The eICU Collaborative Research Database is a large multi-center critical care database made available by Philips Healthcare in partnership with the MIT Laboratory for Computational Physiology.

The eICU Collaborative Research Database holds data associated with over 200,000 patient stays, providing a large sample size for research studies.

The eICU Collaborative Research Database is a large multi-center critical care database made available by Philips Healthcare in partnership with the MIT Laboratory for Computational Physiology.

This is a preprocessed dataset derived from patient records in MIMIC-III and eICU, two large-scale electronic health record (EHR) databases. It contains features and labels for 5 prediction tasks involving 3 adverse outcomes (prediction times listed in parentheses): in-hospital mortality (48h), acute respiratory failure (4h and 12h), and shock (4h and 12h). We extracted comprehensive, high-dimensional feature representations (up to ~8,000 features) using FIDDLE (FlexIble Data-Driven pipeLinE), an open-source preprocessing pipeline for structured clinical data. These 5 prediction tasks were designed in consultation with a critical care physician for their clinical importance, and were used as part of the proof-of-concept experiments in the original paper to demonstrate FIDDLE's utility in aiding the feature engineering step of machine learning model development. The intent of this release is to share preprocessed MIMIC-III and eICU datasets used in the experiments to support and enable reproducible machine learning research on EHR data.

Background: It remains unclear whether the mean vancomycin trough concentration (VTC) derived from the entire course of therapy is of potential benefit for critically ill patients. This study was conducted to explore the association between mean serum VTC and mortality in intensive care units (ICUs).Methods: 3,603 adult patients with two or more VTC records after receiving vancomycin treatment in the eICU Collaborative Research Database were included in this multicenter retrospective cohort study. Mean VTC was estimated using all measured VTCs and investigated as a continuous and categorical variable. Patients were categorised into four groups according to mean VTC: 20 mg/L. Multivariable logistic regression and subgroup analyses were performed to investigate the relationship of mean VTC with mortality.Results: After adjusting for a series of covariates, logistic regression analyses indicated that mean VTC, as a continuous variable, was positively correlated with ICU (odds ratio, 1.038, 95% confidence interval, [1.014–1.063]) and hospital (1.025 [1.005–1.046]) mortalities. As a categorical variable, mean VTC of 10–15 mg/L was not associated with reduced ICU (1.705 [0.975–2.981]) and hospital (1.235 [0.829–1.841]) mortalities. Mean VTC of 15–20 mg/L was not correlated with a lower risk of hospital mortality (1.370 [0.924–2.029]). Moreover, mean VTCs of 15–20 and >20 mg/L were significantly associated with higher ICU mortality (1.924 [1.111–3.332]; 2.428 [1.385–4.258]), and mean VTC of >20 mg/L with higher hospital mortality (1.585 [1.053–2.387]) than mean VTC of

BackgroundThe research on the impact of dynamic corrected anion gap (cAG) on prognosis is scarce.ObjectiveThis study aimed to investigate the relationship between changes in cAG (ΔcAG) during intensive care unit (ICU) hospitalization and mortality.MethodsIn this multicenter, retrospective cohort study, patients with both initial and final records of serum sodium, potassium, chloride, bicarbonate, and albumin were recruited from the eICU Collaborative Research Database. Two cohorts were included in the study: cohort A (final cAG > initial cAG) and cohort B (final cAG < initial cAG). Multivariable logistic regression was utilized to assess the association between mortality and ΔcAG in each cohort. ΔcAG was calculated as shown as follows: ΔcAG=|final cAG ‐ initial cAG|initial cAG×100%.ResultsAmong the 11,216 enrolled patients, 4,147 (37%) individuals were classified into cohort A, while 7,069 (63%) patients were assigned to cohort B. In cohort A, for every 10% increase in ΔcAG, ICU and hospital mortalities increased by 46.1% (odds ratio: 1.461, 95% confidence interval [1.378, 1.548]) and 55.5% (1.555 [1.467, 1.648]), respectively. Interaction and subgroup analyses demonstrated consistent results among patients with different Acute Physiology and Chronic Health Evaluation Ⅳ (APACHE Ⅳ) scores (≤58 vs. >58), time interval (≤97 h vs. >97 h) and initial cAG (≤16 mEq/L vs. >16 mEq/L). Meanwhile, in cohort B, ICU and hospital mortalities decreased by 31.4% (0.686 [0.619, 0.759]) and 29.4% (0.706 [0.651, 0.764]), respectively, with each 10% increase in ΔcAG, especially among patients with higher APACHE IV scores (>62) and initial cAG (>16 mEq/L). When analyzed categorically, the ΔcAG still exhibited a significant risk gradient across quartiles.ConclusionFurther elevated cAG after ICU admission demonstrates a robust association with an increased mortality risk in critically ill patients. ICU patients with higher APACHE Ⅳ scores or initial cAG may benefit from measures aimed at reducing cAG.

Background and ObjectivesFew studies have evaluated the impact of red blood cell distribution width (RDW) on prognosis for critically ill patients with acute stroke according to recent studies. The aim of this study was to investigate the association between RDW and mortality in these patients.MethodsClinical data were extracted from the eICU Collaborative Research Database (eICU-CRD) and analyzed. The exposure of interest was RDW measured at admission. The primary outcome was in-hospital mortality. Binary logistic regression models and interaction testing were performed to examine the RDW-mortality relationship and effect modification by acute myocardial infarction and hypertension (HP).ResultsData from 10,022 patients were analyzed. In binary logistic regression analysis, after adjusting for potential confounders, RDW was found to be independently associated with in-hospital mortality {odds ratio (OR) 1.07, [95% confidence interval (CI) 1.03 to 1.11]; p = 0.001}. Higher RDW linked to an increase in mortality (OR, 1.07; 95% CI, 1.03 to 1.11; P for trend < 0.0001). Subgroup analysis showed that, in patients combined with AMI and without HP (both P-interaction

Objective: Congestive heart failure (CHF) is a clinical syndrome in which heart disease progresses to a severe stage. Risk assessment and early diagnosis of death in patients with CHF are critical to patient prognosis and treatment. The purpose of this study was to establish a nomogram predicting in-hospital death for CHF patients in the ICU. Design: A retrospective observational cohort study. Setting and participants: The data of study from 30,411 CHF patients in the Medical Information Mart for Intensive Care (MIMIC-IV) database and the eICU Collaborative Research Database (eICU-CRD). Primary outcome: In-hospital mortality. Results: The inclusion criteria were met by 15983 subjects, whose in-hospital mortality rate was 12.4%. Multivariate analysis determined that the independent risk factors were age, race, norepinephrine, dopamine, phenylephrine, vasopressin, mechanical ventilation, intubation, HepF, heart rate, respiratory rate, temperature, SBP, AG, BUN, creatinine, chloride, MCV, ...

IntroductionSepsis associated thrombocytopenia (SAT) is a common complication of sepsis. We designed this study to investigate factors influencing SAT.MethodsPatients with sepsis (2984 in Peking union medical college hospital [PUMCH] database, 13165 in eICU Collaborative Research [eICU] database, 11101 in Medical Information Mart for Intensive Care IV [MIMIC-IV] database) were enrolled. Variables included basic information, comorbidities, and organ functions. Multi-variable logistic regression models and artificial neural network model were applied to determine the factors related to SAT.Main resultsAge and body mass index (BMI) were inversely correlated with the incidence of SAT (p-value 0.175 and 0.049 [PUMCH], p-value 0.000 and 0.000 [eICU], p-value 0.000 and 0.000 [MIMIC-IV]). Hematologic malignancies and other malignancies were positively correlated with the incidence of SAT (p-value 0.000 and 0.000 [PUMCH], p-value 0.000 and 0.000 [eICU], p-value 0.000 and 0.020 [MIMIC-IV]) except other malignancies was inversely correlated with the incidence of SAT in PUMCH database. Norepinephrine (NE) equivalents, total bilirubin (TBIL) and creatinine were positively correlated with the incidence of SAT (p-value 0.000, 0.000 and 0.011 [PUMCH], p-value 0.028, 0.000 and 0.013 [eICU], p-value 0.028, 0.000 and 0.027 [MIMIC-IV]). PaO2 / FiO2 was inversely correlated with the incidence of SAT in PUMCH database (p-value 0.021 [PUMCH]), while it was positively correlated with the incidence of SAT (p-value 0.000 [MIMIC-IV]). PaO2 / FiO2 and SAT was not related (p-value 0.111 [eICU]). TBIL, hematologic malignancies, PaO2 / FiO2 and NE equivalents ranked in the top five significant variables in all three datasets.ConclusionsHematologic malignancies and other malignancies were positively correlated with the incidence of SAT. NE equivalents, TBIL and creatinine were positively correlated with the incidence of SAT. TBIL, hematologic malignancies, PaO2 / FiO2 and NE equivalents ranked in the top significant variables in factors influencing SAT.

Pulmonary embolism (PE) is associated with acute kidney injury (AKI). This study aimed to develop a nomogram to predict AKI in PE patients admitted to the intensive care unit. The data of patients with PE were obtained from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) and the eICU Collaborative Research Database, with AKI as the primary outcome. Patients from MIMIC-IV were divided into training (80%) and internal validation (20%) cohorts, and external validation was performed using the eICU. Independent risk factors for AKI were identified using univariable logistic regression and stepwise logistic regression. A nomogram was constructed based on the stepwise analysis. Its performance was evaluated using the receiver operating characteristic (ROC) area under the curve (AUC), calibration plots, decision curve analysis (DCA), and sensitivity analysis, and compared to the simplified acute physiology score (SAPS) II score. Six independent risk factors for AKI were identified. The nomogram’s AUC was 0.717 in the training cohort, 0.758 in the internal validation cohort, and 0.889 in the external validation cohort. The AUC of the nomogram was higher than the SAPS II score (p 

BackgroundThis study aimed to investigate the association between blood urea nitrogen to serum albumin ratio (BAR) and the risk of in-hospital mortality in patients with diabetic ketoacidosis.MethodsA total of 3,962 diabetic ketoacidosis patients from the eICU Collaborative Research Database were included in this analysis. The primary outcome was in-hospital death.ResultsOver a median length of hospital stay of 3.1 days, 86 in-hospital deaths were identified. One unit increase in LnBAR was positively associated with the risk of in-hospital death (hazard ratio [HR], 1.82 [95% CI, 1.42–2.34]). Furthermore, a nonlinear, consistently increasing correlation between elevated BAR and in-hospital mortality was observed (P for trend =0.005 after multiple-adjusted). When BAR was categorized into quartiles, the higher risk of in-hospital death (multiple-adjusted HR, 1.99 [95% CI, (1.1–3.6)]) was found in participants in quartiles 3 to 4 (BAR≥6.28) compared with those in quartiles 1 to 2 (BAR

Demographics of patients with AKI.

ObjectiveIn an era of increasingly expensive intensive care costs, it is essential to evaluate early whether the length of stay (LOS) in the intensive care unit (ICU) of obesity patients with sepsis will be prolonged. On the one hand, it can reduce costs; on the other hand, it can reduce nosocomial infection. Therefore, this study aimed to verify whether ICU prolonged LOS was significantly associated with poor prognosis poor in obesity patients with sepsis and develop a simple prediction model to personalize the risk of ICU prolonged LOS for obesity patients with sepsis.MethodIn total, 14,483 patients from the eICU Collaborative Research Database were randomized to the training set (3,606 patients) and validation set (1,600 patients). The potential predictors of ICU prolonged LOS among various factors were identified using logistic regression analysis. For internal and external validation, a nomogram was developed and performed.ResultsICU prolonged LOS was defined as the third quartile of ICU LOS or more for all sepsis patients and demonstrated to be significantly associated with the mortality in ICU by logistic regression analysis. When entering the ICU, seven independent risk factors were identified: maximum white blood cell, minimum white blood cell, use of ventilation, Glasgow Coma Scale, minimum albumin, maximum respiratory rate, and minimum red blood cell distribution width. In the internal validation set, the area under the curve was 0.73, while in the external validation set, it was 0.78. The calibration curves showed that this model predicted probability due to actually observed probability. Furthermore, the decision curve analysis and clinical impact curve showed that the nomogram had a high clinical net benefit.ConclusionIn obesity patients with sepsis, we created a novel nomogram to predict the risk of ICU prolonged LOS. This prediction model is accurate and reliable, and it can assist patients and clinicians in determining prognosis and making clinical decisions.

PurposeThis study aims to develop and validate a nomogram for predicting the risk of bloodstream infections (BSI) in critically ill patients based on their admission status to the Intensive Care Unit (ICU).Patients and methodsPatients’ data were extracted from the Medical Information Mart for Intensive Care−IV (MIMIC−IV) database (training set), the Beijing Friendship Hospital (BFH) database (validation set) and the eICU Collaborative Research Database (eICU−CRD) (validation set). Univariate logistic regression analyses were used to analyze the influencing factors, and lasso regression was used to select the predictive factors. Model performance was assessed using area under receiver operating characteristic curve (AUROC) and Presented as a Nomogram. Various aspects of the established predictive nomogram were evaluated, including discrimination, calibration, and clinical utility.ResultsThe model dataset consisted of 14930 patients (1444 BSI patients) from the MIMIC-IV database, divided into the training and internal validation datasets in a 7:3 ratio. The eICU dataset included 2100 patients (100 with BSI) as the eICU validation dataset, and the BFH dataset included 419 patients (21 with BSI) as the BFH validation dataset. The nomogram was constructed based on Glasgow Coma Scale (GCS), sepsis related organ failure assessment (SOFA) score, temperature, heart rate, respiratory rate, white blood cell (WBC), red width of distribution (RDW), renal replacement therapy and presence of liver disease on their admission status to the ICU. The AUROCs were 0.83 (CI 95%:0.81-0.84) in the training dataset, 0.88 (CI 95%:0.88-0.96) in the BFH validation dataset, and 0.75 (95%CI 0.70-0.79) in the eICU validation dataset. The clinical effect curve and decision curve showed that most areas of the decision curve of this model were greater than 0, indicating that this model has a certain clinical effectiveness.ConclusionThe nomogram developed in this study provides a valuable tool for clinicians and nurses to assess individual risk, enabling them to identify patients at a high risk of bloodstream infections in the ICU.

PurposeThis study aimed to examine the correlation between lactate levels and 28 days mortality in elderly sepsis patients across different Sequential Organ Failure Assessment (SOFA) score groups following ICU admission.MethodsA multi-center retrospective cohort study utilized data from the eICU Collaborative Research Database, encompassing elderly sepsis patients from 208 ICUs across the United States during 2014–2015. Lactate levels and SOFA scores at admission were collected, with a focus on 28 days mortality post-ICU admission. A two-piece-wise linear regression model was developed to assess the threshold effects of lactate on outcomes and its variation across SOFA score categories. Smooth curve fitting was utilized.ResultsOf the 5,150 patients with a median age of 76 years, 711 (13.8%) died within 28 days of ICU admission. A positive correlation was noted between lactate levels and mortality when lactate was < 3.7 mmol/l, with an adjusted odds ratio (OR) of 1.33 (95% CI: 1.17–1.51, P < 0.0001) for each increment in lactate. For lactate levels ≥ 3.7 mmol/L, mortality increased with an adjusted OR of 1.11 (95% CI: 1.05–1.18, P = 0.0003) for each increment in lactate. Moreover, mortality was low and rose gradually with increasing lactate levels in the SOFA score ≤ 5 group. Conversely, in the SOFA score > 5 group, mortality increased significantly, particularly when lactate levels exceeded 5 mmol/L.ConclusionThis study reveals a non-linear positive relationship between lactate and 28 days mortality among elderly sepsis patients. Furthermore, stratification by SOFA score demonstrated that patients with higher scores exhibited a heightened risk of mortality as lactate levels increased.

Advantages and disadvantages of each models.

BackgroundThe mean perfusion pressure (MPP) was recently proposed to personalize tissue perfusion pressure management in critically ill patients. Severe fluctuation in MPP may be associated with adverse outcomes. We sought to determine if higher MPP variability was correlated with increased mortality in critically ill patients with CVP monitoring.MethodsWe designed a retrospective observational study and analyzed data stored in the eICU Collaborative Research Database. Validation test was conducted in MIMIC-III database. The exposure was the coefficient of variation (CV) of MPP in the primary analyses, using the first 24 hours MPP data recorded within 72 hours in the first ICU stay. Primary endpoint was in-hospital mortality.ResultsA total of 6,111 patients were included. The in-hospital mortality of 17.6% and the median MPP-CV was 12.3%. Non-survivors had significantly higher MPP-CV than survivors (13.0% vs 12.2%, p 19.2%) were associated with increased risk of hospital mortality compared with those in the fifth and sixth decile (adjusted OR: 1.38, 95% Cl: 1.07–1.78). These relationships remained remarkable in the multiple sensitivity analyses. The validation test with 4,153 individuals also confirmed the results when MPP-CV > 21.3% (adjusted OR: 1.46, 95% Cl: 1.05–2.03).ConclusionsSevere fluctuation in MPP was associated with increased short-term mortality in critically ill patients with CVP monitoring.

Comparison of mortality prediction performance among the four models of AKI patients.

Background: Acute respiratory distress syndrome (ARDS) is a common cause of respiratory failure in patients in intensive care unit (ICU). The therapeutic value of glucocorticoids (GCs) in the prognosis of ARDS remains controversial. The aim of this research is studying the impacts of GCs treatment on ARDS patients in ICU.Methods: We retrospectively studied 2,167 ARDS patients whose data were collected from the public eICU Collaborative Research Database, among which 254 patients who received glucocorticoid (GCs) treatment were 1:1 matched by propensity matching analysis (PSM). The primary outcome was ICU mortality. Every oxygenation index (PaO2/FiO2) measurement before death or ICU discharge was recorded. A joint model (JM) which combined longitudinal sub-model (mixed-effect model) and time-to-event sub-model (Cox regression model) by trajectory functions of PaO2/FiO2 was conducted to determine the effects of GCs treatment on both ICU mortality and PaO2/FiO2 level and further PaO2/FiO2's effect on event status. The marginal structural cox model (MSCM) adjusted the overall PaO2/FiO2 of patients to further validate the results.Results: The result of the survival sub-model showed that GCs treatment was significantly associated with reduced ICU mortality in ARDS patients [HR (95% CI) = 0.642 (0.453, 0.912)], demonstrating that GCs treatment was a protective factor of ICU mortality. In the longitudinal sub-model, GCs treatment was not correlated to the PaO2/FiO2. After adjusted by the JM, the HR of GCs treatment was 0.602 while GCs was still not significantly related to PaO2/FiO2 level. The JM-induced association showed that higher PaO2/FiO2 was a significant protective factor of mortality in ARDS patients and the HR was 0.991 which demonstrated that one level increase of PaO2/FiO2 level decreased 0.9% risk of ICU mortality. MSCM results also show that GCs can improve the prognosis of patients.Conclusion: Rational use of GCs can reduce the ICU mortality of ARDS patients in ICU. In addition to the use of GCs treatment, clinicians should also focus on the shifting trend of PaO2/FiO2 level to provide better conditions for patients' survival.

BackgroundPrior 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.MethodPatient 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.ResultThe 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 

BackgroundThe triglyceride-glucose (TyG) index is a marker for insulin resistance (IR) linked to diabetes complications and poor outcomes. Its connection to all-cause mortality in non-diabetic critically ill patients is unknown. This study aims to investigate the TyG index’s impact on mortality in this population, evaluating how IR affects their prognosis.MethodsThis study is retrospective observational research utilizing data from the eICU Collaborative Research Database. A total of 14,089 non-diabetic critically ill patients were included and categorized into three groups based on the TyG index measured on the first day of admission (T1, T2, and T3). Kaplan-Meier survival analysis was performed to compare the 28-day mortality rates among the different groups. Cox proportional hazards models were used to assess the relationship between the TyG index and 28-day mortality. Additionally, we conducted sensitivity analyses, subgroup analyses, and interaction analyses to assess the robustness of the results.ResultsDuring the observation period, 730 patients (5.18%) died in the ICU, while 1,178 patients (8.36%) died in the hospital. The 28-day ICU mortality rate and hospital mortality rate significantly increased with higher TyG index values (P < 0.001). Cox proportional hazards models were used to assess the relationship between the TyG index and 28-day mortality. Specifically, Cox proportional hazards models were used to assess the relationship between the TyG index and 28-day mortality. Furthermore, the analysis showed a nonlinear effect of the TyG index on mortality in non-diabetic critically ill patients, with a critical point at 9.94. While Below 9.94, ICU and hospital mortality rates rose with higher TyG index values. But above 9.94, mortality didn’t significantly increase despite further rises in the TyG index. Sensitivity and subgroup analyses confirmed the robustness of these results, and E-value analysis indicated strong resistance to unmeasured confounding factors.ConclusionThe TyG index demonstrates a significant positive correlation with all-cause mortality in non-diabetic critically ill patients, exhibiting a nonlinear relationship. Consequently, the TyG index serves as a crucial tool for identifying high-risk patients, thereby assisting clinicians in formulating more effective monitoring and intervention strategies.