The eICU Collaborative Research Database, a freely available multi-center database for critical care research

Utilization of lactate trajectory models for predicting acute kidney injury and mortality in patients with hyperlactatemia: insights across three independent cohorts

This study aims to investigate the association between lactate trajectories and the risk of acute kidney injury (AKI) and hospital mortality in patients with hyperlactatemia. We conducted a multicenter retrospective study using data from three independent cohorts. By the lactate levels during the first 48 h of ICU admission, patients were classified into distinct lactate trajectories using group-based trajectory modeling (GBTM) method. The primary outcomes were AKI incidence and hospital mortality. Logistic regression analysis assessed the association between lactate trajectories and clinical outcomes, with adjusting potential confounders. Patients were divided into three trajectories: mild hyperlactatemia with rapid recovery (Traj-1), severe hyperlactatemia with gradual recovery (Traj-2), and severe hyperlactatemia with persistence (Traj-3). Traj-3 was an independent risk factor of both hospital mortality (all p < 0.001) and AKI development (all p < 0.001). Notably, Traj-2 was also associated with increased risk of mortality and AKI development (all p < 0.05) using Traj-1 as reference, except for the result in the Tianjin Medical University General Hospital (TMUGH) cohort for mortality in adjusted model (p = 0.123). Our finding was still robust in subgroup and sensitivity analysis. In the combination cohort, both Traj-2 and Traj-3 were considered as independent risk factor for hospital mortality and AKI development (all p < 0.001). When compared with the Traj-3, Traj-2 was only significantly associated with the decreased risk of hospital mortality (OR 0.17, 95% CI 0.14-0.20, p < 0.001), but no with the likelihood of AKI development (OR 0.90, 95% CI 0.77-1.05, p = 0.172). Lactate trajectories provide valuable information for predicting AKI and mortality in critically ill patients.

A Machine Learning Model to Predict Treatment Effect Associated with Targeted Temperature Management After Cardiac Arrest

BACKGROUND

Targeted temperature management (TTM) has been associated with neurological recovery among comatose survivors of cardiac arrest. The aim of this study is to determine whether models leveraging acute phase multimodal data after intensive care unit admission (hyperacute phase) can predict short-term outcome after TTM.

METHODS

Clinical, physiologic, and laboratory data in the hyperacute phase were analyzed from adult patients receiving TTM after cardiac arrest. Primary end points were survival and favorable neurological outcome. Three machine learning algorithms were trained: generalized linear models, random forest, and gradient boosting. Models with optimal features from forward selection were tenfold cross-validated and resampled 10 times.

RESULTS

The generalized linear model performed best, with an area under the receiver operating characteristic curve ± standard deviation of 0.86 ± 0.04 for the prediction of survival and 0.85 ± 0.03 for the prediction of favorable neurological outcome. Features most predictive of both end points included lower serum chloride concentration, higher serum pH, and greater neutrophil counts.

CONCLUSIONS

We found that in patients receiving TTM after cardiac arrest, short-term outcomes can be accurately determined using machine learning applied to data routinely collected in the first 12 h after intensive care unit admission. With validation, hyperacute prediction could enable personalized decision-making in the postcardiac arrest setting.

Associations of the atherogenic index of plasma with 28-day in-hospital mortality in patients with acute myocardial infarction: a retrospective cohort study from the eICU

BACKGROUND

Despite substantial advancements in treatment strategies, acute myocardial infarction (AMI) continues to exhibit high mortality. Recent research has identified the atherogenic index of plasma (AIP) as a significant measure of cardiovascular outcomes. However, the relationship between the AIP and 28-day mortality during hospitalization in AMI patients remains to be further clarified.

METHODS

A retrospective analysis was conducted based on data sourced from the eICU Collaborative Research Database, encompassing records of 2,517 AMI patients treated in 208 critical care facilities across the U.S. from 2014 to 2015. AIP measurements were derived via log10 (triglyceride/high-density lipoprotein cholesterol) calculations. The primary endpoint was 28-day in-hospital mortality. The analysis utilized adjusted multivariable logistic models with restricted cubic splines for nonlinear associations. Subgroup analyses were performed to evaluate the relationships between AIP and mortality across various demographic and clinical subgroups. These subgroups included age, sex, body mass index (BMI), congestive heart failure, intubation status, mechanical ventilation, pneumonia, diabetes mellitus, antihyperlipidaemic agents, and AMI types.

RESULTS

Among the 2,517 patients enrolled in the cohort (median age: 64.42 years), 138 (5.48%) died within 28 days. The analysis revealed a nonlinear association between the AIP and mortality, presenting a J-curve shape with a threshold of 0.60 (P for nonlinearity = 0.028). Each 0.1-unit elevation above 0.60 corresponded to a 22% increased mortality risk (adjusted OR = 1.22, 95% CI: 1.09-1.36; P = 0.0004). The highest AIP quartile had a 112% greater mortality risk than the lowest quartile (adjusted OR = 2.12, 95% CI: 1.15-3.88; P = 0.0154). Subgroup analyses revealed consistent patterns across the strata.

CONCLUSION

The relationship between the AIP and 28-day hospital mortality in AMI patients may be characterized by a J-shaped curve, where elevated AIP values are associated with increased mortality risk.

Differences in Arterial Blood Gas Testing by Race and Sex across 161 U.S. Hospitals in Four Electronic Health Record Databases

Rationale: Pulse oximetry accuracy varies across races, underscoring the importance of routine arterial blood gas (ABG) testing, the gold standard for assessing oxygen saturation. Objectives: This study aimed to assess disparities in ABG testing among critically ill patients by race and sex. Methods: Records from 2001 to 2019 in 161 U.S. hospitals were analyzed, including Duke, MIMIC-III (Medical Information Mart for Intensive Care), MIMIC-IV, and the eICU Collaborative Research Database. The study evaluated ABG test incidence; time to first test; and frequency of subsequent tests, adjusting for confounders, including the Sequential Organ Failure Assessment, hospital, and age. Subgroup analyses focused on patients with arterial lines and mechanical ventilation. Measurements and Main Results: The cohort included 184,178 ICU admissions (35.0% with ABG test results; 1.9% Asian, 16.5% Black, 3.5% Hispanic or Latino, 78.1% White, 45.7% female). Compared with White patients, Asian, Black, and Hispanic or Latino patients were less likely to have an ABG test (odds ratio [OR] [95% confidence interval (CI)], 0.807 [0.741, 0.879]; 0.859 [0.830, 0.888]; 0.919 [0.865, 0.976], respectively), experienced delays to initial ABG testing (hazard ratio [HR] [95% CI], Asian, 0.855 [0.803, 0.911]; Black, 0.833 [0.814, 0.853]; P < 0.001), and were less likely to have repeated ABG tests (incidence rate ratio [95% CI], Asian 0.913 [0.845, 0.986]; Black 0.913 [0.887, 0.940]). Compared with male patients, female patients underwent fewer ABG tests (OR [95% CI], 0.926 [0.905, 0.948]), had delays in initial testing (HR [95% CI], 0.958 [0.942, 0.974]), and had fewer repeated ABG tests (incidence rate ratio [95% CI], 0.951 (0.931, 0.971)). These findings were consistent among patients who were mechanically ventilated and had arterial lines placed. Conclusions: Asian, Black, and female patients had significantly reduced and delayed rates of ABG testing. Inequitable ABG testing may exacerbate the prevalence of hidden hypoxemia. Until skin tone-corrected pulse oximeters are available, equitable ABG testing remains the best strategy to mitigate hidden hypoxemia.

Initial seizure episodes risk factors identification during hospitalization of ICU patients: A retrospective analysis of the eICU collaborative research database

BACKGROUND

We aimed to identify risk factors for initial seizure episodes in ICU patients using various machine learning algorithms.

METHODS

Using the extensive eICU database, we curated a dataset of 200,859 patient records, with 15,890 patients meeting inclusion and exclusion criteria. Among them, 497 experienced initial seizure episodes during hospitalization. We developed models to identify risk factors associated with these episodes using Logistic Regression, Random Forest, Gradient Boosting, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Decision Tree. After developing and evaluating these individual models, we selected the two best-performing models and combined them using a stacking ensemble learning technique. Additionally, Recursive Feature Elimination (RFE) was used to select the most relevant features. Model performance was evaluated using metrics such as Area Under the Receiver Operating Characteristic Curve (AUC-ROC), accuracy, precision, recall, and F1 score, alongside calibration plots and Decision Curve Analysis (DCA).

RESULTS

The incidence rate of initial seizure episodes was 3.10% (497/15,890), with no significant difference between the training and validation sets. The best-performing individual models were Gradient Boosting (AUC-ROC: 0.78) and Logistic Regression (AUC-ROC: 0.79). The ensemble model achieved an AUC-ROC of 0.80 (95%CI: 0.78-0.82), accuracy of 0.78, precision of 0.80, recall of 0.75, and F1 score of 0.77. Calibration plots demonstrated that the ensemble model's predicted probabilities were well-aligned with observed outcomes. DCA indicated significant net benefit across a range of threshold probabilities, underscoring the model's clinical utility.

CONCLUSION

The ensemble learning model, combining Gradient Boosting and Logistic Regression via a stacking technique, demonstrated superior performance for identifying risk factors for initial seizure episodes in ICU patients. This model was evaluated using a range of performance metrics, including accuracy, sensitivity, specificity, and the AUC-ROC curve, and was validated through 10-fold cross-validation to ensure its robustness and generalizability. These results offer clinically relevant risk factor identification. Key risk factors identified include age, GCS score, glucose levels, hematocrit levels, hyponatremia, stroke history, prothrombin time, potassium levels, and hypertension. The risk estimation table simplifies these complex interactions into a practical tool for clinical use.

Determining the ground truth for the prediction of delirium in adult patients in acute care: a scoping review

Objective

Delirium is a severe condition, often underreported and linked to adverse outcomes such as increased mortality and prolonged hospitalization. Despite its significance, delirium prediction is often hindered by underreporting and inconsistent labeling, highlighting the need for models trained on reliably labeled data (ground truth). This review examines (i) practices for determining labels in delirium prediction models and (ii) how study designs affect label quality, aiming to identify key considerations for improving model reliability.

Materials and Methods

A search of Cochrane, PubMed, and IEEE identified 120 studies that met the inclusion criteria.

Results

To establish the ground truth, 40.8% of studies used routine data, while 42.5% used primary data. The Confusion Assessment Method (CAM) was the most widely used assessment tool (60. 0%). Label and data leakage occurred in 35.0% of studies. High Risk of Bias (RoB) was a recurring issue, with 31.7% of studies lacking sufficient reporting and 36.7% showing inadequate outcome determination. Studies using primary data had lower RoB, whereas those with unclear label sources displayed higher RoB.

Discussion

Our findings underscore the importance of careful planning in determining the ground truth frequently neglected in existing studies. To address these challenges, we provide a decision support flowchart to guide the development of more accurate and reliable prediction models.

Conclusion

This review uncovers significant variability in labeling methods and discusses how this may affect delirium prediction model reliability. Highlighting the importance of addressing underreporting bias and providing guidance for developing more robust models.

Large Language Models in Integrative Medicine: Progress, Challenges, and Opportunities

Integrating Traditional Chinese Medicine (TCM) and Modern Medicine faces significant barriers, including the absence of unified frameworks and standardized diagnostic criteria. While Large Language Models (LLMs) in Medicine hold transformative potential to bridge these gaps, their application in integrative medicine remains underexplored and methodologically fragmented. This review systematically examines LLMs' development, deployment, and challenges in harmonizing Modern and TCM practices while identifying actionable strategies to advance this emerging field. This review aimed to provide insight into the following aspects. First, it summarized the existing LLMs in the General Domain, Modern Medicine, and TCM from the perspective of their model structures, number of parameters and domain-specific training data. We highlighted the limitations of existing LLMs in integrative medicine tasks through benchmark experiments and the unique applications of LLMs in Integrative Medicine. We discussed the challenges during the development and proposed possible solutions to mitigate them. This review synthesizes technical insights with practical clinical considerations, providing a roadmap for leveraging LLMs to bridge TCM's empirical wisdom with modern medical systems. These AI-driven synergies could redefine personalized care, optimize therapeutic outcomes, and establish new standards for holistic healthcare innovation.

Linear association between serum potassium levels and 28-day mortality among ICU patients with diabetes and sepsis: a multicenter study

Background

Dysregulation of serum potassium is a common electrolyte disturbance in critically ill patients, and both hypokalemia and hyperkalemia have been linked to adverse outcomes in sepsis. However, the relationship between serum potassium levels and mortality in ICU patients with diabetes and sepsis remains poorly understood.

Methods

A retrospective cohort study was conducted using data from the eICU Collaborative Research Database (2014-2015). The study included 5,104 adult ICU patients with diabetes and sepsis from 208 hospitals in the U.S. Serum potassium levels measured within 24 h of ICU admission were categorized into hypokalemia (<3.5 mmol/L), normokalemia (3.5-5.0 mmol/L), and hyperkalemia (>5.0 mmol/L). Multivariable logistic regression models were used to assess the association between serum potassium levels and 28-day ICU mortality.

Results

Of the 5,104 patients (mean age, 66.8 years; 49.1% male), 1,046 (20.5%) had hypokalemia, 3,377 (66.2%) had normokalemia, and 681 (13.3%) had hyperkalemia. After adjusting for demographic factors, comorbidities, and treatment measures, each 1 mmol/L increase in serum potassium was associated with a 25% higher risk of 28-day mortality (adjusted OR, 1.25; 95% CI, 1.07-1.47). Compared to hypokalemia, hyperkalemia was associated with significantly higher mortality risk (adjusted OR, 1.86; 95% CI, 1.17-2.96). A linear relationship was observed between serum potassium levels and mortality (P = 0.006), differing from the previously reported U-shaped relationship in general ICU populations.

Conclusions and relevance

Elevated serum potassium levels were independently associated with increased 28-day mortality in ICU patients with diabetes and sepsis. These findings suggest that potassium management strategies should be specifically tailored for this high-risk patient population.

A distributional reinforcement learning model for optimal glucose control after cardiac surgery

This study introduces Glucose Level Understanding and Control Optimized for Safety and Efficacy (GLUCOSE), a distributional offline reinforcement learning algorithm for optimizing insulin dosing after cardiac surgery. Trained on 5228 patients, tested on 920, and externally validated on 649, GLUCOSE achieved a mean estimated reward of 0.0 [-0.07, 0.06] in internal testing and -0.63 [-0.74, -0.52] in external validation, outperforming clinician returns of -1.29 [-1.37, -1.20] and -1.02 [-1.16, -0.89]. In multi-phase human validation, GLUCOSE first showed a significantly lower mean absolute error (MAE) in insulin dosing, with 0.9 units MAE versus clinicians' 1.97 units (p < 0.001) in internal testing and 1.90 versus 2.24 units (p = 0.003) in external validation. The second and third phases found GLUCOSE's performance as comparable to or exceeding that of senior clinicians in MAE, safety, effectiveness, and acceptability. These findings suggest GLUCOSE as a robust tool for improving postoperative glucose management.

ORAKLE: Optimal Risk prediction for mAke30 in patients with sepsis associated AKI using deep LEarning

BACKGROUND

Major Adverse Kidney Events within 30 days (MAKE30) is an important patient-centered outcome for assessing the impact of acute kidney injury (AKI). Existing prediction models for MAKE30 are static and overlook dynamic changes in clinical status. We introduce ORAKLE, a novel deep-learning model that utilizes evolving time-series data to predict MAKE30, enabling personalized, patient-centered approaches to AKI management and outcome improvement.

METHODS

We conducted a retrospective study using three publicly available critical care databases: MIMIC-IV as the development cohort, and SiCdb and eICU-CRD as external validation cohorts. Patients with sepsis-3 criteria who developed AKI within 48 h of intensive care unit admission were identified. Our primary outcome was MAKE30, defined as a composite of death, new dialysis or persistent kidney dysfunction within 30 days of ICU admission. We developed ORAKLE using Dynamic DeepHit framework for time-series survival analysis and its performance against Cox and XGBoost models. We further assessed model calibration using Brier score.

RESULTS

We analyzed 16,671 patients from MIMIC-IV, 2665 from SICdb, and 11,447 from eICU-CRD. ORAKLE outperformed the XGBoost and Cox models in predicting MAKE30, achieving AUROCs of 0.84 (95% CI: 0.83-0.86) vs. 0.81 (95% CI: 0.79-0.83) vs. 0.80 (95% CI: 0.78-0.82) in MIMIC-IV internal test set, 0.83 (95% CI: 0.81-0.85) vs. 0.80 (95% CI: 0.78-0.83) vs. 0.79 (95% CI: 0.77-0.81) in SICdb, and 0.85 (95% CI: 0.84-0.85) vs. 0.83 (95% CI: 0.83-0.84) vs. 0.81 (95% CI: 0.80-0.82) in eICU-CRD. The AUPRC values for ORAKLE were also significantly better than that of XGBoost and Cox models. The Brier score for ORAKLE was 0.21 across the internal test set, SICdb, and eICU-CRD, suggesting good calibration.

CONCLUSIONS

ORAKLE is a robust deep-learning model for predicting MAKE30 in critically ill patients with AKI that utilizes evolving time series data. By incorporating dynamically changing time series features, the model captures the evolving nature of kidney injury, treatment effects, and patient trajectories more accurately. This innovation facilitates tailored risk assessments and identifies varying treatment responses, laying the groundwork for more personalized and effective management approaches.

Analysis of the heterogeneous treatment effect of vasoactive drug dosage and time on hospital mortality across different sepsis phenotypes: a retrospective cohort study

BACKGROUND

The heterogeneity of sepsis poses challenges for the individualized treatment of vasoactive drugs.

METHODS

This study used data from ICUs in MIMIC-IV (2008-2019) and eICU (2014-2015) databases, identified sepsis by sepsis-3 criteria, and stratified sepsis into phenotypes by consensus K-means. The norepinephrine equivalence (NEE) formula balance treatment of different vasoactive drugs, with NEE captured hourly for up to 72 h to record both time of use and dosage. The logistic regression model, including phenotype-dosage-time interactions, examined heterogeneous treatment effects on hospital mortality. To address confounding, three models were fitted: Model 1 unadjusted, Model 2 adjusted for age and sex, and Model 3 additionally included 7 clinical variables identified via machine learning and directed acyclic graph. Nonlinear dosage was further analyzed based on restricted cubic splines. P values and P for interaction were Bonferroni-adjusted.

RESULTS

A total of 54,673 sepsis patients were included for phenotype identification, and 8,803 patients were further analyzed to evaluate heterogeneous treatment effect of vasoactive drugs. Four sepsis phenotypes were identified: A, B, C and D. Phenotype D was the most severe subgroup, followed by phenotype C, while phenotypes A and B were mild subgroups. In Model 3, each 0.05 μg/kg/min increase in NEE dosage was linked to higher hospital mortality (OR 1.328, 95% CI 1.314-1.342; p < 0.001). Longer NEE time of use also significantly increased mortality risk (OR 1.006, 95% CI 1.005-1.007; p < 0.001). In addition, these associations varied significantly by phenotype (P for interaction < 0.001). In RCS model, phenotype A consistently showed higher mortality than the other phenotypes at NEE dosages of 0.1-0.5 µg/kg/min, with this gap increasing over time, showing a clear dosage-time dependence. Phenotype B displayed lower overall mortality but the steepest relative risk of hospital mortality increased as dosage and time (OR of dosage: 1.309; OR of time: 1.005) in Model 3. Phenotype C reached the highest mortality risk when dosages exceeded 0.5 µg/kg/min, which was dosage dependence. Finally, phenotype D followed a U-shaped curve in RCS model, and minimum mortality was around 20% at 0.03-0.05 µg/kg/min.

CONCLUSIONS

Sepsis phenotypes differ significantly in their treatment effects of vasoactive drug dosage and time of use, indicating the need for phenotype-specific treatment strategies to improve outcomes.

Association between aspartate aminotransferase to alanine aminotransferase ratio and 28-day mortality of ICU patients: A retrospective cohort study from MIMIC-IV database

BACKGROUND

Prior studies have linked the aspartate aminotransferase to alanine aminotransferase ratio (AAR) with negative health outcomes in the elderly and specific populations. However, the impact of AAR on the prognosis of the entire population in the intensive care unit (ICU) remains unclear. This study aimed to determine the correlation between AAR and the mortality among adult ICU patients.

METHOD

Patient data were retrieved from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database and stratified into quartiles by AAR. Survival analysis using the Kaplan-Meier curves was conducted to compare survival across quartiles. The primary outcome was 28-day mortality, with secondary outcomes including 60-day, 90-day, and 365-day mortality, along with ICU-free, ventilator-free, and vasopressor-free days within the first 28 days. The association between AAR and mortality was evaluated using Cox proportional hazards regression analysis complemented by a restricted cubic spline. Furthermore, the eICU Collaborative Research Database (eICU-CRD) was used as an external validation cohort for sensitivity analysis.

RESULT

The study included 20,225 patients with a mean age of 63.7 ± 17.5 years. Kaplan-Meier analysis indicated a higher risk of 28-day mortality for patients with higher AAR (log-rank P < 0.001). After adjusting for confounders, the AAR was significantly related to 28-day mortality (HR = 1.04, 95% CI: 1.03-1.06, P < 0.001) and other mortality benchmarks, exhibiting an inverted L-shaped relationship. The inflection point of the AAR for 28-day mortality was 2.60. Below this threshold, each unit increase in the AAR was associated with a 19% rise in the risk of 28-day mortality (HR = 1.19, 95% CI: 1.11-1.27, P < 0.001), with a plateau observed above this threshold. Subgroup and sensitivity analyses further confirmed the robustness and generalizability of the study.

CONCLUSION

AAR demonstrated a significant association with 28-day, 60-day, 90-day, and 365-day mortality, characterized by an inverted L-shaped pattern.

Toward Identifying New Risk Aversions and Subsequent Limitations and Biases When Making De-identified Structured Data Sets Openly Available in a Post-LLM world

Making clinical datasets openly available is critical to promote reproducibility and transparency of scientific research. Currently, few datasets are accessible to the public. To support the open science initiative, we plan to release the structured clinical datasets from the CONCERN study. In this paper, we are presenting our de-identification approaches for structured data, considering the future inclusion of de-identified narrative notes and re-identification risks in the LLM era. Through literature review and collaborative consensus sessions, our team made informed decisions regarding dataset release, weighing the pros and cons of each choice, outlining limitation and bias introduced by the de-identification algorithm. To our best knowledge, this is the first study describing the rationales of de-identification decisions in the LLMs era, delineating the consequent problems that should be considered when using our data set. We advocate for transparent disclosure of de-identification decisions and associated limitations and biases with all openly available datasets.

Enhancing Semantic and Structure Modeling of Diseases for Diagnosis Prediction

Electronic Health Records (EHRs) are valuable healthcare data, aiding researchers and doctors in improving diagnosis accuracy. Researchers have developed several predictive models by learning disease representations to forecast the potential diagnosis that patients may receive. However, existing studies usually ignore the fine-grained semantic and structure information in EHRs (e.g., the hierarchical relations between diseases and ICD-9 codes), which fails to provide accurate disease representation towards effective diagnosis prediction. To this end, we propose to enhance diagnosis prediction through LabCare, a framework with improved semantic and structure modeling of diseases in EHR data. LabCare can simultaneously capture rich semantic and structural relations among diseases and ICD-9 codes, which is achieved by innovatively integrating language models and box embeddings. Extensive experiments on two EHR datasets show that LabCare surpasses competitors, consistently achieving a 4.29% average improvement in Recall and NDCG metrics.

Association between aspartate aminotransferase to alanine aminotransferase ratio and mortality in critically ill patients with end stage renal disease

The aspartate aminotransferase to alanine aminotransferase (AST/ALT) ratio has been extensively studied in relation to mortality, yet its specific association with intensive care unit (ICU) mortality in end stage renal disease (ESRD) patients remains underexplored. The study investigated this relationship in critically ill ESRD patients. This multicenter retrospective cohort study analyzed data from ESRD patients admitted to 208 ICUs across the United States between 2014 and 2015 using the eICU Collaborative Research Database. Smooth curve fitting with Generalized Additive Model and two-piecewise linear regression analyses were utilized to examine nonlinear relationships. Among the 3005 patients (mean age 62.68 ± 14.16 years; 54.48% male), 252 (8.39%) died in the ICU. A significant nonlinear relationship between the AST/ALT ratio and ICU mortality was identified with an inflection point of 1.59. For AST/ALT ratios ≤ 1.59, each unit increase was associated with a 2.02-fold higher risk of ICU mortality (OR 2.02, 95% CI 1.22-3.33, P = 0.0059). For AST/ALT ratios > 1.59, no significant association with mortality was observed (OR  1.07, 95% CI 0.86-1.33, P = 0.5348). Sensitivity analyses confirmed the robustness of these findings. In critically ill ESRD patients, a nonlinear relationship exists between AST/ALT ratio and ICU mortality.

Evaluating the SWIFT algorithm's efficacy in predicting hypoxemia across multiple critical care datasets

BACKGROUND

Machine learning models to predict hypoxia in patients could improve timely interventions. Due to the diversity and limited generalizability of approaches, external validation is required.

OBJECTIVE

This study aimed to validate the generalizability of SpO2 Waveform ICU Forecasting Technique (SWIFT), an LSTM algorithm for predicting SpO2 5 and 30 min in advance, on two external datasets.

METHODS

We trained the SWIFT model on eICU Collaborative Research Database (eICU-CRD) and validated it on Medical Information Mart for Intensive Care IV (MIMIC-IV) and Amsterdam University Medical Centers Database (UMCdb) data. We evaluated SWIFT-5 and SWIFT-30 for ventilated and non-ventilated populations.

RESULTS

The sampling procedure resulted in substantial population size reduction for MIMIC-IV and UMCdb data due to differences in SpO2 measurement frequency. SWIFT performed well on eICU-CRD data but showed reduced performance on MIMIC-IV data, particularly for SWIFT-30. UMCdb validation demonstrated promise, with comparable performance to eICU-CRD for ventilated patients. All datasets exhibited high specificity and NPV, critical for gaining trust in alarms in clinical applications.

CONCLUSIONS

The study highlights challenges in generalizing prediction models across diverse ICU populations, emphasizing need for external validation. Further research should focus on improving model adaptability and interpretability, considering the practical application in clinical settings.

FedWeight: mitigating covariate shift of federated learning on electronic health records data through patients re-weighting

Federated learning (FL) enables collaborative analysis of decentralized medical data while preserving patient privacy. However, the covariate shift from demographic and clinical differences can reduce model generalizability. We propose FedWeight, a novel FL framework that mitigates covariate shift by reweighting patient data from the source sites using density estimators, allowing the trained model to better align with the distribution of the target site. To support unsupervised applications, we introduce FedWeight ETM, a federated embedded topic model. We evaluated FedWeight in cross-site FL on the eICU dataset and cross-dataset FL between eICU and MIMIC III. FedWeight consistently outperforms standard FL baselines in predicting ICU mortality, ventilator use, sepsis diagnosis, and length of stay. SHAP-based interpretation and ETM-based topic modeling reveal improved identification of clinically relevant characteristics and disease topics associated with ICU readmission.

Predicting ICU Mortality Among Septic Patients Using Machine Learning Technique

Introduction: Sepsis leads to substantial global health burdens in terms of morbidity and mortality and is associated with numerous risk factors. It is crucial to identify sepsis at an early stage in order to limit its escalation and sequelae associated with the condition. The purpose of this research is to predict ICU mortality early and evaluate the predictive accuracy of machine learning algorithms for ICU mortality among septic patients. Methods: The study used a retrospective cohort from computerized ICU records accumulated from 280 hospitals between 2014 and 2015. Initially the sample size was 23.47K. Several machine learning models were trained, validated, and tested using five-fold cross-validation, and three sampling strategies (Under-Sampling, Over-Sampling, and Combination). Results: The under-sampled approach combined with augmentation for the Extra Trees model produced the best performance with Accuracy, Precision, Sensitivity, Specificity, F1-Score, and AUC of 90.99%, 84.16%, 94.89%, 88.48%, 89.20%, and 91.69%, respectively, with Top 30 features. For Over-Sampling, the Top 29 combined features showed the best performance with Accuracy, Precision, Sensitivity, Specificity, F1-Score, and AUC of 82.99%, 51.38%, 71.72%, 85.41%, 59.87%, and 78.56%, respectively. For Down-Sampling, the Top 31 combined features produced Accuracy, Precision, Sensitivity, Specificity, F1-Score, and AUC of 81.78%, 49.08%, 79.76%, 82.21%, 60.76%, and 80.98%, respectively. Conclusions: Machine learning models can reliably predict ICU mortality when suitable clinical predictors are utilized. The study showed that the proposed Extra Trees model can predict ICU mortality with an accuracy of 90.99% accuracy using only single-entry data. Incorporating longitudinal data could further enhance model performance.

Multicenter target trial emulation to evaluate corticosteroids for sepsis stratified by predicted organ dysfunction trajectory

Corticosteroids decrease the duration of organ dysfunction in sepsis and a range of overlapping and complementary infectious critical illnesses, including septic shock, pneumonia and the acute respiratory distress syndrome (ARDS). The risk and benefit of corticosteroids are not fully defined using the construct of organ dysfunction duration. This retrospective multicenter, proof-of-concept study aimed to evaluate the association between usage of corticosteroids and mortality of patients with sepsis, pneumonia and ARDS by emulating a target trial framework stratified by predicted organ dysfunction trajectory. The study employed a two staged machine learning (ML) methodology to first subphenotype based on organ dysfunction trajectory then predict this defined trajectory. Once patients were classified by predicted trajectory we conducted a target trial emulation. Our analysis revealed that the association between corticosteroid use and 28-day mortality varied by predicted trajectory and between cohorts.Our findings suggest that matching treatment strategies to empirically observed pathobiology may offer a more nuanced understanding of corticosteroid utility.

Association Between Lactate-to-Albumin Ratio and 28-Day All-Cause Mortality in Critical Care Patients with COPD: Can Both Arterial and Peripheral Venous Lactate Serve as Predictors?

Background

Lactate-to-albumin ratio (LAR) has been reported as a useful predictor for multiple critical illnesses. However, the association between LAR and mortality in patients with chronic obstructive pulmonary disease (COPD) remains unclear. This study aims to clarify the correlation between LAR and 28-day all-cause mortality in patients with COPD and to investigate whether LAR calculated using arterial lactate (AL) or peripheral venous lactate (PVL) can serve as predictive indicators.

Methods

A total of 1428 patients from the Medical Information Mart for Intensive Care (MIMIC) IV database (version 2.2) and 2467 patients from the eICU Collaborative Research Database (eICU-CRD, version 2.0) were included in this study. Propensity score matching (PSM) method was conducted to control confounders. Cox proportional hazards model, Kaplan-Meier survival method, subgroup analysis and receiver operating characteristic (ROC) analysis were performed to assess the predictive ability of LAR. To verify our hypothesis, data from the two databases were analyzed individually.

Results

After adjusting for covariates, LAR calculated using either AL (MIMIC IV, HR = 1.254, 95% CI, 1.013-1.552, P = 0.038) or PVL (eICU-CRD, HR = 1.442, 95% CI, 1.272-1.634, P < 0.001) was independently associated with 28-day all-cause mortality in COPD patients. Kaplan-Meier analysis showed that patients with higher LAR value had significantly higher all-cause mortality (all P < 0.05). This association was consistent across subgroup analyses. In addition, the ROC analysis suggested that LAR calculated using PVL may have better predictive performance compared to using AL.

Conclusion

LAR calculated using both AL and PVL can independently predict the 28-day all-cause mortality after ICU admission in patients with COPD and higher level of LAR is related to higher mortality risk.

Predicting prolonged hospitalization in asthma patients: model development and external validation

PURPOSE

This study aims to develop and validate a machine learning (ML) model to predict prolonged hospitalization in asthma patients.

PATIENTS AND METHODS

This retrospective cohort study included patients with asthma as the primary diagnosis. We randomly divided 2820 asthma patients from Beth Israel Deaconess Medical Center into a training set and an internal validation set (in an 8:2 ratio), and used 1714 asthma patients from 208 other hospitals in the United States as an external validation cohort. Prolonged hospitalization was the primary outcome. Feature selection was conducted using LASSO regression, univariate logistic regression, and multivariate logistic regression analyses. Nine ML algorithms were employed to develop predictive models.

RESULTS

Based on discrimination, calibration, and clinical utility, the Extreme Gradient Boosting (XGBoost) model demonstrated the best overall performance. The nine most important predictors in the model were age, oxygen saturation (SpO2), red blood cell count, hemoglobin count, comorbid pneumonia, chronic obstructive pulmonary disease (COPD), congestive heart failure, anxiety, and use of invasive mechanical ventilation. The XGBoost model achieved an area under the receiver operating characteristic curve (AUC) of 0.829 and a Cohen's Kappa value of 0.439 in the internal validation set, and an AUC of 0.745 and a Cohen's Kappa value of 0.315 in the external validation set. The decision curve analysis indicated good clinical utility of the model.

CONCLUSIONS

The XGBoost model can effectively predict prolonged hospitalization in asthma patients.

Influence of age-adjusted shock index trajectories on 30-day mortality for critical patients with septic shock

Background

Septic shock poses a high mortality risk in critically ill patients, necessitating precise hemodynamic monitoring. While the age-adjusted shock index (ASI) reflects hemodynamic stability, the prognostic value of its dynamic trajectory remains unexplored. This study evaluates whether dynamic 24-h ASI trajectories predict 30-day mortality in septic shock patients.

Methods

This retrospective cohort study extracted data from the MIMIC-IV (derivation cohort, n = 2,559) and eICU-CRD (validation cohort, n = 2,177) databases. The latent category trajectory model (LCTM) classified ASI changes within 24 h of intensive care unit (ICU) admission. The association between ASI trajectory categories and 30-day mortality was evaluated using Kaplan-Meier (KM) method and Cox proportional-hazard models, reported as hazard ratios (HRs) and 95% confidence intervals (CIs).

Result

Three distinct ASI trajectories were explored: persistently low (Classes 1), initial high ASI sharply decreasing followed by instability (Classes 2), and steady ASI increase (Classes 3). KM curve revealed significantly higher 30-day mortality in Class 2 (32.1%) and Class 3 (38.7%) than Class 1 (12.3%) (P < 0.001). After fully adjusting for covariates, Class 2 (HR = 1.68, 95% CI: 1.25-2.25, P = 0.001) and Class 3 (HR = 1.87, 95% CI: 1.26-2.77, P = 0.002) showed elevated mortality risks in the derivation cohort. Validation cohort results were consistent (Class 2: HR = 1.92, 95% CI: 1.38-2.68, P = 0.001) and (Class 3: HR = 1.66, 95% CI: 1.09-2.54, P = 0.019). Triple-robust analyses and subgroup analyses confirmed the reliability of the results.

Conclusion

Dynamic 24-h ASI trajectories independently predict 30-day mortality in patients with septic shock, with unstable or rising patterns signaling high-risk subgroups. This underscores the clinical utility of real-time ASI monitoring for early risk stratification and tailored intervention.

Developing a Scalable Annotation Method for Large Datasets That Enhances Alarms With Actionability Data to Increase Informativeness: Mixed Methods Approach

BACKGROUND

Alarm fatigue, a multifactorial desensitization of staff to alarms, can harm both patients and health care staff in intensive care units (ICUs), especially due to false and nonactionable alarms. Increasing amounts of routinely collected alarm and ICU patient data are paving the way for training machine learning (ML) models that may help reduce the number of nonactionable alarms, potentially increasing alarm informativeness and reducing alarm fatigue. At present, however, there is no publicly available dataset or process that routinely collects information on alarm actionability (ie, whether an alarm triggers a medical intervention or not), which is a key feature for developing meaningful ML models for alarm management. Furthermore, case-based manual annotation is too slow and resource intensive for large amounts of data.

OBJECTIVE

We propose a scalable method to annotate patient monitoring alarms associated with patient-related variables regarding their actionability. While the method is aimed to be used primarily in our institution, other clinicians, scientists, and industry stakeholders could reuse it to build their own datasets.

METHODS

The interdisciplinary research team followed a mixed methods approach to develop the annotation method, using data-driven, qualitative, and empirical strategies. The iterative process consisted of six steps: (1) defining alarm terms; (2) reaching a consensus on an annotation concept and documentation structure; (3) defining physiological alarm conditions, related medical interventions, and time windows to assess; (4) developing mapping tables; (5) creating the annotation rule set; and (6) evaluating the generated content. All decisions were made based on feasibility criteria, clinical relevance, occurrence frequency, data availability and quantity, structure, and storage mode. The annotation guideline development process was preceded by the analysis of the institution's data and systems, the evaluation of device manuals, and a systematic literature review.

RESULTS

In a multidisciplinary consensus-based approach, we defined preprocessing steps and a rule-based annotation method to classify alarms as either actionable or nonactionable based on data from the patient data management system. We have presented our experience in developing the annotation method and provided the generated resources. The method focuses on respiratory and medication management interventions and includes 8 general rules in a tabular format that are accompanied by graphical examples. Mapping tables enable handling unstructured information and are referenced in the annotation rule set.

CONCLUSIONS

Our annotation method will enable a large number of alarms to be labeled semiautomatically, retrospectively, and quickly, and will provide information on their actionability based on further patient data. This will make it possible to generate annotated datasets for ML models in alarm management and alarm fatigue research. We believe that our annotation method and the resources provided are universal enough and could be used by others to prepare data for future ML projects, even beyond the topic of alarms.

Federated Target Trial Emulation using Distributed Observational Data for Treatment Effect Estimation

Target trial emulation (TTE) aims to estimate treatment effects by simulating randomized controlled trials using real-world observational data. Applying TTE across distributed datasets shows great promise in improving generalizability and power but is always infeasible due to privacy and data-sharing constraints. Here we propose a Federated Learning-based TTE framework, FL-TTE, that enables TTE across multiple sites without sharing patient-level data. FL-TTE incorporates federated protocol design, federated inverse probability of treatment weighting, and a federated Cox proportional hazards model to estimate time-to-event outcomes across heterogeneous data. We validated FL-TTE by emulating Sepsis trials using eICU and MIMIC-IV data from 192 hospitals, and Alzheimer's trials using INSIGHT Network across five New York City health systems. FL-TTE produced less biased estimates than traditional meta-analysis methods when compared to pooled results and is theoretically supported. Our FL-TTE enables federated treatment effect estimation across distributed and heterogeneous data in a privacy-preserved way.

A Personalized Predictive Model That Jointly Optimizes Discrimination and Calibration

Precision medicine is accelerating rapidly in the field of health research. This includes fitting predictive models for individual patients based on patient similarity in an attempt to improve model performance. We propose an algorithm which fits a personalized predictive model (PPM) using an optimal size of a similar subpopulation that jointly optimizes model discrimination and calibration, as it is criticized that calibration is not assessed nearly as often as discrimination despite poorly calibrated models being potentially misleading. We define a mixture loss function that considers model discrimination and calibration, and allows for flexibility in emphasizing one performance measure over another. We empirically show that the relationship between the size of subpopulation and calibration is quadratic, which motivates the development of our jointly optimized model. We also investigate the effect of within-population patient weighting on performance and conclude that the size of subpopulation has a larger effect on the predictive performance of the PPM compared to the choice of weight function.

IR-GPT: AI Foundation Models to Optimize Interventional Radiology

Foundation artificial intelligence (AI) models are capable of complex tasks that involve text, medical images, and many other types of data, but have not yet been customized for procedural medicine. This report reviews prior work in deep learning related to interventional radiology (IR), identifying barriers to generalization and deployment at scale. Moreover, this report outlines the potential design of an "IR-GPT" foundation model to provide a unified platform for AI in IR, including data collection, annotation, and training methods-while also contextualizing challenges and highlighting potential downstream applications.

CLEAR-Shock: Contrastive LEARning for Shock

Shock is a life-threatening condition characterized by generalized circulatory failure, which can have devastating consequences if not promptly treated. Thus, early prediction and continuous monitoring of physiological signs are essential for timely intervention. While previous machine learning research in clinical settings has primarily focused on predicting the onset of deteriorating events, the importance of monitoring the ongoing state of a patient's condition post-onset has often been overlooked. In this study, we introduce a novel analytical framework for a prognostic monitoring system that offers hourly predictions of shock occurrence within the next 8 hours preceding its onset, along with forecasts regarding the likelihood of shock continuation within the subsequent hour post-shock occurrence. We categorize the patient's physiological states into four cases: pre-shock (non-shock or shock within the next 8 hours) and post-shock onset (continuation or improvement of shock within the next hour). To effectively predict these cases, we adopt supervised contrastive learning, enabling differential representation in latent space for training a predictive model. Additionally, to extract effective contrastive embeddings, we incorporate a feature tokenizer transformer into our approach. Our framework demonstrates improved predictive performance compared to baseline models when utilizing contrastive embeddings, validated through both internal and external datasets. Clinically, our system significantly improved early detection by identifying shock on average 6 hours before its onset. This framework not only provides early predictions of shock likelihood but also offers real-time assessments of shock persistence risk, thereby facilitating early prevention and evaluation of treatment effectiveness.

Clinical insights: A comprehensive review of language models in medicine

This paper explores the advancements and applications of language models in healthcare, focusing on their clinical use cases. It examines the evolution from early encoder-based systems requiring extensive fine-tuning to state-of-the-art large language and multimodal models capable of integrating text and visual data through in-context learning. The analysis emphasizes locally deployable models, which enhance data privacy and operational autonomy, and their applications in tasks such as text generation, classification, information extraction, and conversational systems. The paper also highlights a structured organization of tasks and a tiered ethical approach, providing a valuable resource for researchers and practitioners, while discussing key challenges related to ethics, evaluation, and implementation.

Advancing Precision Oncology Through Modeling of Longitudinal and Multimodal Data

Cancer evolves continuously over time through a complex interplay of genetic, epigenetic, microenvironmental, and phenotypic changes. This dynamic behavior drives uncontrolled cell growth, metastasis, immune evasion, and therapy resistance, posing challenges for effective monitoring and treatment. However, today's data-driven research in oncology has primarily focused on cross-sectional analysis using data from a single modality, limiting the ability to fully characterize and interpret the disease's dynamic heterogeneity. Advances in multiscale data collection and computational methods now enable the discovery of longitudinal multimodal biomarkers for precision oncology. Longitudinal data reveal patterns of disease progression and treatment response that are not evident from single-timepoint data, enabling timely abnormality detection and dynamic treatment adaptation. Multimodal data integration offers complementary information from diverse sources for more precise risk assessment and targeting of cancer therapy. In this review, we survey methods of longitudinal and multimodal modeling, highlighting their synergy in providing multifaceted insights for personalized care tailored to the unique characteristics of a patient's cancer. We summarize the current challenges and future directions of longitudinal multimodal analysis in advancing precision oncology.

Explainable Machine Learning Model for Predicting Persistent Sepsis-Associated Acute Kidney Injury: Development and Validation Study

BACKGROUND

Persistent sepsis-associated acute kidney injury (SA-AKI) shows poor clinical outcomes and remains a therapeutic challenge for clinicians. Early identification and prediction of persistent SA-AKI are crucial.

OBJECTIVE

The aim of this study was to develop and validate an interpretable machine learning (ML) model that predicts persistent SA-AKI and to compare its diagnostic performance with that of C-C motif chemokine ligand 14 (CCL14) in a prospective cohort.

METHODS

The study used 4 retrospective cohorts and 1 prospective cohort for model derivation and validation. The derivation cohort used the MIMIC-IV database, which was randomly split into 2 parts (80% for model construction and 20% for internal validation). External validation was conducted using subsets of the MIMIC-III dataset and e-ICU dataset, and retrospective cohorts from the intensive care unit (ICU) of Northern Jiangsu People's Hospital. Prospective data from the same ICU were used for validation and comparison with urinary CCL14 biomarker measurements. Acute kidney injury (AKI) was defined based on serum creatinine and urine output, using the Kidney Disease: Improving Global Outcomes (KDIGO) criteria. Routine clinical data within the first 24 hours of ICU admission were collected, and 8 ML algorithms were used to construct the prediction model. Multiple evaluation metrics, including area under the receiver operating characteristic curve (AUC), were used to compare predictive performance. Feature importance was ranked using Shapley Additive Explanations (SHAP), and the final model was explained accordingly. In addition, the model was developed into a web-based application using the Streamlit framework to facilitate its clinical application.

RESULTS

A total of 46,097 patients with sepsis from multiple cohorts were enrolled for analysis. Among 17,928 patients with sepsis in the derivation cohort, 8081 patients (45.1%) showed progression to persistent SA-AKI. Among the 8 ML models, the gradient boosting machine (GBM) model demonstrated superior discriminative ability. Following feature importance ranking, a final interpretable GBM model comprising 12 features (AKI stage, ΔCreatinine, urine output, furosemide dose, BMI, Sequential Organ Failure Assessment score, kidney replacement therapy, mechanical ventilation, lactate, blood urea nitrogen, prothrombin time, and age) was established. The final model accurately predicted the occurrence of persistent SA-AKI in both internal (AUC=0.870) and external validation cohorts (MIMIC-III subset: AUC=0.891; e-ICU dataset: AUC=0.932; Northern Jiangsu People's Hospital retrospective cohort: AUC=0.983). In the prospective cohort, the GBM model outperformed urinary CCL14 in predicting persistent SA-AKI (GBM AUC=0.852 vs CCL14 AUC=0.821). The model has been transformed into an online clinical tool to facilitate its application in clinical settings.

CONCLUSIONS

The interpretable GBM model was shown to successfully and accurately predict the occurrence of persistent SA-AKI, demonstrating good predictive ability in both internal and external validation cohorts. Furthermore, the model was demonstrated to outperform the biomarker CCL14 in prospective cohort validation.

Identification and Validation of an Explainable Prediction Model of Sepsis in Patients With Intracerebral Hemorrhage: Multicenter Retrospective Study

BACKGROUND

Sepsis is a life-threatening condition frequently observed in patients with intracerebral hemorrhage (ICH) who are critically ill. Early and accurate identification and prediction of sepsis are crucial. Machine learning (ML)-based predictive models exhibit promising sepsis prediction capabilities in emergency settings. However, their application in predicting sepsis among patients with ICH is still limited.

OBJECTIVE

The aim of the study is to develop an ML-driven risk calculator for early prediction of sepsis in patients with ICH who are critically ill and to clarify feature importance and explain the model using the Shapley Additive Explanations method.

METHODS

Patients with ICH admitted to the intensive care unit (ICU) from the Medical Information Mart for Intensive Care IV database between 2008 and 2022 were divided into training and internal test sets. The external test was performed using the eICU Collaborative Research Database, which includes over 200,000 ICU admissions across the United States between 2014 and 2015. Sepsis following ICU admission was identified using Sepsis-3.0 through clinical diagnosis combining elevation of the Sequential Organ Failure Assessment by ≥2 points with suspected infection. The Boruta algorithm was used for feature selection, confirming 29 features. Nine ML algorithms were used to construct the prediction models. Predictive performance was compared using several evaluation metrics, including the area under the receiver operating characteristic curve (AUC). The Shapley Additive Explanations technique was used to interpret the final model, and a web-based risk calculator was constructed for clinical practice.

RESULTS

Overall, 2414 patients with ICH were enrolled from the Medical Information Mart for Intensive Care IV database, with 1689 and 725 patients assigned to the training and internal test sets, respectively. An external test set of 2806 patients with ICH from the eICU database was used. Among the 9 ML models tested, the categorical boosting (CatBoost) model demonstrated the best discriminative ability. After reducing features based on their importance, an explainable final CatBoost model was developed using 8 features. The final model accurately predicted sepsis in internal (AUC=0.812) and external (AUC=0.771) tests.

CONCLUSIONS

We constructed a web-based risk calculator with 8 features based on the CatBoost model to assist clinicians in identifying people at high risk for sepsis in patients with ICH who are critically ill.

A U shaped association between the HCT-ALB and hospital mortality in patients with sepsis

The difference between hematocrit and serum albumin (HCT-ALB) demonstrates diagnostic significance in infectious diseases, yet the nonlinear relationship between HCT-ALB and hospital mortality in ICU patients with sepsis remains unexplored. This retrospective multicenter cohort study analyzed 7,546 ICU sepsis patients (mean age 66 ± 16 years) to elucidate the HCT-ALB-mortality relationship. Using Cox proportional hazards models with smooth curve fitting, we identified a U-shaped association: Threshold analysis revealed an inflection point at 6.1. Below this threshold, each unit HCT-ALB increase corresponded to reduced mortality risk (adjusted HR 0.986, 95%CI 0.972-0.999; P = 0.036). Conversely, values ≥ 6.1 predicted escalating risk (adjusted HR 1.048 per unit increase, 95%CI 1.037-1.060; P < 0.0001). Significant age interaction was observed (P for interaction < 0.05), with heightened mortality risk in elderly patients (≥ 65 years: HR 1.022, 95%CI 1.014-1.031). These findings establish HCT-ALB as a non-linear predictor of sepsis outcomes, emphasizing its critical threshold dynamics and age-dependent prognostic implications.

The association between body mass index and ICU 28-day mortality rate in patients with sepsis: A retrospective observational study

OBJECTIVES

Sepsis remains the major cause of mortality among critically ill patients worldwide, indicating the importance of better understanding of its influencing factors for fast recognition and management. Although greater concerns have been raised about the "obesity paradox" and sepsis related mortality, the evidence regarding on overweight or obese septic patients is still controversial. To provide more clinical evidence for the exploration of body mass index (BMI) on sepsis prognostic prediction, we assessed the association of BMI with 28-day mortality of septic patients in intensive care unit (ICU).

METHODS

This was a retrospective observational study with patient data extracted from the eICU Collaborative Research Database. We employed a logistic regression to assess the effect of admission BMI levels on sepsis related mortality risk. Furthermore, the two-piecewise linear model was used to identify BMI mortality thresholds, and BMI-outcome associations were evaluated by interaction tests and subgroup analyses.

RESULTS

Our cohort included a total of 17,454 patients, of whom 1555 (8.91 %) died within 28 days after being admitted to the ICU. The connection between BMI and 28-day mortality in the ICU displayed a U-shaped curve. The threshold effect analysis results in two inflection points of BMI were 23.62 kg/m2 and 45.53 kg/m2. When the BMI was <23.62 kg/m2, the mortality rate decreased by 7 % (95 %CI 0.91, 0.96, P<0.0001) for every 1 increment in the BMI. When the BMI was ≥45.53 kg/m2, the mortality rate increased by 8 % (95 %CI 1.01,1.15, P = 0.0322) for every 1 increment in the BMI. Subgroup analysis showed that neither age nor sex covariates affected the stability of these results (all P for interaction≥0.05).

CONCLUSIONS

In septic ICU patients, the correlation between BMI and 28-day mortality exhibited a U-shaped pattern, indicating that both low and extremely high BMIs were linked to a heightened risk of mortality within 28 days.

Association between osmolality trajectories and mortality in patients with sepsis: a group-based trajectory model in large ICU open access databases

Objective

The regulation of osmolality levels is controlled by the endocrine system, reflecting the body's water and electrolyte balance. However, the relationship between dynamic osmolality trajectories and the prognosis of septic patients has not yet been reported. This study aims to investigate the predictive value of dynamic osmolality trajectories on mortality among patients with sepsis.

Methods

A retrospective analysis was performed using the MIMIC IV and eICU-CRD databases. A total of 19,502 patients were included, 10,263 from MIMIC IV and 9,239 from eICU-CRD. Group-based trajectory modeling (GBTM) analysis was performed to identify distinct osmolality trajectories. The association between these trajectories and in-hospital mortality was assessed by logistic regression analysis and further adjusted for potential confounders. Subgroup analysis was used to identify potential interactive factors and to assess the robustness of the present findings.

Results

Five distinct osmolality trajectories were identified. Patients in the persistent hyperosmolality trajectory (Trajectory-5) had significantly higher in-hospital mortality compared to other trajectories, with an increased risk of in-hospital mortality of 233% (OR 3.33, 95% CI 2.71-4.09) and 150% (OR 2.50, 95% CI 1.97-3.17) in MIMIC IV and eICU-CRD respectively, with Trajectory-2 as reference. A dynamic increase in osmolality (Trajectory-4) was also associated with a 68% (OR 1.68, 95% CI 1.39-2.03) and a 68% (OR 1.68, 95% CI 1.44-1.97) increase in the risk of death, compared with Trajectory-2. Conversely, maintaining osmolality in the range of 290-300 mOsm/L (Trajectory-1 and Trajectory-2) was associated with a lower risk of death. Our results remained stable in the IPWRA and subgroup analyses.

Conclusion

Our findings suggest that dynamic changes in plasma osmolality are significantly associated with in-hospital mortality in septic patients. Osmolality trajectory model provides a potentially effective, easily accessible and cost-effective biomarker for the prognostic assessment and clinical management of sepsis.

Novel pulse pressure pattern monitoring in critical care of elderly sepsis patients

OBJECTIVE

Our research aimed to explore the application of pulse pressure (PP) at the bedside of elderly intensive care unit (ICU) patients with sepsis through a large-scale retrospective cohort study.

METHODS

We obtained data from four heterogeneous datasets, which included information on elderly sepsis patients (≥ 65 years). The data were divided into the inference and validation datasets. Thereby enhancing the generalizability of the study. The primary outcome was mortality at 28 days, and piecewise exponential additive mixed model (PAMM) were employed to estimate the strength of the associations over time.

RESULTS

We included 12,525 elderly patients with sepsis in the initial inference dataset. Based on the PAMM's inference results, we identified a specific low PP phenotype from the time-dependent endpoint dataset. The phenotype indicates an imbalance between the patient's cardiac pumping ability and circulatory resistance, contributing to an increased 28-day mortality (hazard ratio, 2.36; 95% CI, 2.12-2.63). The consistency of these results was validated using data from various sources.

CONCLUSION

Low PP phenotype (PP < 45 mmHg 72 h after intensive care unit admission and lasting for > 3h) may provide an early dynamic warning of the therapeutic effects of resuscitation interventions in long-hospitalized elderly patients with sepsis.

IMPLICATIONS FOR CLINICAL PRACTICE

The results demonstrate acceptable consistency across heterogeneous datasets and hold promise for further development and integration into bedside monitoring systems for elderly sepsis patients.

An interpretable machine learning model for predicting mortality risk in adult ICU patients with acute respiratory distress syndrome

Background

Acute respiratory distress syndrome (ARDS) is a clinical syndrome triggered by pulmonary or extra-pulmonary factors with high mortality and poor prognosis in the ICU. The aim of this study was to develop an interpretable machine learning predictive model to predict the risk of death in patients with ARDS in the ICU.

Methods

The datasets used in this study were obtained from two independent databases: Medical Information Mart for Intensive Care (MIMIC) IV and eICU Collaborative Research Database (eICU-CRD). This study used eight machine learning algorithms to construct predictive models. Recursive feature elimination with cross-validation is used to screen features, and cross-validation-based Bayesian optimization is used to filter the features used to find the optimal combination of hyperparameters for the model. The Shapley additive explanations (SHAP) method is used to explain the decision-making process of the model.

Results

A total of 5,732 patients with severe ADRS were included in this study for analysis, of which 1,171 patients (20.4%) did not survive. Among the eight models, XGBoost performed the best; AUC-ROC was 0.887 (95% CI: 0.863-0.909) and AUPRC was 0.731 (95% CI: 0.673-0.783).

Conclusion

We developed a machine learning-based model for predicting the risk of death of critically ill ARDS patients in the ICU, and our model can effectively identify high-risk ARDS patients at an early stage, thereby supporting clinical decision-making, facilitating early intervention, and improving patient prognosis.

Exploring the prognostic value of the novel nutritional index for in-hospital mortality in acute coronary syndrome: a sex-specific analysis

Background

Emerging evidence suggests that nutritional status plays a pivotal role in determining the prognosis of patients with acute coronary syndrome (ACS). This study aimed to investigate the relationship between a novel nutritional index, Triglycerides × Total Cholesterol × Body Weight Index (TCBI), and short-term prognosis in patients with ACS.

Methods

A retrospective study was conducted using data from 5,277 ACS patients admitted to intensive care units of 208 United States hospitals in the eICU Collaborative Research Database (eICU-CRD) in 2014 and 2015. Patients were divided into three groups based on TCBI tertiles: Group 1 (< 1017.97), Group 2 (1017.97-2069.02), and Group 3 (> 2069.02).

Results

Multivariate logistic regression analysis showed that after adjusting for 17 confounding variables, higher TCBI had significantly lower in-hospital mortality [Tertile 3 vs Tertile 1: OR (95% CI): 0.67 (0.48, 0.94), p = 0.019]. This relationship was significant in the male subgroup but not in the female subgroup. The association between TCBI and in-hospital mortality was more pronounced in male patients and those with blood pressure > 140 mmHg. Subgroup analysis revealed a significant interaction between sex and the predictive value of TCBI (p for interaction < 0.05).

Conclusion

Higher TCBI was independently associated with decreased in-hospital mortality in ACS patients, particularly in male patients. TCBI, as a novel nutritional index, may serve as a practical tool for risk stratification and personalized management of ACS patients.

Development and Validation of a Dynamic Real-Time Risk Prediction Model for Intensive Care Units Patients Based on Longitudinal Irregular Data: Multicenter Retrospective Study

BACKGROUND

Timely and accurate prediction of short-term mortality is critical in intensive care units (ICUs), where patients' conditions change rapidly. Traditional scoring systems, such as the Simplified Acute Physiology Score and Acute Physiology and Chronic Health Evaluation, rely on static variables collected within the first 24 hours of admission and do not account for continuously evolving clinical states. These systems lack real-time adaptability, interpretability, and generalizability. With the increasing availability of high-frequency electronic medical record (EMR) data, machine learning (ML) approaches have emerged as powerful tools to model complex temporal patterns and support dynamic clinical decision-making. However, existing models are often limited by their inability to handle irregular sampling and missing values, and many lack rigorous external validation across institutions.

OBJECTIVE

We aimed to develop a real-time, interpretable risk prediction model that continuously assesses ICU patient mortality using irregular, longitudinal EMR data, with improved performance and generalizability over traditional static scoring systems.

METHODS

A time-aware bidirectional attention-based long short-term memory (TBAL) model was developed using EMR data from the MIMIC-IV (Medical Information Mart for Intensive Care) and eICU Collaborative Research Database (eICU-CRD) databases, comprising 176,344 ICU stays. The model incorporated dynamic variables, including vital signs, laboratory results, and medication data, updated hourly, to perform static and continuous mortality risk assessments. External cross-validation and subgroup sensitivity analyses were conducted to evaluate robustness and fairness. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), accuracy, and F1-score. Interpretability was enhanced using integrated gradients to identify key predictors.

RESULTS

For the static 12-hour to 1-day mortality prediction task, the TBAL model achieved AUROCs of 95.9 (95% CI 94.2-97.5) and 93.3 (95% CI 91.5-95.3) and AUPRCs of 48.5 and 21.6 in MIMIC-IV and eICU-CRD, respectively. Accuracy and F1-scores reached 94.1 and 46.7 in MIMIC-IV and 92.2 and 28.1 in eICU-CRD. In dynamic prediction tasks, AUROCs reached 93.6 (95% CI 93.2-93.9) and 91.9 (95% CI 91.6-92.1), with AUPRCs of 41.3 and 50, respectively. The model maintained high recall for positive cases (82.6% and 79.1% in MIMIC-IV and eICU-CRD). Cross-database validation yielded AUROCs of 81.3 and 76.1, confirming generalizability. Subgroup analysis showed stable performance across age, sex, and severity strata, with top predictors including lactate, vasopressor use, and Glasgow Coma Scale score.

CONCLUSIONS

The TBAL model offers a robust, interpretable, and generalizable solution for dynamic real-time mortality risk prediction in ICU patients. Its ability to adapt to irregular temporal patterns and to provide hourly updated predictions positions it as a promising decision-support tool. Future work should validate its utility in prospective clinical trials and investigate its integration into real-world ICU workflows to enhance patient outcomes.

Association of mean corpuscular volume with 28-day mortality in sepsis patients: A retrospective cohort study using eICU data

INTRODUCTION

The issue of mortality due to sepsis remains a significant concern in the field of medicine. Previous researches have demonstrated an association between mean corpuscular volume (MCV) and mortality from a range of diseases. The objective of this study was to investigate the relationship between MCV and the risk of mortality from sepsis in a large multicentre cohort.

METHOD

A retrospective cohort study was conducted using data from the eICU Collaborative Research Database from 2014-2015. MCV was determined within the initial 24 hours of ICU admission, with patients subsequently classified into quartiles based on their MCV levels. Multivariate regression models were employed to investigate the correlation between MCV and 28-day mortality, with adjustments made for potential confounding factors such as age, sex, body mass index, vital signs and comorbidities. To evaluate the dose-response relationship between MCV and 28-day mortality in patients with sepsis, smoothed curve fitting and threshold effects analysis were utilised.

RESULTS

A total of 9,415 patients with sepsis were included in the study and the 28-day ICU mortality rate of the sepsis patients was 9.38% (883/9415). After adjusting for confounding variables, it was found that the odds ratio (OR) for 28-day mortality was 1.11 (95% CI 1.01, 1.23, P=0.04) increased followed by each 1 fl increase in MCV. The smoothed fitted curves demonstrated a nonlinear positive correlation between MCV and 28-day mortality. The inflection point for the level of MCV was 83 fl. At MCV <83 fl, there was a significant increase in the risk of 28-day mortality with each 1 fl increase in MCV (OR 1.10, 95% CI 1.02, 1.17, P=0.004).

CONCLUSIONS

There is a non-linear positive correlation between MCV and 28-day risk of death in patients with sepsis. Clinicians should be aware of changes in this indicator, especially in patients with high MCV levels.

Anticoagulation in critically ill patients with new-onset atrial fibrillation: Insights from a retrospective cohort study

BACKGROUND

New-onset atrial fibrillation (NOAF) is associated with worse short-term prognosis among critically ill patients, and the benefit of anticoagulation is uncertain.

OBJECTIVE

We aimed to evaluate whether anticoagulation at hospital discharge for critically ill patients presenting with NOAF is associated with improved survival.

METHODS

Retrospective Cohort Study using the Medical Information Mart for Intensive Care (MIMIC)-IV database, which comprises data from patients admitted to the intensive care units (ICUs) at Beth Israel Deaconess Medical Center, Boston, Massachusetts, USA. We included patients diagnosed with NOAF during admission to ICUs between 2008 and 2019 and excluded patients with pre-existing atrial fibrillation (AF), admission related to cardiac surgery and previous anticoagulant use. We did a propensity-score matched (PSM) 1:1 to compare survival between patients discharged on anticoagulation (ACO) vs no anticoagulation and a Cox regression model to assess the primary outcome of long-term survival post-hospital discharge. We censored the patients at 3 different time points and performed another 3-sensitivity analysis to assess the robustness of the results.

RESULTS

A total of 495 patients received ACO therapy, and 2021 patients did not. Matching demonstrated adequate covariate balance, with both groups comprising 495 patients postmatching. Out of hospital mortality rate after PSM was 30.9% vs 38.4% in the ACO and non-ACO groups, respectively. Patients in the ACO group exhibited greater long-term survival (hazard ratio [HR] 0.72, 95% CI 0.58-0.89, P = .003). The benefit was present in the PSM analysis censored at 90 days (HR 0.63, 95% CI 0.43-0.94, P = .02), 1 year (HR 0.55, 95% CI 0.41-0.793, P < .001), and 5 years (HR 0.7, 95% CI 0.56-0.88, P = .002).These findings also remained consistent across the 3 other sensitivity analyses performed.

CONCLUSION

Anticoagulant prescription at hospital discharge for patients with NOAF during ICU stay was associated with longer survival.

Distinct immunological signatures define three sepsis recovery trajectories: a multi-cohort machine learning study

Importance

Understanding heterogeneous recovery patterns in sepsis is crucial for personalizing treatment strategies and improving outcomes.

Objective

To identify distinct recovery trajectories in sepsis and develop a prediction model using early clinical and immunological markers.

Design setting and participants

Retrospective cohort study using data from 28,745 adult patients admitted to 12 intensive care units (ICUs) with sepsis between January 2014 and December 2024.

Main outcomes and measures

Primary outcome was the 28-day trajectory of Sequential Organ Failure Assessment (SOFA) scores. Secondary outcomes included 90-day mortality and hospital length of stay.

Results

Among 24,450 eligible patients (mean [SD] age, 64.5 [15.3] years; 54.2% male), three distinct recovery trajectories were identified: rapid recovery (42.3%), slow recovery (35.8%), and deterioration (21.9%). The machine learning model achieved an AUROC of 0.85 (95% CI, 0.83-0.87) for trajectory prediction. Key predictors included initial SOFA score, lactate levels, and inflammatory markers. Mortality rates were 12.3, 28.7, and 45.6% for rapid, slow, and deterioration groups, respectively.

Conclusions and relevance

Early prediction of sepsis recovery trajectories is feasible and may facilitate personalized treatment strategies. The developed model could assist clinical decision-making and resource allocation in critical care settings.

Inter-organ correlation based multi-task deep learning model for dynamically predicting functional deterioration in multiple organ systems of ICU patients

BACKGROUND

Functional deterioration (FD) of various organ systems is the major cause of death in ICU patients, but few studies propose effective multi-task (MT) model to predict FD of multiple organs simultaneously. This study propose a MT deep learning model named inter-organ correlation based multi-task model (IOC-MT), to dynamically predict FD in six organ systems.

METHODS

Three public ICU databases were used for model training and validation. The IOC-MT was designed based on the routine MT deep learning framework, but it used a Graph Attention Networks (GAT) module to capture inter-organ correlation and an adaptive adjustment mechanism (AAM) to adjust prediction. We compared the IOC-MT to five single-task (ST) baseline models, including three deep models (LSTM-ST, GRU-ST, Transformer-ST) and two machine learning models (GRU-ST, RF-ST), and performed ablation study to assess the contribution of important components in IOC-MT. Model discrimination was evaluated by AUROC and AUPRC, and model calibration was assessed by the calibration curve. The attention weight and adjustment coefficient were analyzed at both overall and individual level to show the AAM of IOC-MT.

RESULTS

The IOC-MT had comparable discrimination and calibration to LSTM-ST, GRU-ST and Transformer-ST for most organs under different gap windows in the internal and external validation, and obviously outperformed GRU-ST, RF-ST. The ablation study showed that the GAT, AAM and missing indicator could improve the overall performance of the model. Furthermore, the inter-organ correlation and prediction adjustment of IOC-MT were intuitive and comprehensible, and also had biological plausibility.

CONCLUSIONS

The IOC-MT is a promising MT model for dynamically predicting FD in six organ systems. It can capture inter-organ correlation and adjust the prediction for one organ based on aggregated information from the other organs.

A Predictive Model for Acute Kidney Injury Based on Leukocyte-Related Indicators in Hepatocellular Carcinoma Patients Admitted to the Intensive Care Unit

Background: This study aimed to develop and validate a straightforward clinical risk model utilizing white blood cell (WBC) counts to predict acute kidney injury (AKI) in critically sick patients with hepatocellular carcinoma (HCC). Methods: Data were taken from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database for the training cohort. Data for an internal validation cohort were obtained from the eICU Collaborative Research Database (eICU-CRD), while patients from our hospital were utilized for external validation. A risk model was created utilizing significant indicators identified through multivariate logistic regression, following logistic regression analysis to determine the primary predictors of WBC-related biomarkers for AKI prediction. The Kaplan-Meier curve was employed to evaluate the prognostic efficacy of the new risk model. Results: A total of 1628 critically sick HCC patients were enrolled. Among these, 23 (23.2%) patients at our hospital, 84 (17.9%) patients in the eICU-CRD database, and 379 (35.8%) patients in the MIMIC-IV database developed AKI. A unique risk model was developed based on leukocyte-related indicators following the multivariate logistic regression analysis, incorporating white blood cell to neutrophil ratio (WNR), white blood cell to monocyte ratio (WMR), white blood cell to hemoglobin ratio (WHR), and platelet to lymphocyte ratio (PLR). This risk model exhibited robust predictive capability for AKI, in-hospital mortality, and ICU mortality across the training set, internal validation set, and external validation set. Conclusion: This risk model seems to have practical consequences as an innovative and accessible tool for forecasting the prognosis of critically ill HCC patients, which may, to some degree, aid in identifying equitable risk assessments and treatment strategies.

Clinical subtypes identification and feature recognition of sepsis leukocyte trajectories based on machine learning

Sepsis is a highly variable condition, and tracking leukocyte patterns may offer insights for tailored treatment and prognosis. We used the MIMIC-IV database to analyze patients diagnosed with Sepsis-3 within 24 h of ICU admission. Latent class mixed models (LCMM) were applied to leukocyte trajectories to identify sepsis subtypes. The primary outcome was 28-day all-cause mortality, with secondary outcomes including the need for life-support therapies. Associations between leukocyte trajectories and outcomes were assessed using multivariate regression, and findings were externally validated with the eICU database. Use the XGBoost model to identify baseline characteristics of high-risk mortality sepsis subgroups for predicting subgroup allocation upon patient admission to the ICU, and apply the SHAP method to interpret the contributing variables of the model. Among 7410 sepsis patients, eight distinct leukocyte trajectory subtypes were identified. Among those subtypes, patients with persistently high leukocyte levels had the poorest prognosis (HR 3.00; 95% CI 2.48-3.62) and a significantly greater need for life-support therapies; Patients with persistently low white blood cell levels had a higher risk of death (HR 1.68; 95% CI 1.24-2.27) but were less likely to receive invasive mechanical ventilation. Incorporating early ICU baseline variables into an XGBoost algorithm enables effective prediction of high-mortality risk subgroups (AUC > 0.8). SHAP method reveals distinct early clinical characteristics between hyperinflammatory subtypes (class 4, 7, and 8) and the hypoinflammatory subtype (class 1). In ICU-admitted sepsis patients, eight leukocyte trajectories are identified, which is the key independent predictors of prognosis, separating from single leukocyte measurements. High-mortality risk subgroups exhibit distinct clinical characteristics at ICU admission, providing valuable insights for their prediction and personalized early intervention.

Comprehensive analyses: Using machine learning models for mortality prediction in the intensive care unit of internal medicine

Mortality prediction in the intensive care unit (ICU) is essential in patient management. Emerging methods such as machine learning (ML) can be employed to predict ICU patients' mortality. Patients receiving treatment in the ICU of the internal medicine department were subjected to ML analysis upon admission, considering demographic, laboratory, and medical scores. Data from 787 internal medicine ICU patients were analyzed, with only a subset (220) included in the study for the 30-day mortality prediction model. The performance of boosting and Logistic Regression models in mortality prediction was compared. Categorical boosting (CatBoost) achieved the highest area under the curve (AUC) of 0.90, while extreme gradient boosting reached a maximum AUC of 0.85, and Logistic Regression attained the highest AUC of 0.83. Incorporating Acute Physiology and Chronic Health Evaluation II, Simplified Acute Physiology Score II, and Sequential Organ Failure Assessment scores with clinical and laboratory values, CatBoost demonstrated the strongest predictive performance with high sensitivity and specificity. In the ICU of the internal medicine department, it was concluded that the ML models successfully predict mortality.

The association between albumin corrected calcium levels and mortality in ICU patients undergoing maintenance hemodialysis

While the relationship between albumin corrected calcium (ACC) levels and unfavourable outcomes has been previously investigated, existing studies have been limited in their specific application to patients undergoing maintenance hemodialysis (MHD) in intensive care unit (ICU). This retrospective cohort study aimed to explore the association between baseline ACC and 28-day in-hospital mortality in ICU patients undergoing MHD. Logistic regression model, smooth curve fitting, piecewise linear regression, subgroup analyses, and a series of sensitivity analyses were employed. Of the 2114 patients with a median age of 64 years, 290 (13.72%) died within 28 days after ICU admission. Multivariate regression analysis revealed that, in comparison with the lowest tertile, the highest tertile of ACC was associated with a higher mortality rate (OR 1.69, 95% CI 1.09-1.53, P = 0.0032). When the ACC levels were < 8.04 mg/dL, the mortality rate decreased with an adjusted OR of 0.44 (95% CI 0.20-0.98, P = 0.0438) for every 1 mg/dL increase in the ACC levels. When the ACC levels were ≥ 8.04 mg/dl, the mortality rate increased with an adjusted OR of 1.36 (95% CI 1.13-1.64, P = 0.0011) for every 1 mg/dl increase in the ACC levels. Non-linear relationship between ACC and 28-day in-hospital mortality were identified in patients undergoing MHD in the ICU. However, the findings of this study need to be confirmed through prospective studies.

Deciphering the impact of sepsis phenotypes on improving clinical outcome predictions: a multicenter retrospective analysis based on critical care in China

Sepsis is a clinically heterogeneous disease with high mortality. It is crucial to develop relevant therapeutic strategies for different sepsis phenotypes, but the impact of phenotypes on patients' clinical outcomes is unclear. This study aimed to identify potential sepsis phenotypes using readily available clinical parameters and assess their predictive value for 28-day clinical outcomes by logistic regression analysis. In this retrospective analysis, researchers extracted clinical data from adult patients admitted to the First Affiliated Hospital of Anhui Medical University between April and August 2022 and from the 2014-2015 eICU Collaborative Study database. K-Means clustering was utilized to identify and refine sepsis phenotypes, and their predictive performance was subsequently evaluated. Logistic regression models were trained independently for each phenotype and five-fold cross-validation was used to predict clinical outcomes. Predictive accuracy was then compared to traditional non-clustered prediction methods using model assessment scores. The study cohort consisted of 250 patients from the First Affiliated Hospital of Anhui Medical University, allocated in a 7:3 ratio for training and testing, respectively, and an external validation cohort of 3100 patients from the eICU Cooperative Research Database. The results of the phenotype-based prediction model demonstrated an improvement in F1 score from 0.74 to 0.82 and AUC from 0.74(95%CI 0.71-0.80) to 0.84(95%CI 0.82-0.87), and these results also highlight the superiority of clinical outcome prediction with the help of sepsis phenotypes over traditional prediction methods. Phenotype-based prediction of 28-day clinical outcomes in sepsis demonstrated significant advantages over traditional models, highlighting the impact of phenotype-driven modeling on clinical outcomes in sepsis.

The Impact of Intraoperative Respiratory Patterns on Morbidity and Mortality in Patients with COPD Undergoing Elective Surgery

Background/Objectives: Surgical procedures in chronic obstructive pulmonary disease (COPD) patients carry a high risk of postoperative respiratory failure, often causing the need for mechanical ventilation and prolonged intensive care unit (ICU) stays. Accompanying COPD with heart failure further increases the risk of complications. This study aimed to identify predictors of mortality, prolonged ICU and hospital stays, the need for mechanical ventilation, and vasoactive drug usage in ICU patients with moderate to severe COPD undergoing elective non-cardiac surgery. Methods: This retrospective cohort study analyzed eICU-CRD data, including adult patients with moderate to severe COPD admitted to the ICU from the operating room following elective non-cardiac surgery. Spearman's correlation analysis was performed to assess associations between intraoperative ventilation parameters and ICU/hospital length of stay, postoperative laboratory parameters, and their perioperative dynamics. Results: This study included 680 patients (21% with severe COPD). Hospital and ICU mortality were 8.6% and 4.4%, respectively. Median ICU and hospital stays were 1.9 and 6.6 days, respectively. Intraoperative tidal volume, expired minute ventilation, positive end-expiratory pressure, mean airway pressure, peak inspiratory pressure, and compliance had no statistically significant association with mortality, postoperative mechanical ventilation, its duration, or the use of vasopressors/inotropes. Tidal volume correlated positively with changes in monocyte count (R = 0.611; p = 0.016), postoperative lymphocytes (R = 0.327; p = 0.017), and neutrophil count (R = 0.332; p = 0.02). Plateau pressure showed a strong positive association with the neutrophil-to-lymphocyte ratio (R = 0.708; p = 0.001). Conclusions: Intraoperative ventilation modes and parameters in COPD patients appear to have no significant impact on the outcomes or laboratory markers, except possibly for the neutrophil-to-lymphocyte ratio, although its elevation cause remains unclear.

Observational Study of Microbial Colonization and Infection in Neurological Intensive Care Patients Based on Electronic Health Records

Background/Objectives: Patients with central nervous system injuries who are hospitalized in intensive care units (ICUs) are at high risk for nosocomial infections. Limited data are available on the incidence and patterns of microbial colonization and infection in this patient population. Methods: To fill this gap, we performed an electronic health record-based study of 1614 chronic patients with brain injury admitted to the ICU from 2017 to 2023. Results: Among the infectious complications, pneumonia was the most common (n = 879; 54.46%). Sepsis was diagnosed in 54 patients, of whom 46 (85%) were diagnosed with pneumonia. The only pathogen that showed an association with the development of pneumonia and sepsis in colonized patients was Pseudomonas aeruginosa (pneumonia: p = 7.2 × 10-9; sepsis: p = 1.7 × 10-5). Bacterial isolates from patients with and without pneumonia did not differ in pathogen titer or dynamics, but patients with monomicrobial culture were more likely to develop pneumonia than patients with polymicrobial culture (1 vs. 2 pathogens, p = 0.014; 1 + 2 pathogens vs. 3 + 4 pathogens, p = 2.8 × 10-6), although the pathogen titer was lower in monoculture than in polyculture. Bacterial isolates from all patients and all culture sites showed high levels of multidrug resistance (Gram-negative bacteria: 88-100%; Gram-positive bacteria: 48-97%), with no differences in multidrug-resistant organism (MDRO) colonization and infection rates. Conclusions: Our results highlight the high burden of MDROs in neurological ICUs and provide novel ecosystem-based insights into mono- and polymicrobial colonization and infection development. These findings may be useful for developing strategies to protect against infections.