Daily Sepsis Research Analysis
Analyzed 60 papers and selected 3 impactful papers.
Summary
Three studies advance precision sepsis/critical care: a prospective multicenter cohort (PHIND) demonstrates 1-hour bedside subphenotyping for ARDS with clear 60-day mortality separation; an externally validated ML model predicts sepsis deterioration trajectories with early warning and real-world mortality reduction; and a nationwide neonatal cohort from China links high Gram-negative resistance to treatment mismatches, informing stewardship.
Research Themes
- Rapid bedside subphenotyping for precision critical care
- Trajectory-aware machine learning for early sepsis deterioration prediction
- Antimicrobial resistance shaping neonatal sepsis therapy and stewardship
Selected Articles
1. Bedside identification of subphenotypes in acute respiratory failure (PHIND): a multicentre, observational cohort study.
This multicenter prospective cohort shows that a 1-hour near-patient assay (IL-6, soluble TNFR1, and bicarbonate) can reliably identify hyper- vs hypoinflammatory ARDS subphenotypes with markedly different 60-day mortality (51% vs 28%). The approach operationalizes precision stratification at the bedside and aligns with prior retrospective biology, supporting subphenotype-stratified interventional trials.
Impact: First prospective, actionable bedside subphenotyping with demonstrated prognostic separation enables precision enrollment and targeted therapy testing in ARDS, much of which is sepsis-related.
Clinical Implications: ICUs can use rapid IL-6/TNFR1+bicarbonate testing to classify ARDS patients early, informing trial enrollment and potentially guiding differential therapies once validated. This supports moving from one-size-fits-all ARDS care to precision approaches, including in sepsis-associated ARDS.
Key Findings
- Near-patient IL-6, sTNFR1, and bicarbonate identified hyperinflammatory (18%) vs hypoinflammatory (82%) subphenotypes within ~1 hour.
- 60-day mortality was 51% in hyperinflammatory vs 28% in hypoinflammatory ARDS (RR 1.8, 95% CI 1.4-2.4; adjusted OR 2.7, 95% CI 1.6-4.4).
- Prospective feasibility across 30 centers with concordance to prior retrospective biology, including higher sepsis prevalence and metabolic acidosis in the hyperinflammatory group.
Methodological Strengths
- Prospective, multicenter design with near-patient, 1-hour biomarker assay
- Pre-specified parsimonious logistic model with objective 60-day mortality endpoint
Limitations
- Observational design without randomized therapeutic allocation
- Generalizability beyond UK/Ireland ICUs and to non-ARDS AHRF populations requires validation
Future Directions: Conduct subphenotype-stratified randomized trials testing targeted therapies and validate the assay across broader settings and platforms.
BACKGROUND: Acute respiratory distress syndrome (ARDS) is a clinically defined, biologically heterogeneous condition with no proven disease-modifying therapies. Retrospective analyses have identified two biologically distinct subphenotypes (hyperinflammatory and hypoinflammatory) of ARDS, with differing outcomes and responses to therapy. Rapid identification of these subphenotypes in an actionable timeframe has previously not been possible. The PHIND study aimed to prospectively identify these subphenotypes and to demonstrate differing 60-day mortality. METHODS: The PHIND study was a prospective, multicentre, observational cohort study conducted in intensive care units (ICUs) within the National Health Service in the UK and the Health Service Executive in Ireland. Adult patients aged 18 years and older with ARDS or acute hypoxaemic respiratory failure (AHRF) were enrolled within 72 h of onset of the syndrome. Eligible patients were required to be receiving invasive mechanical ventilation, non-invasive ventilation, or high-flow nasal oxygen. Plasma interleukin (IL-6) and soluble TNF receptor-1 (TNFR1) were quantified at enrolment using a near-patient benchtop immunoanalyser (Randox multiSTAT) with a run time of approximately 1 h. Together with plasma bicarbonate measured from an arterial blood sample, these values were used to prospectively determine subphenotypes on an individual patient basis using a validated parsimonious logistic regression model. The primary outcome was 60-day mortality. The study was registered on ClinicalTrials.gov, NCT04009330. FINDINGS: Between Nov 22, 2019, and Sept 28, 2023, 1853 patients from 30 centres were screened for eligibility. Of these, 1328 were excluded and 525 were recruited into the study, with 512 individuals included. 308 (60%) patients were male, 204 (40%) were female, and mean age was 57·0 years (SD 15·1). 443 (87%) patients were white, 18 (4%) were Black, and 16 (3%) were Asian. 490 were subphenotyped using the near-patient assay: 89 (18%) were classified as hyperinflammatory and 401 (82%) as hypoinflammatory. The primary outcome of 60-day mortality was measured in 486 patients after four patients withdrew consent for confirmation of vital status. 60-day mortality was significantly higher in the hyperinflammatory group (45 [51%] of 88) than in the hypoinflammatory group (111 [28%] of 398; risk ratio 1·8 [95% CI 1·4-2·4], p<0·0001). After adjustment, hyperinflammatory patients had increased odds of 60-day mortality (adjusted odds ratio 2·7 [95% CI 1·6-4·4], p=0·0002). INTERPRETATION: Rapid identification of ARDS inflammatory subphenotypes using a near-patient assay was feasible and associated with many clinical characteristics and outcomes consistent with those described in earlier retrospective studies, including mortality, prevalence of sepsis, and incidence of metabolic acidosis. These findings support the implementation of precision medicine approaches in ARDS and the urgent need for prospective, subphenotype-stratified interventional trials. FUNDING: Innovate UK, Randox Laboratories, and Belfast Health & Social Care Trust.
2. Machine learning predicts sepsis deterioration trajectories.
Across 47,936 Sepsis-3 ICU encounters, group-based trajectory modeling defined three recovery patterns and an ensemble ML model predicted deterioration with AUROC 0.92 (development) to 0.77 (eICU) and a median 17.6-hour warning. Real-world deployment reduced ICU length of stay, ventilation duration, and 28-day mortality.
Impact: Combines trajectory discovery, external validation, and demonstrated clinical impact after implementation—moving ML from prediction to outcome improvement in sepsis.
Clinical Implications: Trajectory-aware early warning can trigger earlier source control, hemodynamic optimization, and staffing/resource allocation. Integration with EHR to surface dynamic risk and physiologic variability may facilitate personalized, timely interventions.
Key Findings
- Three latent trajectories identified: rapid recovery (41.5%), slow recovery (36.4%), deterioration (22.1%).
- Binary deterioration prediction achieved AUROC 0.92 (development), 0.89 (internal), 0.84 (MIMIC-III), 0.77 (eICU); median warning time 17.6 hours.
- Reduced heart rate variability (SD <10 bpm) independently predicted mortality (adjusted HR 2.17).
- Implementation associated with 1.8-day shorter ICU stay, 2.3-day fewer ventilation days, and 5.7% absolute reduction in 28-day mortality.
Methodological Strengths
- Large multicenter datasets with temporal validation, external testing, and real-world implementation
- Group-based trajectory modeling and ensemble methods capturing dynamic physiological variability
Limitations
- Retrospective observational design and non-randomized implementation introduce potential confounding
- Generalizability may vary across institutions and over time due to data drift and practice differences
Future Directions: Prospective, randomized evaluations of trajectory-informed care pathways and adaptive clinical decision support; calibration monitoring to mitigate data drift.
Sepsis has heterogeneous clinical trajectories, but conventional severity scores offer only static risk estimates. Timely, dynamic prediction could enable personalized intervention. In this multicenter retrospective study of 47,936 ICU patients meeting Sepsis-3 criteria from one institutional and two public datasets (MIMIC-III, eICU; sensitivity in MIMIC-IV), group-based trajectory modeling identified latent recovery patterns. An ensemble machine-learning model incorporating dynamic physiological variability was trained, temporally validated, and externally tested; clinical impact was assessed following implementation. Three trajectories emerged: rapid recovery (41.5%), slow recovery (36.4%), and clinical deterioration (22.1%). In the final binary classification task, AUROC was 0.92 (development), 0.89 (internal), 0.84 (MIMIC-III) and 0.77 (eICU); median warning time before deterioration was 17.6 h (Overall pooled across all cohorts). Reduced heart rate variability (SD < 10 bpm) predicted mortality (adjusted HR 2.17). Implementation reduced ICU stay by 1.8 days, machanical ventilation by 2.3 days, and 28-day mortality by 5.7%. This externally validated trajectory-based model offers accurate, early risk stratification for sepsis, supporting proactive, individualized critical care.
3. Antimicrobial resistance and treatment mismatch in culture-proven sepsis among very preterm infants in Chinese NICUs: a cohort study.
In 38,560 very preterm infants across 88 NICUs, culture-proven sepsis showed Gram-negative predominance and high resistance (e.g., cefotaxime resistance in EOS 69.8%), revealing empiric regimen mismatches (ampicillin+ceftazidime would miss 13.3% of EOS). Overuse of vancomycin/linezolid and poor de-escalation underscore stewardship needs.
Impact: A contemporary, nationwide, prospectively collected cohort quantifies resistance-treatment mismatches in neonatal sepsis at scale, directly informing empiric regimen choices and national guideline updates.
Clinical Implications: Empiric regimens should account for high Gram-negative resistance; gentamicin-based combinations may offer better coverage than some beta-lactam pairings. Emphasize susceptibility-guided definitive therapy and robust de-escalation protocols in NICUs.
Key Findings
- Incidence: EOS 1.1% (case-fatality 24.0%) and LOS 5.8% (case-fatality 12.2%) among 38,560 very preterm infants.
- Gram-negative predominance with high resistance: cefotaxime resistance in EOS 69.8% and higher resistance in LOS.
- Empiric mismatch: ampicillin+ceftazidime would fail to cover 13.3% of EOS, whereas gentamicin-based regimens showed comparable coverage.
- Overuse of vancomycin/linezolid (16.6% in Gram-positive EOS) and insufficient de-escalation after pathogen identification.
Methodological Strengths
- Nationwide, prospectively collected data across 88 tertiary NICUs with large sample size
- Integration of pathogen profiles, resistance patterns, and actual prescribing/practice
Limitations
- Observational design limits causal inference on outcomes and regimen effects
- Potential variability in microbiology methods and prescribing practices across centers
Future Directions: Develop and test context-specific empiric algorithms incorporating local resistance; implement stewardship bundles emphasizing de-escalation and measure impact on mortality and resistance.
BACKGROUND: The real-world effectiveness of guideline-recommended antibiotics for neonatal sepsis remains uncertain in settings with high antimicrobial resistance. Evidence derived from large-scale, contemporary data that integrates pathogens, resistance, and treatment is needed. This study aims to assess the epidemiology, antimicrobial resistance, and antibiotic treatment practices for sepsis among very preterm infants (VPIs) in China. METHODS: We analysed prospectively collected data from all VPIs admitted to 88 tertiary neonatal intensive care units (NICUs) in the Chinese Neonatal Network between 2019 and 2022. FINDINGS: Among 38,560 VPIs, 1.1% (95% CI 1.0-1.2) developed early-onset sepsis (EOS; case-fatality 24.0%, 95% CI 19.8-28.8), and 5.8% (95% CI 5.6-6.1) developed late-onset sepsis (LOS; case-fatality 12.2%, 95% CI 10.9-13.7). Gram-negative bacteria predominated in both EOS (59.0%) and LOS (48.3%), with resistance to cefotaxime and ceftazidime reaching 69.8% and 28.6% in EOS, and even higher rates in LOS. An ampicillin plus ceftazidime regimen would fail to cover 13.3% of EOS cases, while gentamicin-based regimens showed comparable coverage. Nearly all Gram-positive EOS pathogens were susceptible to penicillin/ampicillin, yet 16.6% of definitive therapy involved vancomycin or linezolid. Broad-spectrum antibiotics were frequently continued after pathogen identification in both EOS and LOS, with insufficient de-escalation. INTERPRETATION: Sepsis remains a major burden among VPIs in China, with Gram-negative bacteria exhibiting alarming resistance and mismatches between guideline-recommended regimens and real-world effectiveness. These findings highlight the need to integrate susceptibility and prescribing data to inform national guideline updates and targeted antimicrobial stewardship strategies. FUNDING: This work was funded by China Medical Board (grant #20-370).