Daily Sepsis Research Analysis
Three impactful studies address key gaps in sepsis care: model-informed meropenem dosing for patients on SLEDD, rapid low-cost sepsis detection using ESR kinetics with machine learning, and hospital-wide dynamic prediction of CLABSI risk with temporal validation. Together they advance precision dosing, scalable diagnostics, and infection prevention strategy.
Summary
Three impactful studies address key gaps in sepsis care: model-informed meropenem dosing for patients on SLEDD, rapid low-cost sepsis detection using ESR kinetics with machine learning, and hospital-wide dynamic prediction of CLABSI risk with temporal validation. Together they advance precision dosing, scalable diagnostics, and infection prevention strategy.
Research Themes
- Model-informed precision antibiotic dosing during renal replacement therapy
- Low-cost rapid sepsis detection via ESR kinetics and machine learning
- Hospital-wide dynamic risk prediction for CLABSI to support prevention
Selected Articles
1. Model-informed identification of optimised dosing strategies for meropenem in critically ill patients receiving SLEDD: an observational study.
In a prospective TDM-informed population PK/PD study of 13 critically ill SLEDD patients (178 samples), the authors developed a one-compartment model and evaluated 24 dosing regimens using PTA against a target window (Cmin 8–44.45 mg/L). They propose an easy-to-use nomogram to individualize meropenem dosing across 7-hour on-SLEDD periods considering renal function and toxicity thresholds.
Impact: This work fills a critical dosing gap for meropenem during SLEDD, where underexposure and toxicity risks are high. A model-informed nomogram provides actionable guidance for ICU clinicians.
Clinical Implications: Supports individualized meropenem dosing in SLEDD by targeting bactericidal exposure while avoiding toxicity, potentially reducing treatment failure in septic ICU patients on hybrid dialysis.
Key Findings
- A one-compartment population PK model described meropenem in SLEDD patients using 178 TDM samples from 13 patients.
- PTA analyses across 24 regimens targeted a Cmin window of 8–44.45 mg/L (covering P. aeruginosa R-breakpoint and toxicity threshold).
- An easy-to-use dosing nomogram was proposed to guide dosing during 7-hour on-SLEDD periods, accounting for renal function.
Methodological Strengths
- Prospective TDM with dense sampling enabling robust popPK modeling.
- PK/PD-driven PTA across multiple regimens with a clinically relevant target window; registered protocol.
Limitations
- Small single-center cohort (n=13) limits generalizability.
- External validation and outcome-based confirmation of the nomogram are lacking; toxicity threshold assumptions may vary across patients.
Future Directions: Prospective multicenter validation with clinical outcomes (microbiologic cure, mortality) and adaptive dosing integration into bedside decision support.
PURPOSE: An increasing number of critically ill patients receive slow extended daily dialysis (SLEDD) due to their pathophysiology while suffering from sepsis, necessitating effective and safe antibiotic therapy. Although SLEDD reduces meropenem exposure and increases treatment failure risk, effective and safe dosing regimens are unclear. We aimed to identify optimised meropenem dosing strategies for critically ill SLEDD patients through population pharmacokinetic (PK) modelling and PK/pharmacodynamic (PD)-based probability of target attainment (PTA) analysis. METHODS: Clinical data from a prospective study involving critically ill SLEDD patients receiving meropenem were monitored through routine therapeutic drug monitoring. A total of 178 blood samples from 13 patients (median 14 samples per patient) were analysed. A PK model was developed and utilised to evaluate 24 clinically relevant dosing regimens during SLEDD therapy (7-h on-SLEDD periods q24h) in PTA analyses. The PK/PD target window of minimum meropenem concentration between 8 mg/L (P. aeruginosa; R-breakpoint) and 44.45 mg/L (toxicity threshold) was used. RESULTS: A one-compartment PK model with linear elimination and total clearance (CL) split into renal (CL CONCLUSION: Our easy-to-use dosing nomogram presents a promising tool in optimising meropenem dosing regimens for critically ill SLEDD patients considering their kidney function in clinical practice. TRIAL REGISTRATION: Clinicaltrials.gov NCT03985605. Registered 14 June 2019. https://classic. CLINICALTRIALS: gov/ct2/show/study/NCT03985605.
2. A machine learning approach for assessing acute infection by erythrocyte sedimentation rate (ESR) kinetics.
Among 346 samples, automated ESR methods agreed with Westergren, and early sedimentation slopes (12–20 min) differentiated disease groups. Machine learning models classified sepsis effectively, with LR achieving AUC 0.884 and 0.991 in validation cohorts at high sensitivity/specificity, highlighting ESR-kinetics as a rapid, low-cost sepsis detection signal.
Impact: Demonstrates that early ESR kinetics combined with ML can enable accurate sepsis detection using ubiquitous, inexpensive tests—potentially transformative for triage in resource-limited settings.
Clinical Implications: Supports rapid triage and early sepsis recognition using existing analyzers; could complement lactate and WBC-based screening and be embedded in lab middleware for automated alerts.
Key Findings
- Automated ESR measurements showed good agreement with the Westergren reference method.
- Early ESR kinetics (12–20 min slopes) differed significantly among groups, with distinctive patterns in sepsis.
- ML models achieved strong sepsis classification: LR AUC 0.884 in validation and 0.991 in a second validation cohort with high sensitivity/specificity.
Methodological Strengths
- Multi-analyzer comparison with reference method and incorporation of early-kinetic features.
- Use of multiple ML algorithms with independent validation cohorts, reporting AUC and operating characteristics.
Limitations
- Modest sample size and potential single-center bias; case-mix may not reflect all clinical settings.
- Risk of overfitting despite validation; prospective, real-time deployment studies are needed.
Future Directions: Prospective multicenter triage studies, integration into LIS/EHR for automated alerts, and combined models with CBC/lactate/CRP for enhanced performance.
BACKGROUND: The erythrocyte sedimentation rate (ESR) is a traditional marker of inflammation, valued for its simplicity and low cost but limited by unsatisfactory specificity and sensitivity. This study evaluated the equivalence of ESR measurements obtained from three automated analyzers compared to the Westergren method. Furthermore, various machine learning (ML) techniques were employed to assess the usefulness of early sedimentation kinetics in inflammatory disease classification. METHODS: A total of 346 blood samples from control, rheumatological, oncological, and sepsis/acute inflammatory status groups were analyzed. ESR was measured using TEST 1 (Alifax Spa, Padua, Italy), VESMATIC 5 (Diesse Diagnostica Senese Spa, Siena, Italy), CUBE 30 TOUCH (Diesse Diagnostica Senese Spa, Siena, Italy) analyzers, and the Westergren method. Early sedimentation rate kinetics (within 20 min) obtained with the CUBE 30 TOUCH were assessed. ML models [Gradient Boosting Machine (GBM), Support Vector Machine (SVM), Naïve Bayes (NB), Neural Networks (NN) and logistic regression (LR)] in discriminating groups were trained and validated using ESR, sedimentation slopes, and clinical data. A second validation cohort of control and sepsis samples was used to validate LR models. RESULTS: Automated methods showed good agreement with Westergren's results. Multivariate analyses identified significant associations between ESR values (measured by CUBE 30 TOUCH) and age (p = 0.025), gender (p < 0.001), and, overall, with samples' group (p < 0.001). Sedimentation rate slopes differed significantly across groups, particularly between 12 and 20 min, with sepsis cases showing distinct patterns. ML models achieved moderate accuracy, with GBM performing best (AUC 0.800). LR for sepsis classification in the validation cohort achieved an AUC of 0.884, with high sensitivity (96.9 %) and specificity (74.2 %). In the second validation cohort, LR outperformed prior results, reaching an AUC of 0.991 (95 % CI: 0.973-1.000), with 95.2 % sensitivity and 100 %. CONCLUSIONS: Current automated technologies for ESR measurement well agree with the reference method and provide robust results for evaluating systemic infections. The novelty of this study lies in connecting ESR sedimentation kinetics to disease states, particularly for identifying sepsis/acute inflammatory status. Future studies with larger datasets are needed to validate these approaches and guide clinical application.
3. Hospital-wide, dynamic, individualized prediction of central line-associated bloodstream infections-development and temporal evaluation of six prediction models.
Using 61,629 training and 44,544 temporal validation catheter episodes, dynamic CLABSI prediction models achieved AUROC up to 0.751 with good calibration at <5% risk and clinical utility for standard-care thresholds (0.5–4%). Performance drift over time highlights the need for model updating.
Impact: Provides a hospital-wide, continuously updating risk assessment for CLABSI with temporal validation, informing targeted prevention and resource allocation.
Clinical Implications: Supports real-time monitoring to trigger standard preventive bundles in medium-risk patients while underscoring the need for periodic recalibration to maintain utility.
Key Findings
- Training set: 61,629 catheter episodes with 3.1% CLABSI; test set: 44,544 episodes with 2.4% CLABSI.
- Best individual model AUROC 0.748 (XGBoost); superlearner AUROC up to 0.751 with improved calibration at low-risk ranges.
- Clinical utility demonstrated at 0.5–4% risk thresholds for standard interventions; limited utility for high-threshold advanced interventions.
Methodological Strengths
- Very large hospital-wide dataset with dynamic, time-updating prediction and competing risk handling.
- Temporal external validation within the same institution and ensemble superlearner approach.
Limitations
- Single-center data; model performance deteriorated over time, indicating drift.
- Calibration overestimation at higher predicted risks; limited utility for advanced interventions at high thresholds.
Future Directions: Prospective deployment with continual learning and recalibration, multi-center validation, and integration into EHR workflows with human factors evaluation.
BACKGROUND: Central line-associated bloodstream infections (CLABSI) are preventable hospital-acquired infections. Predicting CLABSI helps improve early intervention strategies and enhance patient safety. AIM: To develop and temporally evaluate dynamic prediction models for continuous CLABSI risk monitoring. METHODS: Data from hospitalized patients with central catheter(s) admitted to University Hospitals Leuven between 2014 and 2017 were used to develop five dynamic models (a landmark cause-specific model, two random forest models, and two XGBoost models) to predict 7-day CLABSI risk, accounting for competing events (death, discharge, and catheter removal). The models' predictions were then combined using a superlearner model. All models were temporally evaluated on data from the same hospital from 2018 to 2020 using performance metrics for discrimination, calibration, and clinical utility. FINDINGS: Among 61629 catheter episodes in the training set, 1930 (3.1%) resulted in CLABSI, while in the test set of 44544 catheter episodes, 1059 (2.4%) experienced CLABSI. Among individual models, one XGBoost model achieved the highest AUROC of 0.748. Calibration was good for predicted risks up to 5%, while the cause-specific and XGBoost models overestimated higher predicted risks. The superlearner displayed a modest improvement in discrimination (AUROC up to 0.751) and better calibration than the cause-specific and XGBoost models, but worse than the random forest models. The models showed clinical utility to support standard care interventions (at risk thresholds between 0.5-4%), but not to support advanced interventions (at thresholds 15-25%). CONCLUSION: Hospital-wide CLABSI prediction models offer clinical utility based on medium-risk thresholds. Clinical utility at present may be limited as the model performance deteriorated over time.