Collaborative Research Database Research Articles

Identification of Distinct Clinical Phenotypes of Heterogeneous Mechanically Ventilated ICU Patients Using Cluster Analysis

This retrospective study aimed to derive the clinical phenotypes of ventilated ICU patients to predict the outcomes on the first day of ventilation. Clinical phenotypes were derived from the eICU Collaborative Research Database (eICU) cohort via cluster analysis and were validated in the Medical Information Mart for Intensive Care (MIMIC-IV) cohort. Four clinical phenotypes were identified and compared in the eICU cohort (n = 15,256). Phenotype A (n = 3112) was associated with respiratory disease, had the lowest 28-day mortality (16%), and had a high extubation success rate (~80%). Phenotype B (n = 3335) was correlated with cardiovascular disease, had the second-highest 28-day mortality (28%), and had the lowest extubation success rate (69%). Phenotype C (n = 3868) was correlated with renal dysfunction, had the highest 28-day mortality (28%), and had the second-lowest extubation success rate (74%). Phenotype D (n = 4941) was associated with neurological and traumatic diseases, had the second-lowest 28-day mortality (22%), and had the highest extubation success rate (>80%). These findings were validated in the validation cohort (n = 10,813). Additionally, these phenotypes responded differently to ventilation strategies in terms of duration of treatment, but had no difference in mortality. The four clinical phenotypes unveiled the heterogeneity of ICU patients and helped to predict the 28-day mortality and the extubation success rate.

Elevated alanine transaminase is nonlinearly associated with in-hospital death in ICU-admitted diabetic ketoacidosis patients

AimsTo investigate the association between alanine transaminase (ALT) and in-hospital death in patients admitted to the intensive care unit for diabetic ketoacidosis (DKA). MethodsA cohort of 2,684 patients was constructed from the eICU Collaborative Research Database. Baseline demographic and clinical characteristics were summarized. Cox regressions with restricted cubic spline functions were modelled to explore the association between alanine transaminase and in-hospital death. Subgroup analyses were conducted between sexes, age groups, and people with/without obesity. ResultsAfter adjusting multiple confounders, a nonlinear, S-shaped association between ALT and in-hospital death was found. Compared to patients at median ALT, patients at the 90th percentile of ALT have a 1.88 (95 % confidence interval [CI]: 1.34–2.62) times higher hazard of in-hospital death in the unstratified cohort. Similar results were found in males (hazard ratio [HR] = 1.69, 95 % CI: 1.24–2.30); patients aged under 65 years (HR = 1.65, 95 % CI: 1.09–2.49); patients aged 65 years or above (HR = 3.45, 95 % CI: 1.67–7.14); non-obese patients (HR = 1.52, 95 % CI: 1.00–2.32); and obese patients (HR = 2.76, 95 % CI: 1.38–5.54). ConclusionsElevated ALT is robustly associated with in-hospital death in ICU-admitted DKA patients across several subgroups. Close monitoring of ALT in these patients is recommended.

Association of early postoperative serum magnesium with acute kidney injury after cardiac surgery

Introduction Dysmagnesemia has been demonstrated to be involved in the pathophysiology of kidney diseases and is common in cardiac surgical patients. It remains unknown whether changes of serum magnesium after cardiac surgery affect AKI. We aimed to investigate the association of early postoperative magnesium with cardiac surgery-associated AKI in adults. Methods We conducted a multicenter retrospective cohort study involving patients who underwent cardiac surgery in the eICU Collaborative Research Database between 2014 and 2015. AKI within 7 days after surgery was defined using both serum creatinine and urine output criteria of Kidney Disease Improving Global Outcomes definition. Postoperative AKI was analyzed using multivariable logistic regression with early postoperative serum magnesium measured within the first 24 h after surgery as a continuous variable and categorically by quartiles. Results Postoperative AKI was identified in 3498 of 6124 (57.1%) patients receiving cardiac surgery. The median (25th–75th percentiles) early postoperative serum magnesium level of the study population was 2.3 (2.0–2.7) mg/dL. Higher serum magnesium level was associated with a higher risk of developing postoperative AKI (adjusted odds ratio (OR), 1.46 per 1 mg/dL increase; 95% confidence interval (CI), 1.31–1.62; p<.001). The multivariable-adjusted ORs (95% CIs) of postoperative AKI across increasing quartiles of serum magnesium were 1.00 (referent), 1.11 (0.95–1.29), 1.30 (1.12–1.52), and 1.72 (1.47–2.02) (p for trend <.001). Conclusions These data demonstrate a significantly higher incidence of AKI in patients with higher early postoperative serum magnesium who underwent cardiac surgery.

Machine learning for prediction of in-hospital mortality in lung cancer patients admitted to intensive care unit.

The in-hospital mortality in lung cancer patients admitted to intensive care unit (ICU) is extremely high. This study intended to adopt machine learning algorithm models to predict in-hospital mortality of critically ill lung cancer for providing relative information in clinical decision-making. Data were extracted from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) for a training cohort and data extracted from the Medical Information Mart for eICU Collaborative Research Database (eICU-CRD) database for a validation cohort. Logistic regression, random forest, decision tree, light gradient boosting machine (LightGBM), eXtreme gradient boosting (XGBoost), and an ensemble (random forest+LightGBM+XGBoost) model were used for prediction of in-hospital mortality and important feature extraction. The AUC (area under receiver operating curve), accuracy, F1 score and recall were used to evaluate the predictive performance of each model. Shapley Additive exPlanations (SHAP) values were calculated to evaluate feature importance of each feature. Overall, there were 653 (24.8%) in-hospital mortality in the training cohort, and 523 (21.7%) in-hospital mortality in the validation cohort. Among the six machine learning models, the ensemble model achieved the best performance. The top 5 most influential features were the sequential organ failure assessment (SOFA) score, albumin, the oxford acute severity of illness score (OASIS) score, anion gap and bilirubin in random forest and XGBoost model. The SHAP summary plot was used to illustrate the positive or negative effects of the top 15 features attributed to the XGBoost model. The ensemble model performed best and might be applied to forecast in-hospital mortality of critically ill lung cancer patients, and the SOFA score was the most important feature in all models. These results might offer valuable and significant reference for ICU clinicians' decision-making in advance.

Malignant cancer may increase the risk of all-cause in-hospital mortality in patients with acute myocardial infarction: a multicenter retrospective study of two large public databases

BackgroundAcute myocardial infarction (AMI) and cancer are diseases with high morbidity and mortality worldwide, bringing a serious economic burden, and they share some risk factors. The purpose of this study was to determine the effect of cancer on the all-cause in-hospital mortality of patients with AMI.MethodsThis multicenter retrospective study analyzed patients with AMI from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database and eICU Collaborative Research Database (eICU-CRD) in the United States. Patients were divided into two groups based on whether they had concomitant malignant cancer: cancer and noncancer groups. The outcome was all-cause in-hospital mortality. The association between the two groups and their outcomes were analyzed using Kaplan–Meier and Cox proportional-hazards regression models. Propensity score matching (PSM) and propensity score based inverse probability of treatment weighting (IPTW) were used to further adjust for confounding variables to verify the stability of the results.ResultsThe study included 3,034 and 5,968 patients with AMI from the MIMIC-IV database and the eICU-CRD, respectively. Kaplan–Meier survival curves indicated that the probability of in-hospital survival was lower in patients with cancer than in those without cancer. After adjusting for potential confounding variables using multivariable Cox proportional hazards regression, the risk of all-cause in-hospital mortality was significantly higher in the cancer than the noncancer group, and the HR (95% CI) values for the cancer group were 1.56(1.22,1.98) and 1.35(1.01,1.79) in the MIMIC-IV database and the eICU-CRD, respectively. The same results were obtained after using PSM and IPTW, which further verified the results.ConclusionsAmong the patients with AMI, the all-cause in-hospital mortality risk of those with cancer was higher than those without cancer. Therefore, when treating such patients, comprehensive considerations should be made from a multidisciplinary perspective involving cardiology and oncology, with the treatment plan adjusted accordingly.

Association between central venous pressure measurement and outcomes in critically ill patients with severe coma

BackgroundThe use of central venous pressure (CVP) measurements among (intensive care unit) ICU patients with severe coma has been questioned. This study aimed to investigate the application value of CVP in this population.MethodsData stored in the ICU Collaborative Research Database (eICU-CRD) and Medical Information Mart for Intensive Care III (MIMIC-III) database were reviewed. Critically ill patients with a Glasgow Coma Scale (GCS) score of 3–8 were included. The primary outcome was the in-hospital mortality rate. The statistical approaches used included multivariable Cox regression, propensity score matching (PSM), inverse probability treatment weighting (IPTW), stabilized IPTW, and restricted cubic splines (RCS) to ensure the robustness of our findings.ResultsIn total, 7386 patients were included in the study. Early CVP measurement was independently associated with in-hospital mortality [hazard ratio, 0.63; p < 0.001] in patients with severe-to-moderate coma. This result was robust in the PSM, sIPTW, and IPTW cohorts. For all patients with CVP measurements, the RCS curves showed that the risk of in-hospital mortality increased as the initial CVP time was delayed. In addition, early CVP measurement was significantly associated with lower ICU mortality, 28-day mortality, and 365-day mortality and a significantly higher number of ventilator-free days.ConclusionEarly CVP measurement could improve clinical outcomes in critically ill patients with severe coma

The prediction of in-hospital mortality in chronic kidney disease patients with coronary artery disease using machine learning models

ObjectiveChronic kidney disease (CKD) patients with coronary artery disease (CAD) in the intensive care unit (ICU) have higher in-hospital mortality and poorer prognosis than patients with either single condition. The objective of this study is to develop a novel model that can predict the in-hospital mortality of that kind of patient in the ICU using machine learning methods.MethodsData of CKD patients with CAD were extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Boruta algorithm was conducted for the feature selection process. Eight machine learning algorithms, such as logistic regression (LR), random forest (RF), Decision Tree, K-nearest neighbors (KNN), Gradient Boosting Decision Tree Machine (GBDT), Support Vector Machine (SVM), Neural Network (NN), and Extreme Gradient Boosting (XGBoost), were conducted to construct the predictive model for in-hospital mortality and performance was evaluated by average precision (AP) and area under the receiver operating characteristic curve (AUC). Shapley Additive Explanations (SHAP) algorithm was applied to explain the model visually. Moreover, data from the Telehealth Intensive Care Unit Collaborative Research Database (eICU-CRD) were acquired as an external validation set.Results3590 and 1657 CKD patients with CAD were acquired from MIMIC-IV and eICU-CRD databases, respectively. A total of 78 variables were selected for the machine learning model development process. Comparatively, GBDT had the highest predictive performance according to the results of AUC (0.946) and AP (0.778). The SHAP method reveals the top 20 factors based on the importance ranking. In addition, GBDT had good predictive value and a certain degree of clinical value in the external validation according to the AUC (0.865), AP (0.672), decision curve analysis, and calibration curve.ConclusionMachine learning algorithms, especially GBDT, can be reliable tools for accurately predicting the in-hospital mortality risk for CKD patients with CAD in the ICU. This contributed to providing optimal resource allocation and reducing in-hospital mortality by tailoring precise management and implementation of early interventions.

Explainable machine learning model for predicting furosemide responsiveness in patients with oliguric acute kidney injury

Background Although current guidelines didn’t support the routine use of furosemide in oliguric acute kidney injury (AKI) management, some patients may benefit from furosemide administration at an early stage. We aimed to develop an explainable machine learning (ML) model to differentiate between furosemide-responsive (FR) and furosemide-unresponsive (FU) oliguric AKI. Methods From Medical Information Mart for Intensive Care-IV (MIMIC-IV) and eICU Collaborative Research Database (eICU-CRD), oliguric AKI patients with urine output (UO) < 0.5 ml/kg/h for the first 6 h after ICU admission and furosemide infusion ≥ 40 mg in the following 6 h were retrospectively selected. The MIMIC-IV cohort was used in training a XGBoost model to predict UO > 0.65 ml/kg/h during 6–24 h succeeding the initial 6 h for assessing oliguria, and it was validated in the eICU-CRD cohort. We compared the predictive performance of the XGBoost model with the traditional logistic regression and other ML models. Results 6897 patients were included in the MIMIC-IV training cohort, with 2235 patients in the eICU-CRD validation cohort. The XGBoost model showed an AUC of 0.97 (95% CI: 0.96–0.98) for differentiating FR and FU oliguric AKI. It outperformed the logistic regression and other ML models in correctly predicting furosemide diuretic response, achieved 92.43% sensitivity (95% CI: 90.88–93.73%) and 95.12% specificity (95% CI: 93.51–96.3%). Conclusion A boosted ensemble algorithm can be used to accurately differentiate between patients who would and would not respond to furosemide in oliguric AKI. By making the model explainable, clinicians would be able to better understand the reasoning behind the prediction outcome and make individualized treatment.

Twenty-eight-day in-hospital mortality prediction for elderly patients with ischemic stroke in the intensive care unit: Interpretable machine learning models.

Risk stratification of elderly patients with ischemic stroke (IS) who are admitted to the intensive care unit (ICU) remains a challenging task. This study aims to establish and validate predictive models that are based on novel machine learning (ML) algorithms for 28-day in-hospital mortality in elderly patients with IS who were admitted to the ICU. Data of elderly patients with IS were extracted from the electronic intensive care unit (eICU) Collaborative Research Database (eICU-CRD) records of those elderly patients admitted between 2014 and 2015. All selected participants were randomly divided into two sets: a training set and a validation set in the ratio of 8:2. ML algorithms, such as Naïve Bayes (NB), eXtreme Gradient Boosting (xgboost), and logistic regression (LR), were applied for model construction utilizing 10-fold cross-validation. The performance of models was measured by the area under the receiver operating characteristic curve (AUC) analysis and accuracy. The present study uses interpretable ML methods to provide insight into the model's prediction and outcome using the SHapley Additive exPlanations (SHAP) method. As regards the population demographics and clinical characteristics, the analysis in the present study included 1,236 elderly patients with IS in the ICU, of whom 164 (13.3%) died during hospitalization. As regards feature selection, a total of eight features were selected for model construction. In the training set, both the xgboost and NB models showed specificity values of 0.989 and 0.767, respectively. In the internal validation set, the xgboost model identified patients who died with an AUC value of 0.733 better than the LR model which identified patients who died with an AUC value of 0.627 or the NB model 0.672. The xgboost model shows the best predictive performance that predicts mortality in elderly patients with IS in the ICU. By making the ML model explainable, physicians would be able to understand better the reasoning behind the outcome.

Association of hypocalcemia with in-hospital mortality in critically ill patients with intracerebral hemorrhage: A retrospective cohort study.

There was little evidence to study the relationship between hypocalcemia and mortality among critically ill patients with intracerebral hemorrhage (ICH) aged ≥16 years. This study aimed to determine the potential association between hypocalcemia and in-hospital and ICU mortality in patients with ICH in the United States. We analyzed 1,954 patients with ICH from the e-Intensive Care Unit Collaborative Research Database and divided them into hypocalcemia and non-hypocalcemia groups. Hypocalcemia was defined as albumin-adjusted total calcium below 8.4 mg/dl. The primary and secondary outcomes were hospital and ICU mortality, respectively. We performed multivariable regression and subgroup analyses to evaluate the association of hypocalcemia with hospital and ICU mortality. Cumulative survival rate analysis was performed using Kaplan-Meier curves with log-rank statistics. We enrolled 1,954 patients with ICH who had been hospitalized in ICU for >24 h and were older than 16 years (average age, 61.8 years; men, 56.7%). We noted that 373 (19%) hospital mortality occurred, including 235 (12%) ICU mortality. In this sample, 195 patients had hypocalcemia. Multivariable logistic regression analyses showed that hypocalcemia was associated with a 67% increased risk of in-hospital and a 72% increased risk of ICU mortality. This association was consistent across subgroup analyses. Hypocalcemia was associated with a high risk of hospital and ICU mortality among critically ill patients with ICH. Future prospective, randomized, controlled studies are needed to confirm our results.

Circadian rhythms of vital signs are associated with in-hospital mortality in critically ill patients: A retrospective observational study

ABSTRACT Vital signs have been widely used to assess the disease severity of patients, but there is still a lack of research on their circadian rhythms. The objective is to explore the circadian rhythms of vital signs in critically ill patients and establish an in-hospital mortality prediction model. Study patients from the recorded eICU Collaborative Research Database were included in the present analyses. The circadian rhythms of vital signs are analyzed in critically ill patients using the cosinor method. Logistic regression was used to screen independent predictors and establish a prediction model for in-hospital mortality by multivariate logistic regression analysis and to show in the nomogram. Internal validation is used to evaluate the prediction model by bootstrapping with 1000 resamples. A total of 29,448 patients were included in the current analyses. The Mesor, Amplitude, and Peak time of vital signs, such as heart rate (HR), temperature, respiration rate (RR), pulse oximetry-derived oxygen saturation (SpO2), and blood pressure (BP), were significant differences between survivors and non-survivors . Logistic regression analysis showed that Mesor, Amplitude, and Peak time of HR, RR, and SpO2 were independent predictors for in-hospital mortality in critically ill patients. The area under the curve (AUC) and c-index of the prediction model for the Medical intensive care unit (MICU) and Surgical intensive care unit (SICU) were 0.807 and 0.801, respectively. The Hosmer-Lemeshow test P-values were 0.076 and 0.085, respectively, demonstrating a good fit for the prediction model. The calibration curve and decision curve analysis (DCA) also demonstrated its accuracy and applicability. Internal validation assesses the consistency of the results. There were significant differences in the circadian rhythms of vital signs between survivors and non-survivors in critically ill patients. The prediction model established by the Mesor, Amplitude, and Peak time of HR, RR, and SpO2 combined with the Acute Physiology and Chronic Health Evaluation (APACHE) IV score has good predictive performance for in-hospital mortality and may eventually support clinical decision-making. Abbreviations: ICU: Intensive care unit; MICU: Medical intensive care unit; SICU: Surgical intensive care unit; HR: Heart rate; RR: Respiration rate; SpO2: Pulse oximetry-derived oxygen saturation; BP: Blood pressure; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; APACHE: Acute Physiology and Chronic Health Evaluation; bpm: beats per min; BMI: Body mass index; OR: Odd ratio; CI: Confidential interval; IQR: Interquartile range; SD: Standard deviation; ROC: Receiver operating characteristic; AUC: area under the curve; DCA: Decision curve analysis; IRB: Institutional review board.

Early Postoperative Acetaminophen Administration and Severe Acute Kidney Injury After Cardiac Surgery

Oxidative stress may contribute to the development of acute kidney injury (AKI) after cardiac surgery. Acetaminophen can be considered an antioxidant because it inhibits hemoprotein-catalyzed lipid peroxidation. We hypothesized that perioperative acetaminophen administration is associated with reduced AKI after cardiac surgery. Retrospective observational cohort study. Patients aged≥18 years who had cardiac surgery were identified from 2 publicly available clinical registries: the Medical Information Mart for Intensive Care III (MIMIC-III) and the eICU Collaborative Research Database (eICU). Administration of acetaminophen in the first 48 hours after surgery. Severe AKI in the first 7 days after surgery, defined as stage 2 or stage 3 AKI according to KDIGO criteria. Multivariable cause-specific hazards regression analysis. We identified 5,791 patients from the MIMIC-III and 3,840 patients from the eICU registries. The overall incidence of severe AKI was 58% (3,390 patients) in the MIMIC-III cohort and 37% (1,431 patients) in the eICU cohort. Acetaminophen was administered in the early postoperative period to 4,185 patients (72%) and 2,737 patients (71%) in these 2 cohorts, respectively. In multivariable regression models, early postoperative use of acetaminophen was associated with a lower risk of severe AKI in both the MIMIC-III (adjusted hazard ratio [AHR], 0.86 [95% CI, 0.79-0.94]) and eICU (AHR, 0.84 [95% CI, 0.72-0.97]) cohorts. The benefit was consistent across sensitivity and subgroup analyses. No data on acetaminophen dose. Early postoperative acetaminophen administration was independently associated with a lower risk of severe AKI in adults recovering from cardiac surgery. Prospective trials are warranted to assess the extent to which the observed association is causal and estimate the extent to which acetaminophen administration might prevent or reduce the severity of AKI.

Impact of Early Tracheostomy on Clinical Outcomes in Trauma Patients Admitted to the Intensive Care Unit: A Retrospective Causal Analysis

To assess the indications, timing, and clinical outcomes that result from the early tracheostomy (ET) administration, by causal inference models. A retrospective observational study. Multiinstitutional intensive care unit in the United States PARTICIPANTS: The study comprised 626 trauma patients. An ET versus late tracheostomy (LT). Trauma patients with tracheostomy were identified from 2 public databases named Medical Information Mart for the Intensive Care-IV and eICU Collaborative Research Database. Tracheostomy was defined as early (≤7 days) or late (>7 days) from intensive care unit admission. A marginal structural Cox model (MSCM) with inverse probability weighting was employed. For comparison, the authors also used time-dependent propensity-score matching (PSM) to account for differences in the probability of receiving an ET or LT. A total of 626 eligible patients were enrolled in the study, of whom 321 (51%) received a ET. The MSCM and time-dependent PSM indicated that the ET group was associated with reduced ventilation-associated pneumonia (VAP) and a shorter mechanical ventilation (MV) duration than the LT group. Yet, mortality did not show any difference between the two groups. The authors' study observed that ET was not associated with reduced mortality in trauma patients, but it was associated with reduced VAP risk and MV duration. The results warrant further validation in randomized controlled trials.

1194: THE MORTALITY CAUSAL EFFECT OF POSITIVE FLUID BALANCES SEPSIS SUBTYPES

Introduction: Millions of deaths occur each year due to sepsis and many questions regarding best practices remain. Volume of fluid resuscitation is one of these questions. It is thought that certain sepsis subtypes may actually fare worse with positive fluid balances, however, the current recommendation for an initial resuscitation of 30cc/kg of crystalloid fluids remains. Randomized controlled trials are the gold standard for determining causal effect, but are time and cost-intensive. Causal inference using inverse probability of treatment weighting (IPTW) provides a cost-effective alternative. Through causal inference, we will evaluate the mortality effect of positive fluid balances on sepsis subtypes with the hypothesis that pulmonary sepsis would have a worse outcome. Methods: An anonymous, publicly available database was analyzed using preexisting IRB approval. The eICU Collaborative Research Database contains data on over 200000 intensive care unit admissions. The analysis used 138986 patients with 30 initial features. We dropped features with more than 50% missing values. Imputation with Sklearn Iterative Imputator. Data scaled with Sklearn Standard Scaler. Selecting features sequentially with 6 final features was performed using MLxtend. The primary outcome was ICU mortality. Aim was to establish Average Treatment Effect for positive fluid balances in pulmonary sepsis compared with urosepsis. Pulmonary sepsis had 6754 patients compared to 4152 in urosepsis. Using domain knowledge, we built a Directed Acyclic Graph (DAG). We selected Apache IV as the confounder covariate based on our DAG. As effect modifiers, we selected creatinine and age. In both groups, IPTW was performed using IBM DoWhy. Results: Mortality analysis of the pulmonary sepsis group demonstrated an ATE of 0.13 (95% CI [0.08,0.19]). The urosepsis group showed an ATE of 0.06 (95% CI [0.054 -0.067]). Positive fluid balances had an increased causal effect on mortality for the pulmonary sepsis group. Conclusions: Positive fluid balance may be more harmful in certain sepsis subtypes. Pulmonary sepsis has a higher fluid balance-related effect on mortality than urosepsis, as determined by causal inference techniques. We hope that this analysis will lead to a further evaluation of fluids’ varying effects on sepsis subtypes.

1258: USE OF CAUSAL INFERENCE TO COMPARE NOREPINEPHRINE- AND VASOPRESSIN-RELATED ICU MORTALITY

Introduction: The choice of vasopressor has been studied extensively in septic shock, however limited data exists in other shock types. Norepinephrine and vasopressin were compared in a retrospective study of the cardiac surgery population. Vasopressin as first-line treatment of post-cardiac surgery vasoplegic shock was found to improve clinical outcomes. However, the study was associated with biases which limits its validity. As there are no prospective studies, we use causal inference to analyze retrospective data. An observational study allows individual observation under one treatment scenario, while a counterfactual outcome is unknown. Targeted Maximum Likelihood Estimation (TMLE) models counterfactual outcomes with limited confounding by generating a pseudo-population. By adjusting for common causes of death, we were able to compare mortality outcomes between norepinephrine and vasopressin. Methods: An anonymous, publicly available database was analyzed using pre-existing IRB approval. The eICU Collaborative Research Database contains data on over 200,000 ICU admissions. The analysis used 27,388 patients with 25 initial features. Missing values greater than 50% were dropped. Machine Learning Extensions Library (MLxtend) was used to select features sequentially. The primary outcome was ICU mortality. Imputation was performed using Sklearn Iterative Imputuer, data scaled using Sklearn Standard Scaler. Our primary model was TMLE with Logistic Regression. We analyzed vasopressin and norepinephrine administration, accounting for their co-administration by using them as covariates. Net fluid balance, Apache IV, albumin, admission weight, age, diabetes, and metastatic cancer were confounder covariates for both groups. Before analysis, a causal diagram was constructed based on these covariates. Our primary metric was Average Treatment Effect (ATE). Results: Mortality analysis on the norepinephrine group showed an ATE of -0.179, 95% CI [-0.188, -0.167]. Vasopressin showed an ATE of 0.03 (95% CI [0.01,0.051]). Compared to vasopressin, norepinephrine showed a reduction in mortality. Conclusions: We found that norepinephrine was associated with decreased mortality compared to vasopressin in this setting. We hope this study’s results will help guide prospective studies moving forward.

A novel risk model based on white blood cell-related biomarkers for acute kidney injury prediction in patients with ischemic stroke admitted to the intensive care unit.

Conventional systemic inflammatory biomarkers could predict prognosis in patients with ischemic stroke (IS) admitted to the intensive care unit (ICU). Acute kidney injury (AKI) is common in patients with IS admitted to ICU, but few studies have used systemic inflammatory biomarkers to predict AKI in critically ill patients with IS. This study aimed to establish a risk model based on white blood cell (WBC)-related biomarkers to predict AKI in critically ill patients with IS. Data were extracted from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) for a training cohort, and data were extracted from the Medical Information Mart for eICU Collaborative Research Database (eICU-CRD) for a validation cohort. Logistic regression analysis was used to determine the significant predictors of WBC-related biomarkers on AKI prediction, and a risk model was established based on those significant indicators in multivariate logistic regression. The receiver operating characteristics (ROC) curve was utilized to obtain the best cut-off value of the risk model. The Kaplan-Meier curve was used to evaluate the prognosis-predictive ability of the risk model. The overall incidence of AKI was 28.4% in the training cohort and 33.2% in the validation cohort. WBC to lymphocyte ratio (WLR), WBC to basophils ratio (WBR), WBC to hemoglobin ratio (WHR), and neutrophil to lymphocyte ratio (NLR) could independently predict AKI, and a novel risk model was established based on WLR, WBR, WHR, and NLR. This risk model depicted good prediction performance both in AKI and other clinical outcomes including hemorrhage, persistent AKI, AKI progression, ICU mortality, and in-hospital mortality both in the training set and in the validation set. A risk model based on WBC-related indicators exhibited good AKI prediction performance in critically ill patients with IS which could provide a risk stratification tool for clinicians in the ICU.

Association between the blood urea nitrogen to creatinine ratio and in‑hospital mortality among patients with acute myocardial infarction: A retrospective cohort study.

The present study aimed to determine the association between the blood urea nitrogen (BUN) and creatinine (Cr) ratio and in-hospital mortality in patients with acute myocardial infarction (AMI). The present retrospective cohort study included adult patients (≥18 years of age) who were admitted to the intensive care unit (ICU) with a primary diagnosis of AMI. Medical records were obtained from the electronic ICU collaborative research database, which includes data from throughout continental USA. Data included demographic characteristics, vital signs, laboratory tests and comorbidities. The clinical endpoint was in-hospital mortality. The Cox proportional hazards model was used to evaluate the prognostic values of the basic BUN/Cr ratio and the Kaplan-Meier method was used to plot survival curves. Subgroup analyses were performed to measure mortality across various subgroups. In total, 5,965 eligible patients were included. In the Cox regression analysis, after being adjusted for age, sex, ethnicity and other confounding factors, the BUN/Cr ratio was found to be a significant risk predictor of in-hospital mortality. There was a non-linear relationship between the BUN/Cr ratio and in-hospital mortality after adjusting for potential confounders. A two-piecewise regression model was used to obtain a threshold inflection point value of 18. Furthermore, after adjusting for additional confounding factors (age, sex, ethnicity, BMI, heart rate, oxygen saturation, platelets, total protein, AMI category, heart failure, history of diabetes, history of hypertension, percutaneous coronary intervention, and administration of norepinephrine, dopamine and epinephrine), the BUN/Cr ratio remained a significant predictor of in-hospital mortality (third vs. first tertile: Hazard ratio, 1.50; 95% CI, 1.08-2.09; P<0.05). The Kaplan-Meier curve for tertiles of the BUN/Cr ratio indicated that in-hospital mortality rates were highest when the BUN/Cr ratio was ≥18.34 after adjustment for age, sex and ethnicity (P<0.05). The present findings demonstrated that a higher BUN/Cr ratio was associated with an increased risk of in-hospital mortality in patients with non-ST-segment elevation myocardial infarction. These results support a revision of how the prognosis of patients with AMI is predicted.

Machine-learning-derived sepsis bundle of care.

Compliance to the Surviving Sepsis Campaign (SSC) guidelines is limited. This is known to be associated with increased mortality. The aim of this retrospective cohort study was to identify among the SCC guidelines the optimal bundle of recommendations that minimize 28-day mortality. We used a training cohort to identify, using a least absolute shrinkage and selection operator penalized machine learning model, this bundle. Patients with sepsis/septic shock admitted to the intensive care unit (ICU) were extracted from two US databases, the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database (training and internal validation cohorts) and the eICU Collaborative Research Database (eICU-CRD) (external validation cohort). In the validation cohorts, we defined a bundle group that includes patients who were treated with at least all the recommendations selected in our bundle and a no-bundle group that includes patients in whom at least one recommendation from our bundle was omitted. All-cause 28-day mortality was the primary outcome measure. A total of 42,735 patients were included. Six recommendations (antimicrobials, balanced crystalloid, insulin therapy, corticosteroids, vasopressin, and bicarbonate therapy) were identified from the training cohort to be included in our bundle. In the propensity score-(PS)-matched internal validation cohort, the bundle group was associated with a lower mortality (OR 0.41 [0.33-0.53]; p < 0.001) compared to the no-bundle group. This was confirmed in the PS-matched external validation cohort (OR 0.75 [0.60-0.94]; p 0.02). Our bundle of six recommendations is associated with a dramatic reduction in mortality in sepsis and septic shock. This bundle needs to be evaluated prospectively.

Association between serum calcium level and in-hospital mortality in patients with acute myocardial infarction: a retrospective cohort study

The association between serum calcium levels and the prognosis of patients with acute myocardial infarction (AMI) remains controversial. This study aimed to explore the association between serum calcium and in-hospital mortality in patients with AMI. The data of this study were extracted from the Philips eICU Collaborative Research Database. A total of 7284 patients were eventually enrolled in this study, of which 799 (10.97%) died during hospitalization. For each patient, serum calcium, corrected to albumin, was calculated and categorized into four groups: Q1 ≤ 8.5, Q2 8.5–9.5, Q3 9.5–10.5, and Q4 > 10.5 mg/dL. Multivariate analysis demonstrated that corrected sCa was an independent predictor of in-hospital death (Q2 vs. Q1, OR 0.5, 95% CI 0.4–0.7, P < 0.001; Q3 vs. Q1, OR 0.8, 95% CI 0.6–1.0, P = 0.035; Q4 vs. Q1, OR 1.6, 95% CI 1.1–2.3, P = 0.008). The association remained stable in the fully adjusted model. A significant U-shaped association between corrected serum calcium and in-hospital mortality was observed in piecewise linear regression model (Corrected sCa < 9.4 mg/dL, OR 0.8, 95% CI 0.7–0.9, P < 0.001; corrected sCa > 9.4 mg/dL, OR 1.5, 95% CI 1.3–1.8, P < 0.001). In conclusion, both decreased and increased corrected serum calcium is associated with increased in-hospital mortality in patients with AMI, and patients may have the lowest risk of in-hospital death when corrected serum calcium is 9.4 mg/dL (2.35 mmol/L).

Machine learning models to predict in-hospital mortality in septic patients with diabetes.

Sepsis is a leading cause of morbidity and mortality in hospitalized patients. Up to now, there are no well-established longitudinal networks from molecular mechanisms to clinical phenotypes in sepsis. Adding to the problem, about one of the five patients presented with diabetes. For this subgroup, management is difficult, and prognosis is difficult to evaluate. From the three databases, a total of 7,001 patients were enrolled on the basis of sepsis-3 standard and diabetes diagnosis. Input variable selection is based on the result of correlation analysis in a handpicking way, and 53 variables were left. A total of 5,727 records were collected from Medical Information Mart for Intensive Care database and randomly split into a training set and an internal validation set at a ratio of 7:3. Then, logistic regression with lasso regularization, Bayes logistic regression, decision tree, random forest, and XGBoost were conducted to build the predictive model by using training set. Then, the models were tested by the internal validation set. The data from eICU Collaborative Research Database (n = 815) and dtChina critical care database (n = 459) were used to test the model performance as the external validation set. In the internal validation set, the accuracy values of logistic regression with lasso regularization, Bayes logistic regression, decision tree, random forest, and XGBoost were 0.878, 0.883, 0.865, 0.883, and 0.882, respectively. Likewise, in the external validation set 1, lasso regularization = 0.879, Bayes logistic regression = 0.877, decision tree = 0.865, random forest = 0.886, and XGBoost = 0.875. In the external validation set 2, lasso regularization= 0.715, Bayes logistic regression = 0.745, decision tree = 0.763, random forest = 0.760, and XGBoost = 0.699. The top three models for internal validation set were Bayes logistic regression, random forest, and XGBoost, whereas the top three models for external validation set 1 were random forest, logistic regression, and Bayes logistic regression. In addition, the top three models for the external validation set 2 were decision tree, random forest, and Bayes logistic regression. Random forest model performed well with the training and three validation sets. The most important features are age, albumin, and lactate.