Daily ReportSep 13, 2026
Sepsis, September 13 edition
We read 37 papers and selected 3.
Summary
Today’s most impactful sepsis-related studies addressed individualized fluid resuscitation, precision antimicrobial dosing during continuous renal replacement therapy, and clinically meaningful phenotyping of surgical sepsis. Together, they combine prospective bedside assessment, physiologically based pharmacokinetic modeling, and externally validated machine-learning approaches to improve early decision-making in critically ill patients.
Research Themes
- Ultrasound-guided assessment of fluid tolerance and responsiveness
- Model-informed precision dosing during continuous renal replacement therapy
- Phenotyping and risk stratification of surgical sepsis
Selected Articles
1. Early assessment of fluid tolerance (VExUS) and stroke volume (LVOT-VTI) to predict adverse outcomes in emergency department patients with suspected sepsis.
In 545 non-intubated, non-vasopressor-treated emergency department patients with suspected sepsis, VExUS congestion scores were strongly associated with 24-hour mortality or ICU admission. Compared with VExUS 0, VExUS 3 was associated with an odds ratio of 4.09 for adverse outcomes and substantially lower fluid responsiveness. Combining VExUS with LVOT-VTI improved characterization of both risk and likely benefit from fluid administration.
Impact: This prospective study provides clinically actionable evidence that early venous congestion assessment may identify patients who are unlikely to benefit from additional fluid and are at increased short-term risk. It directly addresses the challenge of balancing resuscitation against fluid overload in sepsis.
Clinical Implications: VExUS assessment could complement conventional hemodynamic evaluation before additional fluid boluses, particularly in patients at risk of venous congestion. The findings support individualized rather than protocol-driven fluid resuscitation, but implementation should await multicenter validation and outcome-directed trials.
Key Findings
- The prospective cohort included 545 emergency department patients with suspected sepsis who had received no more than 500 mL of fluid and were not intubated or receiving vasopressors.
- Adverse 24-hour outcomes increased progressively with VExUS score; compared with VExUS 0, VExUS 3 had an odds ratio of 4.09 for death or ICU admission.
- Fluid responsiveness decreased from 63.7% in patients with VExUS 0 to 21.1% in those with VExUS 3.
- The combination of VExUS 2-3 and LVOT-VTI below 17 cm identified a group with both high adverse-outcome risk and low fluid responsiveness.
Methodological Strengths
- Prospective recruitment before substantial fluid administration reduced some forms of treatment-related bias.
- The study evaluated both clinical outcomes and physiologic fluid responsiveness using standardized ultrasound measurements.
- Multivariable regression quantified associations across graded VExUS categories and explored combined VExUS-LVOT-VTI phenotypes.
Limitations
- The study was conducted at a single center and enrolled patients with suspected rather than uniformly confirmed sepsis.
- Patients who had already received more than 500 mL of fluid, were intubated, or were receiving vasopressors were excluded, limiting generalizability to more advanced shock.
- The observational design demonstrates association but cannot establish that VExUS-guided fluid management improves outcomes.
Future Directions: Multicenter studies should validate VExUS thresholds across operators and patient populations, assess interobserver reproducibility, and test whether VExUS-guided resuscitation reduces fluid-related complications and improves survival in randomized or pragmatic clinical trials.
BACKGROUND: To determine if early Doppler ultrasound assessments of venous congestion with Venous Excess Ultrasound (VExUS) and stroke volume with left ventricular outflow tract velocity time integral (LVOT-VTI) are associated with 24-h adverse outcomes and fluid-responsiveness in Emergency Department (ED) suspected sepsis patients prior to substantial fluid administration. METHODS: A single-center, prospective cohort study of adult ED patients with suspected sepsis recruited between 5/2020 and 12/2023 (exclusions: receipt of >500 mL fluid, intubated, on vasopressors, or existing do not resuscitate/intubate). The primary outcome was a 24-h ordinal measure of mortality, intensive care unit (ICU) admission, or rapid response team (RRT) activation; and the secondary outcome was fluid-responsiveness (increase in LVOT-VTI ≥10% after a 500 mL fluid bolus).
2. Integrating CRRT as a Virtual Organ Into Physiologically Based Pharmacokinetic Modeling to Optimize Antimicrobial Dosing in Septic Patients.
The study developed a physiologically based pharmacokinetic framework that modeled CRRT as a virtual clearance organ in septic patients. Models for vancomycin, meropenem, and ceftazidime-avibactam adequately reproduced reported pharmacokinetic data, and simulations in 1,000 virtual patients showed that both glomerular filtration rate and CRRT intensity affect drug exposure, with glomerular filtration rate exerting the greater influence. The framework supports model-informed precision dosing across different CRRT modalities and intensities.
Impact: This work offers a generalizable quantitative platform for antimicrobial dosing in a population where standard dosing is unreliable and pharmacokinetic variability is substantial. Treating CRRT as a virtual organ is a methodological advance that can support individualized dosing without requiring drug-specific clinical trials for every combination of renal function and CRRT settings.
Clinical Implications: The framework can inform initial and adaptive dosing of vancomycin, meropenem, and ceftazidime-avibactam in septic patients receiving CRRT. Clinical use should be combined with therapeutic drug monitoring, local susceptibility data, and prospective validation before replacing established dosing protocols.
Key Findings
- A CRRT compartment was integrated into PBPK modeling as a clearance organ and parameterized according to CRRT modality and intensity.
- Models for vancomycin, meropenem, and ceftazidime-avibactam predicted observed exposure within 0.5-2.0-fold, with geometric mean fold errors below 2.
- Simulations of 1,000 virtual patients across GFR values of 0-30 mL/min, five CRRT modalities, and intensities of 20-40 mL/kg/h showed that both GFR and CRRT intensity influence exposure.
- GFR contributed more to exposure variability than CRRT intensity, providing a quantitative basis for model-informed precision dosing.
Methodological Strengths
- The model incorporated sepsis-related pathophysiologic changes and CRRT-specific extracorporeal clearance rather than relying only on conventional renal function adjustment.
- Validation included three clinically important antimicrobials with different pharmacokinetic characteristics.
- Large virtual-patient simulations explored multiple combinations of renal function, CRRT modality, and intensity.
Limitations
- Model validation was based on graphical comparison and reported data rather than a new prospective clinical pharmacokinetic cohort.
- The framework included three antimicrobials and may not directly generalize to other agents without drug-specific model development.
- Clinical outcomes, microbiological cure, toxicity, and target attainment in individual patients were not prospectively tested.
Future Directions: Prospective multicenter studies should validate the model against therapeutic drug monitoring, microbiological outcomes, toxicity, and survival. Future extensions should incorporate additional antimicrobials, dynamic changes in CRRT settings, residual kidney function, and pathogen-specific pharmacodynamic targets.
Continuous renal replacement therapy (CRRT) extracorporeal drug removal, sepsis-related pathophysiological alterations, and reduced pathogen susceptibility to antimicrobials can lead to treatment failure in septic patients undergoing CRRT, highlighting the urgent need for precision dosing. This study aimed to develop a physiologically based pharmacokinetic (PBPK) modeling framework to characterize antimicrobial pharmacokinetics and support dosing optimization in septic patients undergoing CRRT. Drug-specific PBPK models were established in healthy adults and extrapolated to septic patients by scaling four pathophysiological parameters in PK-Sim. A CRRT compartment was subsequently incorporated in MoBi as a clearance organ and parameterized by modality and intensity to construct a PBPK framework for septic patients undergoing CRRT.
3. Identification of Clinical Subtypes of Surgical Sepsis in Critically Ill Patients Based on K-Means Clustering and Development of Parsimonious Classifier Model for High-Risk Subtype.
Using data from 4,027 adults with surgical sepsis and external validation in MIMIC-IV, the study identified three clinical subtypes with markedly different one-year mortality. The highest-risk subtype, comprising 13% of patients, had severe organ dysfunction and 73.3% one-year mortality. A parsimonious four-variable classifier using prothrombin time, norepinephrine dose, lactate, and pH achieved AUCs of 0.970 in derivation and 0.889 in external validation.
Impact: The study moves beyond a single severity continuum by identifying clinically distinct surgical sepsis phenotypes with different long-term prognoses. External validation and a compact classifier increase the potential for rapid bedside risk stratification and future phenotype-directed trials.
Clinical Implications: Early measurement of prothrombin time, lactate, pH, and norepinephrine requirement may help identify surgical sepsis patients at particularly high risk of long-term mortality. The model should be viewed as a risk-stratification tool rather than a treatment-selection instrument until prospective implementation and impact studies are completed.
Key Findings
- K-means clustering identified three clinical subtypes among 4,027 adult surgical sepsis patients.
- Subtype 1, representing 13% of patients, had severe organ dysfunction and the highest one-year mortality at 73.3%; subtype 2 represented 59% and had 26.4% one-year mortality.
- A four-variable logistic model using prothrombin time, norepinephrine dose, lactate, and pH predicted the high-risk subtype with an AUC of 0.970 in the derivation cohort.
- External validation in MIMIC-IV produced a robust AUC of 0.889 and showed similar subtype distributions and clinical profiles.
Methodological Strengths
- The derivation cohort was large and used data from a defined surgical intensive care population.
- Cluster selection incorporated the elbow method and silhouette coefficient rather than relying on an arbitrary number of groups.
- The classifier underwent external validation in the independent MIMIC-IV database and included interpretability analysis using SHAP values.
Limitations
- The derivation cohort was retrospective and obtained from a single institution, allowing potential selection, measurement, and treatment biases.
- The clustering variables were based on worst values within a broad 48-hour window around diagnosis, which may reduce temporal interpretability.
- The model predicts subtype membership rather than directly predicting treatment response or individual patient mortality, and prospective clinical utility remains untested.
Future Directions: Prospective multicenter studies should assess whether real-time subtype assignment changes triage, monitoring, antimicrobial and source-control decisions, or long-term outcomes. Future work should also evaluate dynamic trajectories and whether subtype-specific interventions improve survival.
BACKGROUND: Identifying the phenotypes of patients with surgical sepsis is crucial for improving their management. This study aims to identify the clinical subtypes of critically ill surgical sepsis patients and develop parsimonious classifier model for rapid subtype identification. METHODS: A retrospective cohort study was conducted on 4,027 adult surgical sepsis patients admitted to the Surgical Intensive Care Unit at the First Affiliated Hospital of Sun Yat-sen University from 2018 to 2023, with external validation using the MIMIC-IV (the Medical Information Mart for Intensive Care) database. We extracted the worst values of clinical variables within 24 hours before and after diagnosis and performed K-means clustering. The elbow method and the silhouette coefficient was used to determine the optimal number of clusters. Survival analysis was used to evaluate the prognostic differences among different subtypes.