Predict ICU Mortality Research Articles

Discovering Time-Aware Hidden Dependencies with Personalized Graphical Structure in Electronic Health Records

Over the past decade, significant advancements in mining electronic health records (EHRs) have enabled a broad range of decision-support applications, and offered an unprecedented capacity for predicting critical events such as disease prognosis and mortality in healthcare. Despite the availability of comprehensive coding systems in EHRs (e.g., ICD-9), which are designed to record diverse information on diseases, procedures, and medications over time, the complex and dynamic dependencies among the recorded data are usually not captured. This limitation often hinders the contextual understanding of medical observations for effective EHR representation learning. Therefore, there is a compelling need to discover a hidden “EHR graph” that represents the medical relationship between the observed features according to a patient’s history. These hidden graphs consisting of the medical codes from the same visits can offer a comprehensive insight derived from disease-to-disease, disease-to-drug, and drug-to-drug dependencies. However, it is still unclear how to address the challenge that the dependencies may vary from patient to patient, and they can dynamically evolve from one visit to another. To this end, we propose Timeaware Personalized Graph Transformer (TPGT), a novel attention-based time-aware hidden graph model, that captures the personalized graphical structures among observed medical codes and summarizes the temporal code dependencies over time to improve patient representation for outcome prediction. Built upon an intra-visit and an inter-visit dual-attention mechanism to model patients’ EHR graphs, our model offers an interpretability of what diagnosis or medication in a patient’s history can interact, and how those interactions may change over time. We conduct extensive experiments on two real-world EHR datasets for different healthcare predictive tasks: acute kidney injury (AKI) prediction and ICU mortality prediction. The experimental results demonstrate a significant performance improvement of the proposed model over baselines through multi-aspect quantitative evaluation. Furthermore, we perform various qualitative studies to validate the interpretability of the model which highlights the application of the proposed method in the context of personalized medicine.

Machine Learning-Based Prediction Model for ICU Mortality After Continuous Renal Replacement Therapy Initiation in Children.

Continuous renal replacement therapy (CRRT) is the favored renal replacement therapy in critically ill patients. Predicting clinical outcomes for CRRT patients is difficult due to population heterogeneity, varying clinical practices, and limited sample sizes. We aimed to predict survival to ICUs and hospital discharge in children and young adults receiving CRRT using machine learning (ML) techniques. Patients less than 25 years of age receiving CRRT for acute kidney injury and/or volume overload from 2015 to 2021 (80%). Internal validation occurred in a testing group of patients from the dataset (20%). Retrospective international multicenter study utilizing an 80/20 training and testing cohort split, and logistic regression with L2 regularization (LR), decision tree, random forest (RF), gradient boosting machine, and support vector machine with linear kernel to predict ICU and hospital survival. Model performance was determined by the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC) due to the imbalance in the dataset. Of the 933 patients included in this study, 538 (54%) were male with a median age of 8.97 years and interquartile range (1.81-15.0 yr). The ICU mortality was 35% and hospital mortality was 37%. The RF had the best performance for predicting ICU mortality (AUROC, 0.791 and AUPRC, 0.878) and LR for hospital mortality (AUROC, 0.777 and AUPRC, 0.859). The top two predictors of ICU survival were Pediatric Logistic Organ Dysfunction-2 score at CRRT initiation and admission diagnosis of respiratory failure. These are the first ML models to predict survival at ICU and hospital discharge in children and young adults receiving CRRT. RF outperformed other models for predicting ICU mortality. Future studies should expand the input variables, conduct a more sophisticated feature selection, and use deep learning algorithms to generate more precise models.

AI-Enhanced Strategies for COVID-19 Vaccination and Booster Prioritization: A Comprehensive Framework

Background: Distributing vaccines efficiently during the COVID-19 pandemic presented significant logistical challenges. To address the need for identifying populations at risk of breakthrough infections and those requiring booster shots, Emirates Health Services (EHS) developed a framework utilizing AI-driven digital solutions. Objective: To develop a machine learning (ML) model to identify individuals at risk of breakthrough infections and in need of booster doses, aiming to prioritize booster administration and reduce repeated infections among fully vaccinated individuals in the Northern Emirates of the UAE. Methods: A monitoring dashboard was developed using the EHS Intelligence (PaCE) platform. The study, conducted in three phases, created models to predict infection risk, COVID-19 severity in ICU patients and breakthrough infection risk, using data from the Wareed EMR system. Results: The AI models accurately identified high-risk individuals and predicted ICU mortality, achieving AUCs of 75% and 74% for infection risk, 94% and 91% for ICU mortality in training and validation datasets, observed 79% AUC with 85% accuracy for identifying high-risk groups for booster vaccination. Conclusion: The integration of AI in vaccination prioritization demonstrated its potential to enhance public health initiatives and improve pandemic management in the UAE.

Abstract 4123371: Hyperoxia is not a significant predictor of ICU mortality and is likely a surrogate parameter for reduced native cardiac function in patients supported with VA-ECMO

Intro: Hyperoxia has been associated with adverse outcomes in VA-ECMO patients. Whether hyperoxia represents a mechanism of mortality or is a surrogate for worse cardiac function is unknown. Aim: To determine patient characteristics, lab values, and hemodynamic parameters associated with increased mortality in VA-ECMO patients. Methods: Single-center retrospective cohort study at UCSF in VA-ECMO patients (2020-2023). Patients were categorized into 3 groups: Dead, Advanced Therapy-Free Survivors, and Advanced Therapy Survivors (survived to ICU discharge with LVAD or heart transplant). Wilcox rank sum and two sample proportion tests were performed. In-hospital mortality was evaluated using multiple logistic regression. Results: Of 179 patients, 87 died and 92 survived to ICU discharge. Of the survivors, 22 received advanced therapies. Compared to Advanced Therapy-Free Survivors, Dead patients had higher PaO2 levels (297[95% CI,164-400], p = 0.01). Dead and Advanced Therapy Survivors groups had similarly high PaO2 levels, but the latter group showed less acidosis and were more likely to have HFrEF (59%; 13/22; p = 0.02). ECMO support and VIS scores were similar. The multivariate model showed PaO2 is not associated with mortality (adjusted OR, 1.002 [95% CI, 0.99-1.004]; p = 0.131). Conclusion: Hyperoxia was associated with both death and advanced therapies, but did not predict mortality in a multivariate model. Advanced therapy survivors and patients that died were more likely to have HFrEF, suggesting that high PaO2 at cannulation reflects a proximal mixing cloud due to weak native cardiac function. Future work will analyze ECHO parameters to examine the association between hyperoxia and reduced cardiac output in VA-ECMO patients.

Monitoring Immune Dysfunction in Critically Ill Patients with Carbapenem-Resistant Acinetobacter baumannii Sepsis Treated with Regimens Including Cefiderocol: A Pilot Study to Identify Accessible Biomarkers to Stratify Patients' Prognosis.

Background: Multidrug-resistant Acinetobacter baumannii (CRAB) infections are a serious problem in critical care. This study aims to develop an early prognostic score for immune paralysis, using practical and cost-effective parameters, to predict ICU mortality in patients with CRAB infections being treated with Cefiderocol. Methods: We carried out an observational pilot study on consecutive patients hospitalized in the ICU with ensuing septic Acinetobacter baumannii infections treated with Cefiderocol monotherapy or Cefiderocol including combinations. We investigated the predictive power of lymphocyte counts, lymphocyte subpopulations, serum cholinesterase levels, and reactivation of herpes viruses. Results: Overall, 36 of 39 patients entered in our analysis: 20 survivors and 16 deceased. A total of 12 patients developed bacteremia, 19 patients had HAP/VAP, and 5 patients had a soft tissue infection. Univariate analyses of factors associated with unfavorable outcome revealed a significant association for age (OR: 1.5, CI: 1.11-2.02), SAPS II (OR: 1.05, CI: 1.01-1.1), SOFA score (OR: 1.37, CI: 1.06-1.76), lymphocytopenia (OR: 32.5, CI: 3.45-306.4), viral reactivation (OR: 9.75, CI: 1.72-55.4), and cholinesterase drop <1600 U/L (OR: 39.7, CI: 5.8-271.6). At variance, monotherapy or associations with Cefiderocol were not associated. In the final multivariable model, the only independent predictors of death were age (OR: 1.42, CI: 0.98-2.05), lymphocytopenia (OR: 18.2, CI: 0.87-371), and cholinesterase drop to below 1600 U/L (OR: 9.7, CI: 0.77-123.7). Conclusions: Age, lymphocytopenia, and serum cholinesterase drops, which were nearly significantly associated with an unfavorable outcome, may help pinpoint patients with acute immune paralysis during sepsis. Knowledge of such an immune state may in turn directly influence patients' care.

Feature Selection Methods for ICU Mortality Prediction Model

Feature Selection Methods for ICU Mortality Prediction Model

Evaluating the effectiveness of a sliding window technique in machine learning models for mortality prediction in ICU cardiac arrest patients

Extensive research has been devoted to predicting ICU mortality, to assist clinical teams managing critical patients. Electronic health records (EHR) contain both static and dynamic medical data, with the latter accumulating during ICU stays. Existing models often rely on a fixed time window (e.g., first 24 h) for prediction, potentially missing vital post-24-hour data. The present study aims to improve mortality prediction for ICU patients following Cardiac Arrest (CA) using a dynamic sliding window approach that accommodates evolving data characteristics.Our cohort included 2331 CA patients, of whom 684 died in the ICU and 1647 survived. Applying the sliding window technique, we created six different time windows and used each separately for model training and validation. We compared our results to a baseline accumulative window.The different time windows created by the sliding window technique differed in their prediction performance and outperformed the baseline 24-hour window significantly. The XGBoost model outperformed all other models, with the 30–42 h time window achieving the best results (AUC = 0.8, accuracy = 0.77).Our work shows that the sliding window technique is effective in improving mortality prediction. We demonstrated how important time-window selection is and showed that enhancing it can save time and thus improve mortality prediction. These findings promise to improve the clinical team’s efficiency in prioritizing patients and giving greater attention to higher-risk patients.To conclude, mortality prediction in the ICU can be improved if we consider alternative time windows instead of the 24-hour window, which is currently the most widely accepted among scoring systems today.

Machine Learning-Based Prediction Models of Mortality for Intensive Care Unit Patients Using Nursing Records.

This study aimed to develop ICU mortality prediction models using a conceptual framework, focusing on nurses' concerns reflected in nursing records from the MIMIC IV database. We included 46,693 first-time ICU admissions of adults over 18 years with a minimum 24-hour stay, excluding those receiving hospice or palliative care. Predictors included demographics, clinical characteristics, and nursing documentation frequencies related to nurses' concerns. Four models were trained with 10-fold cross-validation after adjusting class imbalance. The random forest (RF) model was identified as the best-performing, with key predictors of mortality in this model being the frequency of vital signs, the frequency of nursing note documentation, and the frequency of monitoring-related nursing notes. This suggests that predictive models using nursing records, which reflect nurses' concerns as represented by the frequency of nursing documentation, may be integrated into clinical decision support tools, potentially enhancing patient outcomes in ICUs.

SOFA in sepsis: with or without GCS

PurposeSepsis is a global public health burden. The sequential organ failure assessment (SOFA) is the most commonly used scoring system for diagnosing sepsis and assessing severity. Due to the widespread use of endotracheal intubation and sedative medications in sepsis, the accuracy of the Glasgow Coma Score (GCS) is the lowest in SOFA. We designed this multicenter, cross-sectional study to investigate the predictive efficiency of SOFA with or without GCS on ICU mortality in patients with sepsis.MethodsFirst, 3048 patients with sepsis admitted to Peking Union Medical College Hospital (PUMCH) were enrolled in this survey. The data were collected from June 8, 2013 to October 12, 2022. Second, 18,108 patients with sepsis in the eICU database were enrolled. Third, 2397 septic patients with respiratory system ≥ 3 points in SOFA in the eICU database were included. We investigated the predictive efficiency of SOFA with or without GCS on ICU mortality in patients with sepsis in various ICUs of PUMCH, and then we validated the results in the eICU database.Main resultsIn data of ICUs in PUMCH, the predictive efficiency of SOFA without GCS (AUROC [95% CI], 24 h, 0.724 [0.688, 0.760], 48 h, 0.734 [0.699, 0.769], 72 h, 0.748 [0.713, 0.783], 168 h, 0.781 [0.747, 0.815]) was higher than that of SOFA with GCS (AUROC [95% CI], 24 h, 0.708 [0.672, 0.744], 48 h, 0.721 [0.685, 0.757], 72 h, 0.735 [0.700, 0.757], 168 h, 0.770 [0.736, 0.804]) on ICU mortality in patients with sepsis, and the difference was statistically significant (P value, 24 h, 0.001, 48 h, 0.003, 72 h, 0.004, 168 h, 0.005). In septic patients with respiratory system ≥ 3 points in SOFA in the eICU database, although the difference was not statistically significant (P value, 24 h, 0.148, 48 h, 0.178, 72 h, 0.132, 168 h, 0.790), SOFA without GCS (AUROC [95% CI], 24 h, 0.601 [0.576, 0.626], 48 h, 0.625 [0.601, 0.649], 72 h, 0.639 [0.615, 0.663], 168 h, 0.653 [0.629, 0.677]) had a higher predictive efficiency on ICU mortality than SOFA with GCS (AUROC [95% CI], 24 h, 0.591 [0.566, 0.616], 48 h, 0.616 [0.592, 0.640], 72 h, 0.628 [0.604, 0.652], 168 h, 0.651 [0.627, 0.675]).ConclusionsIn severe sepsis, it is realistic and feasible to discontinue the routine GCS for SOFA in patients with a respiratory system ≥ 3 points, and even better predict ICU mortality.

Dynamic assessment of prealbumin for nutrition support effectiveness in critically ill patients

Background & AimsSerum prealbumin is considered to be a sensitive predictor of clinical outcomes and a quality marker for nutrition support. However, its susceptibility to inflammation restricts its usage in critically ill patients according to current guidelines. We assessed the performance of the initial value of prealbumin and dynamic changes for predicting the ICU mortality and the effectiveness of nutrition support in critically ill patients. MethodsThis monocentric study included patients admitted to the ICU between 2009-2016, having at least one initial prealbumin value available. Prospectively recorded data were extracted from the electronic ICU charts. We used both univariable and multivariable logistic regressions to estimate the performance of prealbumin for the prediction of ICU mortality. Additionally, the association between prealbumin dynamic changes and nutrition support was assessed via a multivariable linear mixed-effects model and multivariable linear regression. Performing subgroup analysis assisted in identifying patients for whom prealbumin dynamic assessment holds specific relevance. ResultsWe included 3136 patients with a total of 4942 prealbumin levels available. Both prealbumin measured at ICU admission (adjusted odds-ratio (aOR) 0.04, confidence interval (CI) 95% 0.01-0.23) and its change over the first week (aOR 0.02, CI95 0.00-0.19) were negatively associated with ICU mortality. Throughout the entire ICU stay, prealbumin dynamic changes were associated with both cumulative energy (estimate: 33.2, standard error (SE) 0.001, p < 0.01) and protein intakes (1.39, SE 0.001, p<0.01). During the first week of stay, prealbumin change was independently associated with mean energy (6.03e-04, SE 2.32e-04, p<0.01) and protein intakes (1.97e-02, SE 5.91e-03, p<0.01). Notably, the association between prealbumin and energy intake was strongest among older or malnourished patients, those suffering from increased inflammation and those with high disease severity. Finally, prealbumin changes were associated with a positive mean nitrogen balance at day 7 only in patients with SOFA<4 (p=0.047). ConclusionPrealbumin measured at ICU admission and its change during the first-week serve as an accurate predictor of ICU mortality. Prealbumin dynamic assessment may be a reliable tool to estimate the effectiveness of nutrition support in the ICU, especially among high-risk patients.

Understanding the Impact of COVID-19 on Roma Vulnerable Communities in Western Romania: Insights and Predictive Factors from a Retrospective Study.

The COVID-19 pandemic disproportionately affected vulnerable populations like Roma patients in Western Romania due to marginalization and limited healthcare access. A retrospective study analyzed COVID-19 cases between March 2020 and August 2022 using data from the Directorate of Public Health in Timis county. Demographic, epidemiological, clinical, and laboratory data were assessed, along with risk factors and biomarkers for ICU admission and mortality prediction. The following biomarkers were assessed: C-reactive protein (CRP), ferritin (FER), IL-6, D-dimers, lactate dehydrogenase (LDH), high density lipoprotein cholesterol (HDL), and 25-OH vitamin D (25-OHD). In comparison with the general population (GP), Roma patients were more overweight (p = 0.0292), came from rural areas (p = 0.0001), could not recall transmission source (p = 0.0215), were admitted to the intensive care unit (ICU, p = 0.0399) more frequently, had worse symptomatology (p = 0.0490), showed more elevated levels of CRP (p = 0.0245) and IL-6 (p < 0.0001) and lower levels of HDL (p = 0.0008) and 25-OHD (p = 0.0299). A stronger, significant correlation was observed between CRP and severity (rho = 0.791 vs. 0.433 in GP), and an inverse stronger significant one was observed between HDL and severity (rho = -0.850 vs. -0.734 in GP) in the Roma patients. The male sex continues to be an important risk factor for ICU admission (OR = 2.379) and death (OR = 1.975), while heavy smoking was more important in relation to ICU admission (OR = 1.768). Although the Roma ethnicity was 1.454 times more at risk of ICU admission than the GP, this did not prove statistically significant (p = 0.0751). CRP was the most important predictive factor in regards to admission to the ICU for both Roma (OR = 1.381) and the GP (OR = 1.110) and in regards to death (OR = 1.154 for Roma, OR = 1.104 for GP). A protective effect of normal values of HDL and 25-OHD was observed in the GP for both ICU admission (OR = 0.947, 0.853, respectively) and death (OR = 0.920, 0.921, respectively), while for the Roma group, normal 25-OHD values were only considered protective in regards to death (OR = 0.703). Cutoff values for ICU admission were 28.98 mg/L for Roma and 29.03 mg/L for GP patients, with high specificity for both groups (over 95). Higher rates of ICU admissions, severe symptomatology, and distinct laboratory biomarker profiles among Roma patients emphasize the critical importance of personalized care strategies and targeted interventions to mitigate the disproportionate burden of COVID-19 on vulnerable communities. CRP values at admission have had a clear impact as a risk assessment biomarker for Roma patients, while the significance of IL-6, HDL, and 25-OHD should also not be overlooked in these patients.

A tertiary care center-based study of a novel ‘ICU Mortality and Prolonged Stay Risk Scoring System’

Background &amp; objective: Intensive care has been associated with high cost and resource-intensive medical care. Therefore, a risk prediction model is required to plan time allocation, human resources, and the required equipment. Various risk predictions for ICU mortality and ‘Prolonged Length of Stay’ (PLOS) scores are already available. Still, the established model, such as the APACHE IV score or SAPS II, sometimes became impractical since they required many laboratory parameters. A model based on co-morbidities and demographic factors may be more useful in limited resources setting. Hence, we developed a simple ICU mortality and PLOS risk prediction model based on co-morbidities and demographic data. Methodology: This retrospective cohort study was performed to develop a risk scoring for mortality and PLOS, using data from Dr. Sardjito Hospital Yogyakarta database between January 01-December 31, 2019. Logistic regression and bootstrap methods were used to create a risk score for estimating the risk. The discrimination performance of the model was evaluated using the area under the curve (AUC) of the receiver operating characteristic (ROC). The Hosmer-Lemeshow test was employed to assess the model’s calibration. Results: A total of 415 patients were included in this study. The risk factors for mortality were perioperative support medication, kidney failure, neurologic disorder, respiratory failure, and intraoperative blood transfusion. The mortality score of 6 was associated with a 100% probability of mortality. Medical cases, GCS &lt; 8, vasoactive/inotropic medication, sepsis, respiratory failure, and kidney failure were the risk factors for PLOS. PLOS score of 3 was associated with a 100% probability of PLOS. The discrimination for either mortality or PLOS was considered excellent with the AUC (± 95% CI) for mortality 0.896 (0.853-0.94), while for PLOS 0.878 (0.80-0.90). The calibration test found that both models had good calibration with P values of 0.53 and 0.55 for mortality and PLOS, respectively. Conclusion: The ‘Mortality and Prolonged Length of Stay Prediction Score’ based on co-morbidities and demographic data upon admission to ICU had good accuracy and can be applied as a potential new scoring system in healthcare institutions. Abbreviations: APACHE- Acute Physiologic Assessment and Chronic Health Evaluation; AUC; Area Under the Curve GCS- Glasgow Coma Scale; ICU- Intensive Care Unit; PLOS- ‘Prolonged Length of Stay’; PRC- Packed Red Cells; SAPS- Simplified Acute Physiology Score Keywords: Risk Scoring; Mortality; Prolonged Length of Stay; ICU Citation: Widyastuti Y, Jufan AY, Widodo U, Wisudarti CFR, Sudadi, Fauzi RA, Ardiansyah F. A tertiary care center-based study of a novel ‘ICU Mortality and Prolonged Stay Risk Scoring System’. Anaesth. pain intensive care 2024;28(1):100−107; DOI: 10.35975/apic.v28i1.2382 Received: October 02, 2023; Reviewed: November 27, 2023; Accepted: December 17, 2023

Hematological Indices for Identifying Adverse Outcomes in Children Admitted to Pediatric ICUs.

The pediatric ICU (PICU) is a specialized area where critically sick children are managed. The mortality rates in PICUs are higher in developing countries as compared to developed nations.Many of these deaths could be prevented if very sick children were identified soon after they arrived at the health facility. Hematological indices like platelet lymphocyte ratio (PLR) and neutrophil-lymphocyte ratio (NLR) have been frequently used in adults as indicators of mortality. However, their use in the pediatric population is limited due to a lack of validated reference intervals. The objective of the study is to assess the role of hematological indices in identifying adverse outcomes in terms of mortality in children admitted to the PICU. It is a prospective, observational study done at a tertiary care hospital. All children aged one year to 12 years admitted to the PICU were enrolled in the study. A sample for complete blood count was taken within one hour of admission to the PICU. Children who had received blood products in the last two months, those on chronic medications (>two weeks) that can affect bone marrow cellularity, and known cases ofhematological disorders such as megaloblastic anemia, hematological malignancies, immune thrombocytopenia, and aplastic anemia were excluded from the study.PLR, NLR, and platelets to mean platelet volume ratio (PLT/MPV) were determined and compared among the survivors and non-survivors. Out of 275 enrolled patients, 119 (43.3%) patients expired during the study period. While PLR had high sensitivity and NLR had high specificity (85.71% and 92.31%, respectively) for predicting mortality, none of these parameters had a good area under the curve (AUC) in our study.PLT/MPV of ≥32 had a sensitivity of 39.5% and a specificity of 56.41%for predicting mortality. Hematological parameters have been used across the world to predict ICU mortality. PLR and NLR are simple hematological biomarkers, easy to calculate, and cost-effective, and ratios are better than individual parameters. More studies and stratified samples are required to evaluate the role of hematological markers in identifying the risk of mortality in children admitted to PICUs.

Comprehensive performance assessment of Multi-neural ensemble model for mortality prediction in ICU

Comprehensive performance assessment of Multi-neural ensemble model for mortality prediction in ICU

Lactate and base excess (BE) as markers of hypoperfusion and mortality in traumatic hemorrhagic shock in patients undergoing Damage Control: a historical cohort.

hemorrhagic shock is a significant cause of trauma-related deaths in Brazil and worldwide. This study aims to compare BE and lactate values at ICU admission and twenty-four hours after in identifying tissue hypoperfusion and mortality. examines a historical cohort of trauma patients over eitheen years old submittet to damage control resuscitation approch upon hospital admission and were then admitted to the ICU. We collected and analyzed ISS, mechanism and type of trauma, need for renal replacement therapy, massive transfusion. BE, lactate, pH, bicarbonate at ICU admission and twenty-four hours later, and mortality data. The patients were grouped based on their BE values (≥-6 and <-6mmol/L), which were previously identified in the literature as predictors of severity. They were subsequently redivided using the most accurate values found in this sample. In addition to performing multivariate binary logistic regression. The data were compared using several statistical tests due to diversity and according to the indication for each variable. there were significant changes in perfusion upon admission to the Intensive Care Unit. BE is a statistically significant value for predicting mortality, as determined by using values from previous literature and from this study. the results demonstrate the importance of monitoring BE levels in the prediction of ICU mortality. BE proves to be a valuable bedside marker with quick results and wide availability.

Factor analysis based on SHapley Additive exPlanations for sepsis-associated encephalopathy in ICU mortality prediction using XGBoost - a retrospective study based on two large database.

Sepsis-associated encephalopathy (SAE) is strongly linked to a high mortality risk, and frequently occurs in conjunction with the acute and late phases of sepsis. The objective of this study was to construct and verify a predictive model for mortality in ICU-dwelling patients with SAE. The study selected 7,576 patients with SAE from the MIMIC-IV database according to the inclusion criteria and randomly divided them into training (n = 5,303, 70%) and internal validation (n = 2,273, 30%) sets. According to the same criteria, 1,573 patients from the eICU-CRD database were included as an external test set. Independent risk factors for ICU mortality were identified using Extreme Gradient Boosting (XGBoost) software, and prediction models were constructed and verified using the validation set. The receiver operating characteristic (ROC) and the area under the ROC curve (AUC) were used to evaluate the discrimination ability of the model. The SHapley Additive exPlanations (SHAP) approach was applied to determine the Shapley values for specific patients, account for the effects of factors attributed to the model, and examine how specific traits affect the output of the model. The survival rate of patients with SAE in the MIMIC-IV database was 88.6% and that of 1,573 patients in the eICU-CRD database was 89.1%. The ROC of the XGBoost model indicated good discrimination. The AUCs for the training, test, and validation sets were 0.908, 0.898, and 0.778, respectively. The impact of each parameter on the XGBoost model was depicted using a SHAP plot, covering both positive (acute physiology score III, vasopressin, age, red blood cell distribution width, partial thromboplastin time, and norepinephrine) and negative (Glasgow Coma Scale) ones. A prediction model developed using XGBoost can accurately predict the ICU mortality of patients with SAE. The SHAP approach can enhance the interpretability of the machine-learning model and support clinical decision-making.

Abstract 18281: Identifying Predictors of ICU Mortality Outcomes Among Critically Ill Patients With Atrial Fibrillation Using Machine Learning Models

Introduction: New onset atrial fibrillation (Afib) is a frequent complication of intensive care admission and is linked to higher mortality in the ICU. We aimed to develop machine learning (ML) models to predict the risk of death in the ICU among these patients and assess their long-term prognosis. Methods: Eligible patients with Afib from the Medical Information Mart for Intensive Care IV database were included. The cases were randomly divided into training (75%) and test (25%) groups. Machine learning algorithms such as light GBM, random forest, gradient boosted classifier, and XGBoost were used. In addition, we utilized a tabular deep learning model architecture comprised of an embedding layer for categorical variables. The robustness of the results was ensured through K-fold cross-validation. AUROC and accuracy were used to compare model performance. Models accounted for variables from demographics, labs, charting events, and comorbidity domains. A web-based calculator was also developed to encourage the model's practical application. Results: The sample (n = 8455) consisted of 3408 females (40.3%), 6004 whites (71.0%), and 4876 (57.7%) patients with Medicare insurance. The ICU mortality was 9.5% (n = 802). The median age (79 vs. 75 years; p &lt; 0.001) and ICU LOS (4 vs. 2 days; p &lt; 0.001) were higher among the patients who expired. Congestive heart failure was reported in 15% of cases. The lightGBM model (AUROC: 84%) performed best and showed that mean arterial BP, WBC, PO2, platelet count, age, hemoglobin, BUN, pH, ICU LOS and anion gap were the top predictors of ICU mortality among this population. Conclusion: Death in the ICU among patients with Afib was associated with several clinical features, especially BP, WBC count, PO2, platelet count, age, and hemoglobin. Predictors of ICU mortality were identified with high accuracy and this may aid in the creation of a real time risk score for selective intervention to reduce the risk of death in this vulnerable patient population.

A Transcriptomic Severity Classifier IMX-SEV-3b to Predict Mortality in Intensive Care Unit Patients with COVID-19: A Prospective Observational Pilot Study.

The prediction of disease outcomes in COVID-19 patients in the ICU is of critical importance, and the examination of host gene expressions is a promising tool. The 29-host mRNA Inflam-matix-Severity-3b (IMX-SEV-3b) classifier has been reported to predict mortality in emergency department COVID-19 patients and surgical ICU patients. The accuracy of the IMX-SEV-3b in predicting mortality in COVID-19 patients admitted to the ICU is yet unknown. Our aim was to investigate the accuracy of the IMX-SEV-3b in predicting the ICU mortality of COVID-19 patients. In addition, we assessed the predictive performance of routinely measured biomarkers and the Sequential Organ Failure Assessment (SOFA) score as well. This was a prospective observational study enrolling COVID-19 patients who received mechanical ventilation on the ICU of the Erasmus MC, the Netherlands. The IMX-SEV-3b scores were generated by amplifying 29 host response genes from blood collected in PAXgene® Blood RNA tubes. A severity score was provided, ranging from 0 to 1 for increasing disease severity. The primary outcome was the accuracy of the IMX-SEV-3b in predicting ICU mortality, and we calculated the AUROC of the IMX-SEV-3b score, the biomarkers C-reactive protein (CRP), D-dimer, ferritin, leukocyte count, interleukin-6 (IL-6), lactate dehydrogenase (LDH), neutrophil-to-lymphocyte ratio (NLR), procalcitonin (PCT) and the SOFA score. A total of 53 patients were included between 1 March and 30 April 2020, with 47 of them being included within 72 h of their admission to the ICU. Of these, 18 (34%) patients died during their ICU stay, and the IMX-SEV-3b scores were significantly higher in non-survivors compared to survivors (0.65 versus 0.57, p = 0.05). The Area Under the Receiver Operating Characteristic Curve (AUROC) for prediction of ICU mortality by the IMX-SEV-3b was 0.65 (0.48-0.82). The AUROCs of the biomarkers ranged from 0.52 to 0.66, and the SOFA score had an AUROC of 0.81 (0.69-0.93). The AUROC of the pooled biomarkers CRP, D-dimer, ferritin, leukocyte count, IL-6, LDH, NLR and PCT for prediction of ICU mortality was 0.81 (IQR 0.69-0.93). Further validation in a larger interventional trial of a point-of-care version of the IMX-SEV-3b classifier is warranted to determine its value for patient management.

Prognostic Significance of Nucleated RBCs in Predicting Mortality Among ST-Elevation Myocardial Infarction Patients Admitted to the ICU.

Background The nucleated red blood cells (NRBCs) are a readily available hematological parameter with potential for risk stratification for mortality. Therefore, our objective was to assess the predictive significance of NRBCs for ICU mortality among ST-elevation myocardial infarction (STEMI) patients admitted to an ICU. Additionally, we aimed to compare the predictive capacity of NRBCs with that of the acute physiology and chronic health evaluation (APACHE) II score and the sequential organ failure assessment (SOFA) score. Methodology This descriptive cross-sectional study was conducted in the ICU of the National Institute of Cardiovascular Diseases (NICVD) in Karachi, Pakistan, from the 1st of February to the 30th of June, 2023. We included adult patients (≥18 years) diagnosed with STEMI who were subsequently admitted to the ICU. NRBCs were assessed in all patients over up to five days at 24-hour intervals, and the highest NRBC levels were used for the final analysis. Furthermore, the APACHE II score and the SOFA score were also documented. Patients were monitored throughout their ICU stay, and any adverse events or complications, such as re-intubation, bleeding necessitating transfusion, requirement for renal replacement therapy, arrhythmias, re-infarction, and mortality, were recorded. Results This study included 151 patients, of whom 97 (64.2%) were male, with an average age of 61.1 ± 10.7 years. Patients with positive NRBCs had higher mean SOFA scores (7.4 ± 2.9 vs. 5.4 ± 2.6; p < 0.001) and APACHE II scores (14.6 ± 6.3 vs. 12.6 ± 5.5; p = 0.037) compared to those with negative NRBCs. The culprit vessel showed greater mean stenosis (%) in patients with positive NRBCs (98.8 ± 3.0% vs. 96.8 ± 5.7%; p = 0.004). Post-procedure thrombolysis in myocardial infarction (TIMI) flow grade III was lower in patients with positive NRBCs (77.8% vs. 91.8% for positive vs. negative NRBCs, respectively). Moreover, patients with positive NRBCs experienced significantly higher mortality rates (63% vs. 8.2%; p < 0.001), a higher occurrence of arrhythmias (35.2% vs. 19.6%; p = 0.034), and an increased requirement for vasopressors/inotropic support (96.3% vs. 71.1%; p < 0.001) compared to those with negative NRBCs. NRBCs demonstrated superior discriminatory ability compared to the SOFA and APACHE II scores, with an area under the curve of 0.818 (95% CI: 0.738-0.899) for NRBCs, 0.774 (95% CI: 0.692-0.857) for SOFA, and 0.707 (95% CI: 0.613-0.801) for APACHE II. Positive NRBCs exhibited a sensitivity of 81.0% and a specificity of 81.7% in predicting ICU mortality. Conclusion In conclusion, positive NRBCs emerge as a robust and reliable prognostic indicator, strongly associated with an elevated risk of ICU mortality in STEMI patients. Moreover, the predictive power of positive NRBCs surpasses that of both SOFA and APACHE II scoring systems.

Predicting ICU Mortality in Acute Respiratory Distress Syndrome Patients Using Machine Learning: The Predicting Outcome and STratifiCation of severity in ARDS (POSTCARDS) Study.

To assess the value of machine learning approaches in the development of a multivariable model for early prediction of ICU death in patients with acute respiratory distress syndrome (ARDS). A development, testing, and external validation study using clinical data from four prospective, multicenter, observational cohorts. A network of multidisciplinary ICUs. A total of 1,303 patients with moderate-to-severe ARDS managed with lung-protective ventilation. None. We developed and tested prediction models in 1,000 ARDS patients. We performed logistic regression analysis following variable selection by a genetic algorithm, random forest and extreme gradient boosting machine learning techniques. Potential predictors included demographics, comorbidities, ventilatory and oxygenation descriptors, and extrapulmonary organ failures. Risk modeling identified some major prognostic factors for ICU mortality, including age, cancer, immunosuppression, Pa o2 /F io2 , inspiratory plateau pressure, and number of extrapulmonary organ failures. Together, these characteristics contained most of the prognostic information in the first 24 hours to predict ICU mortality. Performance with machine learning methods was similar to logistic regression (area under the receiver operating characteristic curve [AUC], 0.87; 95% CI, 0.82-0.91). External validation in an independent cohort of 303 ARDS patients confirmed that the performance of the model was similar to a logistic regression model (AUC, 0.91; 95% CI, 0.87-0.94). Both machine learning and traditional methods lead to promising models to predict ICU death in moderate/severe ARDS patients. More research is needed to identify markers for severity beyond clinical determinants, such as demographics, comorbidities, lung mechanics, oxygenation, and extrapulmonary organ failure to guide patient management.