduced short-term (28–30 days) mortality (RR 0.82 [95

BACKGROUND:Circulating neutrophil-derived extracellular vesicles (NEVs) may contribute to the pathophysiology of acute kidney injury by causing glomerular endothelial inflammation. METHODS:NEVs were first isolated from ex vivo, lipopolysaccharide stimulation of whole blood taken from healthy volunteers (median age [interquartile range {IQR}]: 32 [26-42] yr; 47% female), and also from plasma of COVID-19 patients with acute respiratory distress syndrome (median age [IQR]: 59 [52-66] yr; 45% female). NEVs were incubated for 4 h in a co-culture of peripheral blood mononuclear cells and either human umbilical vein endothelial cells or renal glomerular endothelial cells. Enzyme-linked immunoassays (tumour necrosis factor-alpha [TNF]) and flow cytometry (median fluorescence intensity [MFI]) were used to quantify cell-specific markers of inflammation/activation, in the presence/absence of pharmacological inhibitors. RESULTS:NEVs were internalised by monocytes, leading to their activation via the p38 mitogen-activated protein kinase pathway and increased release of TNF (median [IQR]: 676 [474-1731] pg ml-1 after NEV internalisation, compared with controls (27 [18-29] pg ml-1, P=0.003). This proinflammatory response increased cell adhesion molecule expression (E-selectin) on human umbilical vein endothelial cells (median MFI [IQR]; NEVs: 4120 [3671-4858] vs untreated: 1438 [1252-1708], P=0.008) and human renal glomerular endothelial cells (median MFI [IQR]; NEVs: 2960 [2471-4991] vs untreated: 931 [881-1181], P=0.003). NEVs contained substantial amounts of matrix metalloproteinase-8 and -9, inhibitors of which (doxycycline, MMP-8 inhibitor or MMP-9 inhibitor) prevented endothelial cell inflammation. CONCLUSIONS:Circulating NEVs may contribute to acute kidney injury through renal endothelial inflammation in a monocyte-dependent fashion in patients with acute respiratory distress syndrome.

Evidence-based effective treatments for hospitalized patients with influenza have yet to be identified. Traditional randomized controlled trials have struggled to provide definitive guidance due in part to small sample sizes and logistical challenges. Adaptive platform trials, such as REMAP-CAP (Randomised Embedded Multifactorial Adaptive Platform for Community-Acquired Pneumonia) and RECOVERY (Randomised Evaluation of COVID-19 Therapy), offer a transformative approach to evaluating influenza therapeutics. REMAP-CAP and RECOVERY utilize flexible, efficient designs that enable the simultaneous assessment of multiple interventions, adaptation to emerging data, and large-scale recruitment. Both platforms are currently evaluating antiviral and immunomodulatory therapies for severe influenza, building on their success in identifying effective treatments for COVID-19. Establishing global platform trials for influenza will facilitate the generation of high-quality evidence to guide seasonal influenza treatment and enhance pandemic preparedness. A coordinated international effort to sustain platform trials beyond pandemic periods is essential for improving clinical outcomes, optimizing resource utilization, and ensuring readiness for future pandemics.

IntroductionThe fight against sepsis is an ongoing healthcare challenge, where digital tools are increasingly used with some promising results. The experience of survivors and their family members can help optimize digital alerts for sepsis/deterioration. This study pairs the experiences of survivors of their sepsis journey and family members with their knowledge and views on the role of digital alerts.MethodsA qualitative study with online, semi-structured interviews and focus groups with sepsis survivors and family members in England. Data were analyzed inductively using thematic analysis.ResultsWe included 11 survivors, and 5 family members recruited via sepsis charities and other social media, for a total of 15 sepsis cases. Identified categories correspond to the three stages of the sepsis journey: 1. Pre-hospital, onset symptoms and help-seeking; 2. Hospital admission and stay; 3. Post-sepsis syndrome. The role of digital alerts at each stage of the sepsis journey is discussed. Participants’ experiences were varied, previous sepsis awareness scant, and knowledge of digital alerts minimal. However, participants were confident in the potential of alerts contributing along the sepsis journey. They perceived digital alerts as important in healthcare professionals’ decision-making to expedite identification and treatment of sepsis and suggested their expansion across healthcare services. Participants expressed that awareness should be increased among the general public about digital alerts for sepsis/deterioration.DiscussionIn light of sepsis’ insidious and variable manifestation, the involvement of patients and family members in the development of digital alerts is crucial to optimize their design and deployment towards improving outcomes. Digital alerts should enhance the connection across healthcare services as well as the care quality. They should also enhance the communication between patients and healthcare professionals.Clinical trial registrationThe ClinicalTrials.gov registration identifier for this study is NCT05741801; the protocol ID is 16347.

Introduction The National Health Service (NHS) ‘move to digital’ incorporating electronic patient record systems (EPR) facilitates the translation of paper-based screening tools into digital systems, including digital sepsis alerts. We evaluated the impact of sepsis screening tools on in-patient 30-day mortality across four multi-hospital NHS Trusts, each using a different algorithm for early detection of sepsis.Methods Using quasi-experimental methods, we investigated the impact of the screening tools. Individual-level EPR data for 718 000 patients between 2010 and 2020 were extracted to assess the impact on a target cohort and control cohort using interrupted time series analysis, based on a binomial regression model. We included one Trust which uses a paper-based screening tool to compare the impact of digital and paper-based interventions, and one Trust which did not introduce a sepsis screening tool, but did introduce an EPR.Results All Trusts had lower odds of mortality, between 5% and 12%, after the introduction of the sepsis screening tool, before adjustment for pre-existing trends or patient casemix. After adjustment for existing trends, there was a significant reduction in mortality in two of the three Trusts which introduced sepsis screening tools. We also observed age-specific effects across Trusts.Conclusion Our findings confirm that patients with similar profiles have a lower mortality risk, consistent with our previous work. This study, conducted across multiple NHS Trusts, suggests that alerts could be tailored to specific patient groups based on age-related effects. Different Trusts may require unique indicators, thresholds, actions and treatments. Including additional EPR information could further enhance personalised care.

Sepsis gene-expression sub-phenotypes with prognostic and theranostic potential have been discovered. These have been identified retrospectively and have not been translated to methods that could be deployed at the bedside. We aimed to identify subgroups of septic patients at high-risk of poor outcome, using a rapid, multiplex RNA-based test. Adults with sepsis, in the intensive care unit (ICU) were recruited from 17 sites in the United Kingdom, Sweden and France. Blood was collected at days 2–5 (S1), 6–8 (S2) and 13–15 (S3) after ICU admission and analyzed centrally. Patients were assigned into ‘high’ and ‘low’ risk groups using two models previously developed for the Immune-Profiling Panel prototype on the bioMérieux FilmArray® system. 357 patients were recruited (March 2021–November 2022). 69

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is a respiratory pathogen that emerged in December 2019 and caused a global pandemic by March 2020, with >7 million deaths due to coronavirus disease 2019 (COVID-19) globally as of September 2025. The clinical syndrome of COVID-19 ranges from asymptomatic infection to severe disease with pneumonia and death. SARS-CoV-2 variant type, inoculum, previous exposure and host factors influence the clinical trajectory. Identification of key structural proteins of SARS-CoV-2 and insights into the pathophysiology of the immune response to infection led to the development of effective preventive (vaccines and monoclonal antibodies) and therapeutic (antivirals and immunomodulatory agents) agents. Antiviral agents, such as remdesivir and nirmatrelvir-ritonavir, inhibit viral replication and immunomodulatory agents, such as tocilizumab and baricitinib, act to reduce a dysregulated immune response to SARS-CoV-2. The pandemic had economic and socio-cultural consequences that affected the quality of life and overall life expectancy of individuals. As the emergency phase of the pandemic concludes, robust monitoring and surveillance systems must be sustained and research to improve vaccines and therapeutics must continue to maintain control of SARS-CoV-2 in the population and be prepared for emerging pathogens with pandemic potential. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is the respiratory pathogen that caused the coronavirus disease 2019 (COVID-19) pandemic. In this Primer, Sherman et al. focus on the acute stage of SARS-CoV-2 infection, describing its epidemiology and pathogenesis, discussing its diagnosis and management, and summarizing effects on quality of life as well as areas for further research.

Machine learning has shown promise to detect useful subgroups of patients with sepsis from gene expression and protein data. This approach has rarely been deployed in metabolomic datasets. Metabolomic data are of interest as they capture effects from the genome, proteome, and environmental. We aimed to discover metabolic sub-phenotypes of septic shock, examine their temporal stability and association with clinical outcome. Analysis was performed in two double-blind randomized trials in septic shock (LeoPARDS (1402 samples from 470 patients) and VANISH (493 samples from 173 patients)). Patients were included soon after the onset of shock and had serum collected at up to four time points. Metabolic clusters were identified from 474 metabolites using k-means clustering in LeoPARDS and predicted in VANISH with an elastic net classifier. Three sub-phenotypes were found. The main determinants of cluster membership were lipid species, especially lysophospholipids. Low lysophospholipid sub-phenotypes were associated with higher circulating cytokine levels. Persistence of low lysophospholipid sub-phenotypes was associated with higher mortality compared to the high lysophospholipid sub-phenotype (LeoPARDS: cluster 2 odds ratio 3.66 (95

PURPOSE:The landiolol and organ failure in patients with septic shock (STRESS-L study) included a pre-planned sub-study to assess the effect of landiolol treatment on inflammatory and metabolomic markers. METHODS:Samples collected from 91 patients randomised to STRESS-L were profiled for immune and metabolomic markers. A panel of pro- and anti-inflammatory cytokines were measured through commercially acquired multiplex Luminex assays and statistically analysed by individual and cluster-level analysis (patient). Metabolite fingerprinting was carried out by flow infusion electrospray ionisation high-resolution mass spectrometry and metabolomic data were analysed using the R-based platform MetaboAnalyst. The metabolites were identified using DIMEdb (dimedb.ibers.aber.ac.uk) from their mass/charge ratios. These metabolomic data were also re-analysed using individual and cluster-level analysis. The individual-level models were adjusted for confounders, such as age, sex, noradrenaline dosage and patient (random effect). RESULTS:Analysis was undertaken at cluster- and individual-level. There were no significant differences in cytokine concentration level between trial arms nor survivors and non-survivors over the duration of the observations from day 1 to day 4. Metabolomic analysis showed some separation in the levels of ceramides and cardiolipins between those who survived and those who died. Following adjusted analysis for confounders, plasma metabolite concentrations remained statistically different between landiolol and standard care arms for succinic acid, L-tryptophan, L-alanine, 2,2,2-trichloroethanol, lactic acid and D-glucose. CONCLUSIONS:In a study of ICU patients with established septic shock and a tachycardia, landiolol treatment used to reduce the heart rate from above 95 to a range between 80 and 94 beats per minute did not induce significant cytokine changes. D-Glucose, lactic acid, succinic acid, L-alanine, L-tryptophan and trichloroethanol were pathways that may merit further investigation. TRIAL REGISTRATION:EU Clinical Trials Register Eudra CT: 2017-001785-14 ( https://www.clinicaltrialsregister.eu/ctr-search/trial/2017-001785-14/GB ); ISRCTN registry Identifier: ISRCTN12600919 ( https://www.isrctn.com/ISRCTN12600919 ).

Importance:For hospitalized critically ill adults with suspected sepsis, procalcitonin (PCT) and C-reactive protein (CRP) monitoring protocols can guide the duration of antibiotic therapy, but the evidence of the effect and safety of these protocols remains uncertain. Objective:To determine whether decisions based on assessment of CRP or PCT safely results in a reduction in the duration of antibiotic therapy. Design, Setting, and Participants:A multicenter, intervention-concealed randomized clinical trial, involving 2760 adults (≥18 years), in 41 UK National Health Service (NHS) intensive care units, requiring critical care within 24 hours of initiating intravenous antibiotics for suspected sepsis and likely to continue antibiotics for at least 72 hours. Intervention:From January 1, 2018, to June 5, 2024, 918 patients were assigned to the daily PCT-guided protocol, 924 to the daily CRP-guided protocol, and 918 assigned to standard care. Main Outcomes and Measures:The primary outcomes were total duration of antibiotics (effectiveness) and all-cause mortality (safety) to 28 days. Secondary outcomes included critical care unit data and hospital stay data. Ninety-day all-cause mortality was also collected. Results:Among the randomized patients (mean age 60.2 [SD, 15.4] years; 60.3% males), there was a significant reduction in antibiotic duration from randomization to 28 days for those in the daily PCT-guided protocol compared with standard care (mean duration, 10.7 [SD, 7.6] days for standard care and 9.8 [SD, 7.2] days for PCT; mean difference, 0.88 days; 95% CI, 0.19 to 1.58, P = .01). For all-cause mortality up to 28 days, the daily PCT-guided protocol was noninferior to standard care, where the noninferiority margin was set at 5.4% (19.4% [170 of 878] of patients receiving standard care; 20.9% [184 of 879], PCT; absolute difference, 1.57; 95% CI, -2.18 to 5.32; P = .02). No difference was found in antibiotic duration for standard care vs daily CRP-guided protocol (mean duration, 10.6 [7.7] days for CRP; mean difference, 0.09; 95% CI, -0.60 to 0.79; P = .79). For all-cause mortality, the daily CRP-guided protocol was inconclusive compared with standard care (21.1% [184 of 874] for CRP; absolute difference, 1.69; 95% CI, -2.07 to 5.45; P = .03). Conclusions and Relevance:Care guided by measurement of PCT reduces antibiotic duration safely compared with standard care, but CRP does not. All-cause mortality for CRP was inconclusive. Trial Registration:isrctn.org Identifier: ISRCTN47473244.

The safety of Artificial Intelligence (AI) systems is as much one of human decision-making as a technological question. In AI-driven decision support systems, particularly in high-stakes settings such as healthcare, ensuring the safety of human-AI interactions is paramount, given the potential risks of following erroneous AI recommendations. To explore this question, we ran a safety-focused clinician-AI interaction study in a physical simulation suite. Physicians were placed in a simulated intensive care ward, with a human nurse (played by an experimenter), an ICU data chart, a high-fidelity patient mannequin and an AI recommender system on a display. Clinicians were asked to prescribe two drugs for the simulated patients suffering from sepsis and wore eye-tracking glasses to allow us to assess where their gaze was directed. We recorded clinician treatment plans before and after they saw the AI treatment recommendations, which could be either 'safe' or 'unsafe'. 92% of clinicians rejected unsafe AI recommendations vs 29% of safe ones. Physicians paid increased attention (+37% gaze fixations) to unsafe AI recommendations vs safe ones. However, visual attention on AI explanations was not greater in unsafe scenarios. Similarly, clinical information (patient monitor, patient chart) did not receive more attention after an unsafe versus safe AI reveal suggesting that the physicians did not look back to these sources of information to investigate why the AI suggestion might be unsafe. Physicians were only successfully persuaded to change their dose by scripted comments from the bedside nurse 5% of the time. Our study emphasises the importance of human oversight in safety-critical AI and the value of evaluating human-AI systems in high-fidelity settings that more closely resemble real world practice.