MIMIC-III Database Research Articles

CEHMR: Curriculum learning enhanced hierarchical multi-label classification for medication recommendation

The medication recommendation (MR) or medication combination prediction task aims to predict effective prescriptions given accurate patient representations derived from electronic health records (EHRs), which contributes to improving the quality of clinical decision-making, especially for patients with multi-morbidity. Although in recent years deep learning technology has achieved great success in MR, the performance of current multi-label based MR solutions is unsatisfactory. They mainly focus on improving the patient representation module and modeling the medication label dependencies such as drug–drug interaction (DDI) correlation and co-occurrence relationship. However, the hierarchical dependency among medication labels and diversity of difficulty among MR training examples lack sufficient consideration. In this paper, we propose a framework of Curriculum learning Enhanced Hierarchical multi-label classification for MR (CEHMR). Motivated by the category hierarchy of medications which organizes standard medication codes in a hierarchical structure, we utilize it to provide more trustworthy prior knowledge for modeling label dependency. Specifically, we design a hierarchical multi-label classifier with a learnable gate fusion layer, to simultaneously capture the level-independent (local) and level-dependent (global) hierarchical information in the medication hierarchy. In addition, to overcome the diversity of training example difficulties, and progressively achieve a smoother training process, we introduce a bootstrap-based curriculum learning strategy. Hence, the example difficulty can be measured based on the predictive performance of the MR model, and then all training examples would be retrained from easy to hard under the guidance of a predefined training scheduler. Experiments on the real-world medical MIMIC-III database demonstrate that the proposed framework can achieve state-of-the-art performance compared with seven representative baselines, and extensive ablation studies validate the effectiveness of each component of CEHMR.

3-Hydroxybutyrate ameliorates sepsis-associated acute lung injury by promoting autophagy through the activation of GPR109α in macrophages

BackgroundSepsis is a systemic inflammatory disease caused by multiple pathogens, with the most commonly affected organ being the lung. 3-Hydroxybutyrate plays a protective role in inflammatory diseases through autophagy promotion; however, the exact mechanism remains unexplored. MethodOur study used the MIMIC-III database to construct a cohort of ICU sepsis patients and figure out the correlation between the level of ketone bodies and clinical prognosis in septic patients. In vivo and in vitro models of sepsis were used to reveal the role and mechanism of 3-hydroxybutyrate in sepsis-associated acute lung injury (sepsis-associated ALI). ResultHerein, we observed a strong correlation between the levels of ketone bodies and clinical prognosis in patients with sepsis identified using the MIMIC- III database. In addition, exogenous 3-hydroxybutyrate supplementation improved the survival rate of CLP-induced sepsis in mice by promoting autophagy. Furthermore, 3-hydroxybutyrate treatment protected against sepsis-induced lung damage. We explored the mechanism underlying these effects. The results indicated that 3-hydroxybutyrate upregulates autophagy levels by promoting the transfer of transcription factor EB (TFEB) to the macrophage nucleus in a G-protein-coupled receptor 109 alpha (GPR109α) dependent manner, upregulating the transcriptional level of ultraviolet radiation resistant associated gene (UVRAG) and increasing the formation of autophagic lysosomes. Conclusion3-Hydroxybutyrate can serve as a beneficial therapy for sepsis-associated ALI through the upregulation of autophagy. These results may provide a basis for the development of promising therapeutic strategies for sepsis-associated ALI.

Early prediction of MODS interventions in the intensive care unit using machine learning

BackgroundMultiple organ dysfunction syndrome (MODS) is one of the leading causes of death in critically ill patients. MODS is the result of a dysregulated inflammatory response that can be triggered by various causes. Owing to the lack of an effective treatment for patients with MODS, early identification and intervention are the most effective strategies. Therefore, we have developed a variety of early warning models whose prediction results can be interpreted by Kernel SHapley Additive exPlanations (Kernel-SHAP) and reversed by diverse counterfactual explanations (DiCE). So we can predict the probability of MODS 12 h in advance, quantify the risk factors, and automatically recommend relevant interventions.MethodsWe used various machine learning algorithms to complete the early risk assessment of MODS, and used a stacked ensemble to improve the prediction performance. The kernel-SHAP algorithm was used to quantify the positive and minus factors corresponding to the individual prediction results, and finally, the DiCE method was used to automatically recommend interventions. We completed the model training and testing based on the MIMIC-III and MIMIC-IV databases, in which the sample features in the model training included the patients’ vital signs, laboratory test results, test reports, and data related to the use of ventilators.ResultsThe customizable model called SuperLearner, which integrated multiple machine learning algorithms, had the highest authenticity of screening, and its Yordon index (YI), sensitivity, accuracy, and utility_score on the MIMIC-IV test set were 0.813, 0.884, 0.893, and 0.763, respectively, which were all maximum values of eleven models. The area under the curve of the deep–wide neural network (DWNN) model on the MIMIC-IV test set was 0.960, and the specificity was 0.935, which were both the maximum values of all these models. The Kernel-SHAP algorithm combined with SuperLearner was used to determine the minimum value of glasgow coma scale (GCS) in the current hour (OR = 0.609, 95% CI 0.606–0.612), maximum value of MODS score corresponding to GCS in the past 24 h (OR = 2.632, 95% CI 2.588–2.676), and maximum score of MODS corresponding to creatinine in the past 24 h (OR = 3.281, 95% CI 3.267–3.295) were generally the most influential factors.ConclusionThe MODS early warning model based on machine learning algorithms has considerable application value, and the prediction efficiency of SuperLearner is superior to those of SubSuperLearner, DWNN, and other eight common machine learning models. Considering that the attribution analysis of Kernel-SHAP is a static analysis of the prediction results, we introduce the DiCE algorithm to automatically recommend counterfactuals to reverse the prediction results, which will be an important step towards the practical application of automatic MODS early intervention.

Less risk factor at AMI onset, higher risk in prognosis---the paradoxical phenomenon from MIMIC-III database

Abstract Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National Natural Science Foundation of China (NSFC). Introduction Numerous epidemiologic studies have confirmed that Standard Modifiable Cardiovascular Risk Factors (SMuRFs) are independent risk factors for coronary heart disease (CHD) and are recommended by guidelines for the primary and secondary prevention of CHD. However, it is interesting to find that the increasing proportion of individuals without SMuRFs in CHD and acute myocardial infarction (AMI). Moreover, accumulating clinical trials have reported the inverse association between the number of risk factors and crude mortality in AMI. It is unknown whether this paradoxical phenomenon can be seen in critical ill patients. Purpose To analyze and validate whether the number of risk factors contradicts the mortality outcome in patients with AMI in the MIMIC-III (Medical Information Mart for Intensive Care III) database. Methods A total of 2484 AMI patients (62.8% male, mean age 71.2 [59.3;80.8] years) admitting to cardiac care unit for the first time from the MIMIC-III database were enrolled. Risk factors were defined as hypertension, diabetes,dyslipidemia, and smoking history. They were grouped according to the number of risk factors. Main outcome was mortality at 2-year following hospitalization. Kaplan-Meier curve and Cox proportional hazards models adjusting for baseline clinical characteristics and in-hospital treatment were constructed to evaluate the impact of numbers of risk factors on mortality in critical AMI patients. Results At 2-year follow-up, death was observed in 714 patients (28.7%). A total of 474(19.1%),1541(62%),469(18.9%) patients had 0,1-2,3-4 SMuRFs respectively. No risk group has a significantly higher 2-year mortality (Picture 1). Compared with no risk group, patients with 1-2 and 3-4 SMuRFs were correlated to an HR of 0.72(95%CI, 0.61-0.86) and 0.66(95%CI, 0.52-0.83) for all-cause mortality in the unadjusted model. Interestingly, after multivariate adjustment, less risk factor at AMI onset was still associated with poor prognosis. Conclusions Taking into confounders, critical AMI patients with less risk factors have higher risk in poor prognosis. Further studies on this paradoxical phenomenon are needed.

Red blood cell distribution width and all-cause mortality in congestive heart failure patients: a retrospective cohort study based on the Mimic-III database.

The red blood cell distribution width (RDW) is a metric that measures the variation in the size of red blood cells and is presented as the red blood cell volume coefficient of variation. Increased RDW levels are closely linked to an elevated risk of death from congestive heart failure (CHF) and might be a new risk marker for cardiovascular disease. This research sought to evaluate the possible link between RDW levels and all-cause mortality in CHF patients after controlling for other covariates. The publicly accessible Mimic-III database served as the source of data for our research. We employed ICU admission scoring systems to gather information on each patient's demographical data, laboratory test results, comorbid conditions, vital signs, and scores. Among CHF patients, the link between baseline RDW levels and short-, medium-, and long-term all-cause mortality was evaluated by Cox proportional hazard analysis, smooth curve fitting, and Kaplan-Meier survival curves. In total, 4,955 participants were selected for the study with an average age of 72.3 ± 13.5 years (old) and with males accounting for 53.1%. The findings recorded from the fully adjusted Cox proportional hazard model showed that higher RDW was associated with a greater risk of 30-day, 90-day, 365-day, and 4-year all-cause death; the HRs and 95% confidence intervals were 1.11 (1.05, 1.16), 1.09 (1.04, 1.13), 1.10 (1.06, 1.14), and 1.10 (1.06, 1.13), respectively. The results were stable and reliable using subgroup analysis. Smooth curve fitting and the K-M survival curve method further validated our results. The RDW levels had a u-shaped connection with 30-day mortality. The RDW level was linked to an elevated risk of short-, medium-, and long-term all-cause death among CHF patients.

Development and validation of a novel nomogram of 1-year mortality in the elderly with hip fracture: a study of the MIMIC-III database

ObjectiveHip fracture is a prevalent condition with a significant death rate among the elderly. We sought to develop a nomogram-based survival prediction model for older patients with hip fracture.DesignA retrospective...

Association Between Percutaneous Oxygen Saturation and Mortality of Patients with Mild Traumatic Brain Injury at ICU Admission: An Analysis of the MIMIC-III Database.

Research related to the result of patients with mild traumatic brain injury and the role of percutaneous oxygen saturation (SpO2) level is rarely reported. Our study investigated the relationship between SpO2 and the 30- and 90-day mortality among patients with mild TBI. A total of 1027 patients with mild TBI[Glasgow Coma Scale (GCS) > 12] were enrolled from the Medical Information Mart for Intensive Care (MIMIC-III) database. The patients were classified into low-SpO2 (< 95%), moderate-SpO2 (95-98%), and high-SpO2 groups (> 98%). With 30- and 90-day mortality rates as the main outcomes, Cox regression and confined cubic spline models were adopted. There was a U-shaped curve in confined cubic splines for the relationship between SpO2 and mortality. Compared with the moderate-SpO2 group, the high-SpO2 group exhibited a much higher risk of mortality after modification, with hazard proportions of 2.108 (95%CI 1.211-3.670, P < 0.05) for the 30-day mortality and 1.760 (95%CI 1.140-2.720, P < 0.05) for the 90-day mortality, and the low-SpO2 group exhibited a much higher risk of mortality after modification, with hazard proportions of 2.215 (95%CI 1.194-4.110, P < 0.05) for the 90-day mortality. Among patients with mild TBI, the correlation between SpO2 level and 30- and 90-day mortality followed a U-shaped curve. Both low and high SpO2 level exerted potential harmful effects on the outcomes of patients with mild TBI. The SpO2 range of 95-98% could be the optimal SpO2 level for mild patients with TBI.

Early human albumin administration is associated with reduced mortality in septic shock patients with acute respiratory distress syndrome: A retrospective study from the MIMIC-III database.

Background: Sepsis-induced acute respiratory distress syndrome (ARDS) was associated with higher mortality. It is unclear whether albumin supplementation early in the course of ARDS can affect the prognostic outcomes of septic shock (SS) patients with ARDS. Methods: The MIMIC-III database was employed to identify SS patients with ARDS. The effect of early application (<24h after ICU admission) of human albumin on 28-day mortality in SS patients with ARDS was explored. The propensity score matching was used to minimize the bias between the non-albumin and early albumin treatment groups. Results: The analysis for all eligible patients who received human albumin showed significantly lower 28-hospital mortality rates than the non-albumin group (37% versus 47%, p = 0.018). After propensity matching, the difference between the two groups also significantly (34.8% versus 48.1%, p = 0.031). Moreover, we found that the relationship between albumin use and reduced 28-day mortality was inconsistent across SOFA score subgroups (Pinteraction = 0.004, non-adjustment for multiple testing). Conclusion: Early human albumin administration in SS patients with ARDS was independently associated with a reduction of 28-day mortality. Furthermore, the benefit of human albumin treatment appeared to be more pronounced in patients with a SOFA score of ≤ 10.

Association between different MAP levels and 30-day mortality in sepsis patients: a propensity-score-matched, retrospective cohort study

BackgroundSepsis is a life-threatening organ dysfunction caused by the infection-related host response disorder. Adequate mean arterial pressure is an important prerequisite of tissue and organ perfusion, which runs through the treatment of sepsis patients, and an appropriate mean arterial pressure titration in the early-stage correlates to the positive outcome of the treatment. Therefore, in the present study, we aimed to elucidate the relationship between early mean arterial pressure levels and short-term mortality in sepsis patients.MethodsWe included all suspected sepsis patients from MIMIC-III database with average mean arterial pressure ≥ 60 mmHg on the first day of intensive care unit stay. Those patients were then divided into a permissive low-mean arterial pressure group (60–65 mmHg) and a high-mean arterial pressure group (> 65 mmHg). Multivariate Cox regression analysis was conducted to analyze the relationship between MAP level and 30-day, 60-day, and 100-day mortality of suspected sepsis patients in the two groups. Propensity score matching, inverse probability of treatment weighing, standardized mortality ratio weighting, PA weighting, overlap weighting, and doubly robust analysis were used to verify our results.ResultsA total of 14,031 suspected sepsis patients were eligible for inclusion in our study, among which 1305 (9.3%) had an average first-day mean arterial pressure of 60–65 mmHg, and the remaining 12,726 patients had an average first-day mean arterial pressure of more than 65 mmHg. The risk of 30-day mortality was reduced in the high mean arterial pressure group compared with the permissive low-mean arterial pressure group (HR 0.67 (95% CI 0.60–0.75; p < 0.001)). The higher mean arterial pressure was also associated with lower 60-day and 100-day in-hospital mortality as well as with shorter duration of intensive care unit stay. Patients in the high-mean arterial pressure group also had more urine output on the first and second days of intensive care unit admission.ConclusionsAfter risk adjustment, the initial mean arterial pressure of above 65 mmHg was associated with reduced short-term mortality, shorter intensive care unit stay, and higher urine volume in the first two days among patients with sepsis.

Significance of platelets in the early warning of new-onset AKI in the ICU by using supervise learning: a retrospective analysis

Objective To explore a machine learning model for the early prediction of acute kidney injury (AKI) and to screen the related factors affecting new-onset AKI in the ICU. Methods A retrospective analysis was performed used the MIMIC-III data source. New onset of AKI defined based on the serum creatinine changed. We included 19 variables for AKI assessment using four machine learning models: support vector machines, logistic regression, and random forest. and XGBoost, using accuracy, specificity, precision, recall, F1 score, and AUROC (area under the ROC curve) to evaluate model performance. The four models predicted new-onset AKI 3–6–9–12 h ahead. The SHapley Additive exPlanation (SHAP) value is used to evaluate the feature importance of the model. Results We finally respectively extracted 1130 AKI patients and non-AKI patients from the MIMIC-III database. With the extension of the early warning time, the prediction performance of each model showed a downward trend, but the relative performance was consistent. The prediction performance comparison of the four models showed that the XGBoost model performed the best in all evaluation indicators in all the time point at new-onset AKI 3–6–9–12 h ahead (accuracy 0.809 vs 0.78 vs 0.744 vs 0.741, specificity 0.856 vs 0.826 vs 0.797 vs 0.787, precision 0.842 vs 0.81 vs 0.775 vs 0.766, recall 0.759 vs 0.734 vs 0.692 vs 0.694, Fl score 0.799 vs 0.769 vs 0.731 vs 0.729, AUROC 0.892 vs 0.857 vs 0.827 vs 0.818). In the prediction of AKI 6, 9 and 12 h ahead, the importance of creatinine, platelets, and height was the most important based on SHapley. Conclusions The machine learning model described in this study can predict AKI 3–6–9–12 h before the new-onset of AKI in ICU. In particular, platelet plays an important role. Key message The new-onset of AKI in ICU is a common and important problem, which early be identified the risk of AKI can improve patients’ outcomes. We explored MIMIC-III and determined the exact time point of occurrence of AKI as the basis for the new-onset of AKI in ICU. XGBoost model performed the best prediction in all the time point at new-onset AKI 3–6–9–12 h ahead. For patients with the new-onset of AKI in ICU, platelets become an important factor associated with AKI.

Timing of vasopressin initiation and mortality in patients with septic shock: analysis of the MIMIC-III and MIMIC-IV databases

Backgroundvasopressin is commonly used as a second-line vasopressor for patients with septic shock, but the optimal timing of initiation is uncertain. This study was designed to investigate when vasopressin initiation may be beneficial for 28-day mortality in septic shock patients.MethodsThis was a retrospective observational cohort study from the MIMIC-III v1.4 and MIMIC-IV v2.0 databases. All adults diagnosed with septic shock according to Sepsis-3 criteria were included. Patients were stratified into two groups based on norepinephrine (NE) dose at the time of vasopressin initiation, defined as the low doses of NE group (NE<0.25 µg/kg/min) and the high doses of NE group (NE ≥ 0.25 µg/kg/min). The primary end‐point was 28‐day mortality after diagnosis of septic shock. The analysis involved propensity score matching (PSM), multivariable logistic regression, doubly robust estimation, the gradient boosted model, and an inverse probability‐weighting model.ResultsA total of 1817 eligible patients were included in our original cohort (613 in the low doses of NE group and 1204 in the high doses of NE group). After 1:1 PSM, 535 patients from each group with no difference in disease severity were included in the analysis. The results showed that vasopressin initiation at low doses of NE was associated with reduced 28-day mortality (odds ratio [OR] 0.660, 95% confidence interval [CI] 0.518–0.840, p < 0.001). Compared with patients in the high doses of NE group, patients in the low doses of NE group received significantly shorter duration of NE, with less intravenous fluid volume on the first day after initiation of vasopressin, more urine on the second day, and longer mechanical ventilation-free days and CRRT-free days. Nevertheless, there were no significant differences in hemodynamic response to vasopressin, duration of vasopressin, and ICU or hospital length of stay.ConclusionsAmong adults with septic shock, vasopressin initiation when low-dose NE was used was associated with an improvement in 28-day mortality.

PREDICTING MORTALITY RATE IN ICU USING MACHINE LEARNING: A STUDY

Mortality rate is the measure of number of death in a limited population or by a particular cause within a certain time period. In healthcare system Intensive Care unit (ICU) plays an important role for critical condition patients. Mortality prediction of critical condition ICU patients who needs special care is a major problem of concern. The focus of this work is to predict ICU patient’s mortality by the use of health record from ICU. Nowadays, machine learning plays an important role to resolve many health related issues which includes handling of patient’s health related data and records, development of new medical procedures and the treatment of disease like cancer, heart disease, stroke, diabetes and arthritis etc. Various machine learning models are used to analyze health records to come up with solutions for different health related issues. In this work, four popular supervised machine learning algorithms, Decision Tree(DT), Random Forest (RF), K-Nearest Neighbors (KNN) and Logistic Regression(LR) has been used to predict patients mortality in ICU. In this work, In Hospital Mortality Prediction dataset which is part of MIMIC-III database has been used. The dataset is available to download and free to use from Kaggle. In our work of mortality prediction, a maximum accuracy of 0.87 has been achieved.

Association between red blood cell distribution width and all-cause mortality in unselected critically ill patients: Analysis of the MIMIC-III database.

Although red cell distribution width (RDW) is widely observed in clinical practice, only a few studies have looked at all-cause mortality in unselected critically ill patients, and there are even fewer studies on long-term mortality. The goal of our study was to explore the prediction and inference of mortality in unselected critically ill patients by assessing RDW levels. We obtained demographic information, laboratory results, prevalence data, and vital signs from the unselected critically ill patients using the publicly available MIMIC-III database. We then used this information to analyze the association between baseline RDW levels and unselected critically ill patients using Cox proportional risk analysis, smoothed curve fitting, subgroup analysis, and Kaplan-Meier survival curves for short, intermediate, and long-term all-cause mortality in unselected critically ill patients. A total of 26,818 patients were included in our study for the final data analysis after screening in accordance with acceptable conditions. Our study investigated the relationship between RDW levels and all-cause mortality in a non-selected population by a smoothed curve fit plots and COX proportional risk regression models integrating cubic spline functions reported results about a non-linear relationship. In the fully adjusted model, we found that RDW values were positively associated with 30-day, 90-day, 365-day, and 4-year all-cause mortality in 26,818 non-selected adult patients with HRs of 1.10 95%CIs (1.08, 1.12); 1.11 95%CIs (1.10, 1.13); 1.13 95%CIs (1.12, 1.14); 1.13 95%CIs (1.12, 1.14). In unselected critically ill patients, RDW levels were positively associated with all-cause mortality, with elevated RDW levels increasing all-cause mortality.

A Machine Learning Approach to the Non-Invasive Estimation of Continuous Blood Pressure Using Photoplethysmography

Blood pressure is an important vital sign that sometimes requires continuous measurement. The current methods include cuff measurements (manual auscultation and oscillometric techniques) for non-continuous measurement and invasive arterial cannulation for continuous measurement. The use of photoplethysmography as a cuffless, non-invasive, and continuous blood pressure measurement system is investigated through the use of four neural networks. These predict the systolic blood pressure, diastolic blood pressure, mean arterial blood pressure, and waveform shape. The models are trained on 890 h of data from 1669 patients in the MIMIC-III database. Feature-trained artificial neural networks predict the systolic blood pressure to 5.26 ± 6.53 mmHg (mean error ± standard deviation), the diastolic blood pressure to 2.96 ± 3.31 mmHg, and the mean arterial pressure to 3.27 ± 3.55 mmHg. These are used to shift and scale the predicted waveform, allowing the waveform prediction neural network to optimise for the wave shape rather than the amplitude. The waveform prediction has 86.4% correlation with the actual arterial blood pressure waveform. All results meet international clinical blood pressure measurement standards and could potentially change how blood pressure is measured in both clinical and research settings. However, more data from healthy individuals and analysis of the models’ biases based on clinical features is required.

A pattern mixture model with long short-term memory network for acute kidney injury prediction

Acute kidney disease is a serious complication characterized by poor short- and long-term outcomes in the intensive care unit. Impairment in renal function of the kidney significantly increases the mortality rate. Early detection of acute kidney disease could lead to preventive interventions, therefore deep learning systems can detect it before its symptoms and consequences appear. We developed a novel deep learning architecture like Stacked long short-term memory network with pattern mixture approach for kidney injury prediction. A total of 33,754 patients encountered were retrospectively analyzed from the MIMIC-III database. A selection and pattern mixture model was used for preprocessing the time-series data. We compared the proposed result with conventional algorithms like gradient boosted trees and long short-term memory model. Our model was trained on patient time-series data for different time windows and obtained the highest accuracy of 92.4% for 12 h and 92.6% for 24 h. A novel stacked long short-term memory model outperforms the machine learning model, revealing superior performance in predicting kidney injury 24 h before onset.

Effects of propofol, benzodiazepines, and opioids on survival in cancer patients: a retrospective cohort study based on MIMIC-III

ABSTRACT Sedative and analgesic drugs are commonly used in the diagnosis and treatment of cancer patients. Analyzing the impact of these drugs on the prognosis of cancer patients can help improve patient outcomes. This study aimed to analyze the use of propofol, benzodiazepines, and opioids on the survival of cancer patients in the intensive care unit (ICU) based on the Medical Information Mart for Intensive Care III (MIMIC-III) database. A total of 2,567 cancer patients from the MIMIC-III database between 2001 and 2012 were included in this retrospective cohort study. Logistic regression analyses were utilized to assess the relationship between propofol, benzodiazepine, and opioid and survival in cancer patients. The follow-up was 1 year from the patient’s first admission to the ICU. Outcomes were ICU mortality, 28-day mortality, and 1-year mortality. Stratified analyses were based on patients’ metastatic status. The use of propofol [odds ratio (OR) = 0.66; 95% confidence interval (CI), 0.53–0.80] and opioids (OR = 0.65; 95%CI, 0.54–0.79) were associated with a decreased risk of 1-year mortality. Both benzodiazepines and opioids use were related to an increased risk of ICU mortality and 28-day mortality (all P < 0.05), whereas propofol use was associated with a decreased risk of 28-day mortality (OR = 0.59; 95%CI, 0.45–0.78). Compared with the use of benzodiazepines combined with opioids, patients who used propofol and opioids were related to a decreased risk of 1-year mortality (OR = 0.74; 95%CI, 0.55–0.98). Similar results were found in patients with metastasis and metastasis-free. Cancer patients who used propofol may experience a lower risk of mortality than benzodiazepine use.

Statin Use is Associated with Reduced Mortality in Mechanically Ventilated Patients: A Retrospective Propensity-Matched Analysis of MIMIC-III Database

BackgroundWe sought to determine whether statin treatment has a protective effect on the outcome of critically ill patients on mechanical ventilation.MethodsPatients who underwent mechanical ventilation were selected from the MIMIC-III database. Patients with statin usage were allocated into the statin cohort. Patients without any statin use were matched to the statin cohort in a 1:1 ratio by propensity score. To ensure the robustness of the findings, we utilized the gradient boosted model, propensity score analysis, doubly robust estimation and an inverse probability‐weighting model in the statistical procedure.ResultsBefore matching, 17,452 patients were enrolled in the non-statin group and 3,999 in the statin group. After matching, 3,363 patients were enrolled in each group. Compared with nonusers, the use of statins was associated with improved 28-day survival in the unmatched cohort (HR 0.85 95% CI 0.80–0.90, p < 0.01) and matched cohort (HR 0.79 95% CI 0.73–0.85, p < 0.01). Statin use was associated with longer ventilator-free days (VFD, 14.93 ± 13.11 vs 12.06 ± 13.26, p < 0.01) and longer ICU-free days (IFD, 13.41 ± 12.14 vs 10.86 ± 12.19, p < 0.01) in the matched cohort. Different types of statins were all associated with significantly improved 28-day survival.ConclusionsIn a population of mechanically ventilated patients, the use of statins may be associated with improved survival, longer VFDs and longer IFDs.

Integrating Structured and Unstructured EHR Data for Predicting Mortality by Machine Learning and Latent Dirichlet Allocation Method.

An ICU is a critical care unit that provides advanced medical support and continuous monitoring for patients with severe illnesses or injuries. Predicting the mortality rate of ICU patients can not only improve patient outcomes, but also optimize resource allocation. Many studies have attempted to create scoring systems and models that predict the mortality of ICU patients using large amounts of structured clinical data. However, unstructured clinical data recorded during patient admission, such as notes made by physicians, is often overlooked. This study used the MIMIC-III database to predict mortality in ICU patients. In the first part of the study, only eight structured variables were used, including the six basic vital signs, the GCS, and the patient's age at admission. In the second part, unstructured predictor variables were extracted from the initial diagnosis made by physicians when the patients were admitted to the hospital and analyzed using Latent Dirichlet Allocation techniques. The structured and unstructured data were combined using machine learning methods to create a mortality risk prediction model for ICU patients. The results showed that combining structured and unstructured data improved the accuracy of the prediction of clinical outcomes in ICU patients over time. The model achieved an AUROC of 0.88, indicating accurate prediction of patient vital status. Additionally, the model was able to predict patient clinical outcomes over time, successfully identifying important variables. This study demonstrated that a small number of easily collectible structured variables, combined with unstructured data and analyzed using LDA topic modeling, can significantly improve the predictive performance of a mortality risk prediction model for ICU patients. These results suggest that initial clinical observations and diagnoses of ICU patients contain valuable information that can aid ICU medical and nursing staff in making important clinical decisions.

Application of Neuromuscular Blockers in Patients with ARDS in ICU: A Retrospective Study Based on the MIMIC-III Database

Although neuromuscular blocker agents (NMBAs) are recommended by guidelines as a treatment for ARDS patients, the efficacy of NMBAs is still controversial. Our study aimed to investigate the association between cisatracurium infusion and the medium- and long-term outcomes of critically ill patients with moderate and severe ARDS. We performed a single-center, retrospective study of 485 critically ill adult patients with ARDS based on the Medical Information Mart for Intensive Care III (MIMIC-III) database. Propensity score matching (PSM) was used to match patients receiving NMBA administration with those not receiving NMBAs. The Cox proportional hazards model, Kaplan-Meier method, and subgroup analysis were used to evaluate the relationship between NMBA therapy and 28-day mortality. A total of 485 moderate and severe patients with ARDS were reviewed and 86 pairs of patients were matched after PSM. NMBAs were not associated with reduced 28-day mortality (hazard ratio (HR) 1.44; 95% CI: 0.85~2.46; p = 0.20), 90-day mortality (HR = 1.49; 95% CI: 0.92~2.41; p = 0.10), 1-year mortality (HR = 1.34; 95% CI: 0.86~2.09; p = 0.20), or hospital mortality (HR = 1.34; 95% CI: 0.81~2.24; p = 0.30). However, NMBAs were associated with a prolonged duration of ventilation and the length of ICU stay. NMBAs were not associated with improved medium- and long-term survival and may result in some adverse clinical outcomes.

Association between Mean Arterial Pressure during the First 24 Hours and Clinical Outcome in Critically Ill Stroke Patients: An Analysis of the MIMIC-III Database

Abnormal blood pressure is common in critically ill stroke patients. However, the association between mean arterial pressure (MAP) and mortality of critically ill stroke patients remains unclear. We extracted eligible acute stroke patients from the MIMIC-III database. The patients were divided into three groups: a low MAP group (MAP ≤ 70 mmHg), a normal MAP group (70 mmHg < MAP ≤ 90 mmHg), and a high MAP group (MAP > 90 mmHg). The Cox proportional hazards model and restricted cubic splines were used to assess the association between MAP and mortality. Sensitivity analyses were conducted to investigate whether MAP had different effects on mortality in different subpopulations. A total of 2885 stroke patients were included in this study. The crude 7-day and 28-day mortality was significantly higher in the low MAP group than that in the normal MAP group. By contrast, patients in the high MAP group did not have higher crude 7-day and 28-day mortality than those in the normal MAP group. After multiple adjustments using the Cox regression model, patients with low MAP were consistently associated with higher 7-day and 28-day mortality than those with normal MAP in the following subgroups: age > 60 years, male, those with or without hypertension, those without diabetes, and those without CHD (p < 0.05), but patients with high MAP were not necessarily associated with higher 7-day and 28-day mortality after adjustments (most p > 0.05). Using the restricted cubic splines, an approximately L-shaped relationship was established between MAP and the 7-day and 28-day mortality in acute stroke patients. The findings were robust to multiple sensitivity analyses in stroke patients. In critically ill stroke patients, a low MAP significantly increased the 7-day and 28-day mortality, while a high MAP did not, suggesting that a low MAP is more harmful than a high MAP in critically ill stroke patients.