Daily ReportSep 14, 2026
Sepsis, September 14 edition
We read 11 papers and selected 3.
Summary
The strongest paper was a multicenter pediatric sepsis analysis showing that an initial venous lactate threshold of 3.5 mmol/L had only fair discrimination for extracorporeal life support or death within 72 hours, supporting cautious rather than isolated use of lactate for risk stratification. A retrospective neonatal machine-learning study demonstrated excellent internal test performance, while a quality-improvement initiative showed that electronic order-set redesign substantially increased balanced-fluid use and shortened time to antibiotics.
Research Themes
- Pediatric sepsis risk stratification
- Machine learning for neonatal sepsis diagnosis
- Implementation of balanced-fluid resuscitation
Selected Articles
1. Secondary analysis of the 2021-2023 Sepsis Epidemiology in Australian and New Zealand Children Observational Dataset: Association Between Presenting Venous Lactate Concentration and 72-Hour Outcomes.
This planned secondary analysis included 4,806 children with suspected community-acquired sepsis from 11 emergency departments in Australia and New Zealand. An initial venous lactate concentration of at least 3.5 mmol/L had 72.7% sensitivity, 84.9% specificity, and an AUC of 0.79 for extracorporeal life support or death within 72 hours, but increased the estimated probability of this outcome only from 0.7% to 3.2%.
Impact: This is one of the largest contemporary multicenter pediatric sepsis datasets assessing a widely used biomarker against a clinically severe short-term outcome. Its modest posttest probability increase cautions against using lactate alone to determine escalation of care.
Clinical Implications: A venous lactate threshold of 3.5 mmol/L may support early risk assessment, but should be interpreted with examination findings, hemodynamics, organ dysfunction, and serial trends rather than used as a standalone trigger for extracorporeal support or prognostic decisions.
Key Findings
- Among 4,806 children, 33 (0.7%) required extracorporeal life support or died within 72 hours.
- A lactate threshold of at least 3.5 mmol/L identified 24 of the 33 severe outcomes, with 72.7% sensitivity and 84.9% specificity.
- The area under the receiver-operating characteristic curve was 0.79, and the estimated outcome probability increased from 0.7% to only 3.2%.
Methodological Strengths
- Large multicenter dataset spanning Australia and New Zealand.
- Prespecified secondary analysis with a clinically meaningful composite outcome and confidence intervals for diagnostic performance.
Limitations
- Only 33 children experienced the composite outcome, limiting precision and threshold stability.
- The observational design cannot establish that lactate-guided management improves outcomes, and generalizability beyond the participating health systems is uncertain.
Future Directions: Future studies should externally validate the 3.5 mmol/L threshold, evaluate serial lactate trajectories, integrate lactate with pediatric organ-dysfunction scores and clinical examination, and test whether biomarker-guided escalation improves patient-centered outcomes.
OBJECTIVES: In pediatric patients with sepsis, we hypothesized that the initial venous lactate concentration would be associated with subsequent need for extracorporeal life support (ECLS) or death within 72 hours. DESIGN: A planned secondary analysis of data from the multicenter, two-country Sepsis Epidemiology Emergency Departments (SENTINEL) study, collected 2021-2023. SETTING: Acute care hospitals in Australia and New Zealand. PATIENTS: SENTINEL participants were children of 0 to less than 18 years with suspected sepsis admitted to the hospital through 11 emergency departments. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We used the initial venous lactate concentration (sampling within 4 hr of hospital arrival) as an explanatory variable for subsequent need for ECLS or death within 72 hours. In 4806 children, with a median (interquartile range [IQR]) age of 1.9 years (IQR, 0.2-6.7 yr), 2651 children (55.2%) were males, and 33 children (0.7%) received ECLS or died within 72 hours.
2. Machine Learning-Based Prediction of Culture-Confirmed Neonatal Sepsis in a Tertiary Neonatal Intensive Care Unit: Retrospective Cohort Study.
This retrospective cohort study evaluated XGBoost, decision-tree, and neural-network models using electronic health record data from 3,274 neonates in a Jordanian tertiary neonatal intensive care unit. XGBoost performed best in the independent test set, with 94% accuracy, 98% sensitivity, and an AUC of 0.98; C-reactive protein, platelet count, and gestational age were important predictors.
Impact: The study demonstrates a potentially useful diagnostic-support approach in a resource-limited setting and explicitly addresses class imbalance and model interpretability. Its high internal performance is promising, but the absence of external validation prevents immediate clinical adoption.
Clinical Implications: An interpretable sepsis prediction tool could support earlier evaluation and targeted testing in neonatal intensive care units, particularly where laboratory turnaround is slow. It should not replace clinical assessment or empiric treatment decisions until prospective and external validation are completed.
Key Findings
- The study included 3,274 neonates, divided into a training set of 2,619 and an independent test set of 655.
- XGBoost achieved 94% accuracy, 98% sensitivity, and an AUC of 0.98 in the test set.
- C-reactive protein, platelet count, and gestational age were among the most important predictive features; neural networks had lower discrimination with an AUC of 0.81.
Methodological Strengths
- Used an independent held-out test set rather than reporting training performance alone.
- Applied stratified sampling and synthetic minority oversampling to address class imbalance, while comparing multiple model classes.
Limitations
- The retrospective single-center design may encode local practice patterns and limit generalizability.
- There was no external or prospective validation, and the abstract does not establish whether predictions improve treatment decisions or clinical outcomes.
Future Directions: External validation across neonatal intensive care units with different patient populations and resource levels should be followed by prospective silent deployment and clinical-impact trials. Calibration, decision-curve analysis, fairness across gestational-age groups, and transparent integration into electronic health records should also be assessed.
BACKGROUND: Neonatal sepsis remains a major cause of neonatal morbidity and mortality in low- and middle-income countries (LMICs). Early diagnosis is challenging because of nonspecific clinical manifestations and delays in laboratory confirmation. Machine learning (ML) approaches using structured electronic health record (EHR) data may improve early risk stratification in neonatal intensive care units (NICUs). OBJECTIVE: This study aimed to evaluate ML models for predicting culture-confirmed neonatal sepsis among neonates admitted to a tertiary NICU in Jordan, with the objective of addressing diagnostic gaps in resource-limited settings. Specifically, we aimed to identify key predictors through feature importance analysis, evaluate model performance with class-imbalanced data, and propose strategies to improve interpretability and generalizability in LMICs. METHODS: A retrospective cohort study was conducted using structured EHRs of 3274 neonates admitted to a tertiary NICU in Jordan between 2018 and 2024. Neonates who underwent blood culture testing were included. The dataset was divided into training (n=2619, 80%) and testing (n=655, 20%) subsets using stratified sampling.
3. Implementing Balanced Fluid Resuscitation for Sepsis Patients in a Tertiary Care Pediatric Emergency Department: A Quality Improvement Initiative.
This single-center quality-improvement initiative used process mapping, failure mode and effects analysis, electronic order-set changes, reminders, and improved access to lactated Ringer's solution. Balanced-fluid administration increased from 4.6% to 35.4%, completion of sepsis huddle documentation increased from 42.8% to 54.2%, and time to antibiotics decreased from 97.7 to 65.6 minutes.
Impact: The study provides a practical example of translating fluid-resuscitation evidence into routine pediatric emergency care through systems engineering rather than relying only on clinician education. It also shows collateral improvement in documentation and antibiotic timeliness, although patient outcomes were not reported.
Clinical Implications: Pediatric emergency departments can consider prechecked balanced-fluid options, opt-out order sets, updated sepsis pathways, accessible supplies, and structured team huddles to improve adherence. Implementation should include safety monitoring and evaluation of kidney injury, acid-base status, fluid overload, and patient-centered outcomes.
Key Findings
- Administration of at least one balanced-fluid bolus increased from 4.6% to 35.4%, exceeding the target of 30%.
- Completion of electronic sepsis huddle forms increased from 42.8% to 54.2%.
- Time from arrival to antibiotic administration decreased from 97.7 minutes to 65.6 minutes.
Methodological Strengths
- Used multidisciplinary quality-improvement methods including process mapping and failure mode and effects analysis.
- Combined clinical pathway revision with electronic health-record redesign, reminders, and supply-access interventions, with special-cause variation reported.
Limitations
- The single-center pre-post quality-improvement design cannot separate intervention effects from secular trends or other concurrent changes.
- The reported outcomes are process measures; effects on mortality, acute kidney injury, fluid overload, and other patient outcomes were not established.
Future Directions: Future work should evaluate sustainability, spread to inpatient units, balancing measures for fluid-related harm, and patient outcomes using interrupted time-series or stepped-wedge designs. Multicenter implementation studies could identify which electronic and workflow components are transferable across pediatric emergency settings.
INTRODUCTION: National data support the use of balanced fluid during volume resuscitation in the setting of sepsis, as it improves morbidity and mortality, particularly preventing acidosis and kidney injury. This single-center quality improvement project aimed to increase the percentage of patients with a positive sepsis screen treated in the tertiary care pediatric emergency department (ED) who received lactated Ringer's (LR) fluid resuscitation from 4.6% to 30% by June 2024. METHODS: A multidisciplinary team examined barriers using a fishbone diagram, process mapping, and failure mode and effects analysis (FMEA). Interventions included order set and clinical practice guideline changes with subsequent electronic medical record changes, huddle form reminders, increased accessibility to LR, as well as discussion forums. RESULTS: Data showed patients receiving at least one balanced fluid bolus in June 2023 increased from 4.6% to 35.4% with special cause variation detected. Completed sepsis huddle forms documented in the electronic medical record (EMR) improved from 42.8% to 54.2% with special cause variation detected. The time to antibiotics from arrival decreased from 97.7 to 65.6 minutes in December 2023 with special cause variation detected.