Daily Sepsis Research Analysis
Analyzed 23 papers and selected 3 impactful papers.
Summary
Today’s most impactful sepsis research spans three complementary advances: prospective development of a validated tool for early candidaemia risk assessment in intensive care, longitudinal transcriptomic characterization of recovery versus treatment failure, and explainable machine-learning prediction of bloodstream infection in carbapenem-resistant Enterobacterales carriers. Collectively, these studies strengthen individualized risk stratification while underscoring the need for external validation and prospective evaluation of clinical benefit.
Research Themes
- Early infection risk prediction and antimicrobial decision support
- Longitudinal host-response profiling in sepsis
- Explainable machine learning for bloodstream infection prediction
Selected Articles
1. Development of Candidaemia Score (CanDi-Score) in Decision of Early Empirical Antifungal Treatment for Patients in Intensive Care Unit: The International Prospective Observational ID-IRI Study.
In 2,704 ICU patients, 204 developed candidaemia. Independent predictors included higher Charlson Comorbidity Index, concurrent infection, neutropenia, prolonged ICU stay, mechanical ventilation, central venous catheter duration, and prolonged total parenteral nutrition; the model achieved an AUC of 0.86 in the development cohort with comparable validation performance.
Impact: This study translates multiple time-dependent ICU risk factors into a practical prediction tool with internal validation, directly addressing the diagnostic delay that contributes to candidaemia mortality. It could improve selection of patients for early empirical antifungal therapy while reducing indiscriminate treatment.
Clinical Implications: CanDi-Score may support daily reassessment of ICU patients for early empirical antifungal treatment, particularly when central venous catheter use, total parenteral nutrition, concurrent infection, or prolonged critical illness is present. It should complement, not replace, microbiological testing and clinical judgment.
Key Findings
- Among 2,704 ICU patients, 204 developed candidaemia.
- Independent predictors included comorbidity burden, concurrent infection, neutropenia, prolonged ICU stay, mechanical ventilation, central venous catheter duration, and prolonged total parenteral nutrition.
- The CanDi-Score showed an AUC of 0.86 in the development cohort, with comparable performance in the validation cohort; prolonged central venous catheter use showed increasing risk across duration categories.
Methodological Strengths
- International, multicentre, prospective observational design with 30-day follow-up.
- Random development and validation cohorts, multivariable modeling, and sensitivity analysis incorporating center-specific random effects.
Limitations
- The observational design cannot establish that CanDi-Score-guided empirical antifungal treatment improves mortality or other patient outcomes.
- The abstract does not report external validation in an independent healthcare system or prospective impact analysis after implementation.
Future Directions: External validation across different ICU populations should be followed by prospective implementation studies evaluating antifungal exposure, time to treatment, confirmed candidaemia, mortality, adverse effects, and antifungal resistance. Calibration and decision-curve analyses would further clarify the threshold for treatment.
BACKGROUND: Candidaemia remains a common, life-threatening infection among intensive care unit (ICU) patients, with high mortality, particularly in patients with delayed diagnosis and treatment. OBJECTIVES: This international, multicentre, prospective, observational study aimed to identify risk factors for candidaemia in ICU patients and to develop an easy-to-use predictive tool, the CanDi-Score, for early initiation of empirical antifungal treatment. PATIENTS/METHODS: All adults over 18 years hospitalised for more than 48 h in the ICUs were included in the study and followed for 30 days. Data on demographics, comorbidities, clinical severity scores including APACHE II and SOFA scores, Charlson comorbidity index (CCI), established risk factors for candidaemia and their time-dependent effects such as total parenteral nutrition (TPN), mechanical ventilation (MV), and central venous catheter (CVC) were collected.
2. Construction of a sepsis recovery-failure axis and single-cell anchoring using public multi-cohort whole-blood and single-cell transcriptomic data.
Using public longitudinal whole-blood and single-cell datasets, the study developed a recovery-failure axis based on 20 recovery-related and 20 failure-related genes. Survivors moved toward recovery by day 5, whereas non-survivors shifted progressively toward failure; the strongest donor-level signal localized to classical monocytes, which showed reduced interferon and oxidative phosphorylation programs and increased inflammatory, IL6-JAK-STAT3, and coagulation programs.
Impact: This work shifts transcriptomic sepsis assessment from static diagnosis or mortality labeling toward longitudinal host-state monitoring. Its cross-cohort and single-cell analyses generate a biologically interpretable framework for identifying non-recovery trajectories and potential monocyte-centered mechanisms.
Clinical Implications: The recovery-failure score could eventually complement bedside severity scores by identifying divergent biological trajectories before conventional outcomes become apparent. It is currently best viewed as a research and hypothesis-generation tool rather than a treatment-selection biomarker or universal clinical cutoff.
Key Findings
- A recovery-failure axis was constructed from 20 recovery-related genes and 20 failure-related genes using longitudinal transcriptomic data.
- Survivors shifted toward recovery by day 5, whereas non-survivors progressively shifted toward failure; the time-outcome interaction remained significant after adjustment for neutrophil proportion and baseline APACHE II.
- The strongest single-cell anchoring occurred in classical monocytes, with non-survivors showing reduced interferon and oxidative phosphorylation programs and increased inflammatory, IL6-JAK-STAT3, and coagulation programs.
Methodological Strengths
- Longitudinal trajectory rules were used rather than relying only on cross-sectional classification or endpoint labels.
- Findings were examined across multiple cohorts and anchored to single-cell data, with null and cross-cohort structure analyses used to assess robustness.
Limitations
- The analysis relies on public datasets with cohort heterogeneity, and cohort-wise standardization limits interpretation as a universal absolute cutoff.
- The score showed only weak correlations with bedside severity scores, and the interferon-activated monocyte displacement analysis had limited cellular coverage and was exploratory.
Future Directions: Prospective serial sampling with standardized processing should test whether the axis predicts clinical deterioration early enough to alter treatment. Integration with proteomics, metabolomics, immune phenotyping, and organ-specific outcomes may establish whether monocyte programs are causal, modifiable mechanisms or merely correlates of non-recovery.
Sepsis exhibits marked heterogeneity in host responses. Although prior time-course transcriptomic studies have provided important information about sepsis diagnosis and mortality-associated gene-expression dynamics, many bioinformatic signatures remain optimized for cross-sectional classification or endpoint labels. Using public multi-cohort whole-blood transcriptomic and single-cell datasets, we constructed a sepsis recovery-failure axis constrained by longitudinal trajectory rules and evaluated its cross-cohort transferability, clinical relevance, and single-cell anchoring. In GSE54514, we identified 20 recovery-related genes and 20 failure-related genes and defined a recovery-failure score. Survivors showed a non-monotonic but ultimately upward movement by day 5, whereas non-survivors progressively shifted toward the failure end of the axis, with clear separation by day 5.
3. Machine learning-based prediction of bloodstream infection in rectal carbapenem-resistant
Among 639 rectal carbapenem-resistant Enterobacterales carriers, 11.3% developed bloodstream infection within 30 days and 7.2% within 14 days. XGBoost models achieved AUROCs of 0.917 for 30-day and 0.854 for 14-day prediction; carbapenem minimum inhibitory concentration was the leading 30-day predictor, while neutrophil count was most influential for 14-day prediction.
Impact: The study combines antimicrobial susceptibility information with longitudinal clinical data and uses explainable machine learning to predict a high-consequence infection in a precisely defined high-risk population. This approach could support targeted surveillance and more rational preemptive antimicrobial decisions.
Clinical Implications: Patients with rectal CRE carriage and high predicted risk may benefit from intensified clinical monitoring, earlier diagnostic blood cultures, and individualized consideration of preemptive therapy. Because the model was developed using a Korean multicentre dataset with random rather than temporal or external validation, implementation should await prospective evaluation in other settings.
Key Findings
- Among 639 carbapenem-resistant Enterales carriers, 72 developed bloodstream infection within 30 days and 46 within 14 days.
- The 30-day and 14-day models achieved AUROCs of 0.917 and 0.854, respectively, using XGBoost with bootstrap assessment.
- Imipenem or meropenem minimum inhibitory concentration was the leading 30-day predictor, while neutrophil count of at least 7.50×10³/μL was the leading 14-day predictor; SHAP enabled patient-level interpretation.
Methodological Strengths
- Multicentre dataset with a clinically specific high-risk population and separate prediction windows for 14-day and 30-day outcomes.
- Use of bootstrap performance estimates and SHAP-based explainable artificial intelligence to characterize nonlinear and interacting predictors.
Limitations
- The dataset was randomly divided into training and test sets, without temporal or fully independent external validation, raising concerns about transportability and optimism.
- The retrospective design and low event count may permit residual confounding and overfitting despite feature reduction and model interpretation.
Future Directions: Future studies should perform temporal and geographically external validation, assess calibration and clinical utility, and compare the model with infectious disease specialist judgment and simpler rule-based scores. Prospective trials should determine whether model-guided surveillance or preemptive therapy improves outcomes without increasing unnecessary antibiotic use or resistance.
BACKGROUND: Carbapenem-resistant METHODS: We included patients with first-time positive rectal CRE surveillance cultures during hospitalization at three hospitals in Korea from January 2014 to December 2023. A total of 173 features-laboratory data including the carbapenem minimum inhibitory concentrations (MIC) of rectal CRE isolates and complete blood cell count, demographics, vital signs, comorbidities, medications, transfusions, and procedures -were collected for the 30-day and 14-day follow-up periods, of which 151 remained after removing redundant features. The entire dataset was randomly split into training (80%) and test (20%) sets, and XGBoost ML models were developed. Model performance was evaluated using the area under the receiver operator characteristic curve (AUROC) with 1, 000 bootstrap iterations. SHAP (SHapley Additive exPlanations) was employed as the XAI method. RESULTS: Among 639 included CRE carriers, 72 (11.3%) and 46 (7.2%) developed BSI within 30 and 14 days, respectively.