Severity of illness scoring systems in the intensive care unit. Crit Care Med. 2011;39:163-169. [PMID: 20838329 DOI: 10.1097/ccm.0b013e3181f96f81]

Association of hydration status and in-hospital mortality in critically ill patients with ischemic stroke: Data from the MIMIC-IV database

BACKGROUND

Hydration plays a critical role in the pathophysiological progression of ischemic stroke. However, the impact of extreme hydration on the mortality of critically ill patients with ischemic stroke remains unclear. Therefore, our objective was to evaluate the association between hydration, as indicated by the blood urea nitrogen to creatinine ratio (UCR), and in-hospital mortality in critically ill patients with ischemic stroke.

METHODS

Data from the Medical Information Mart for Intensive Care (MIMIC-IV) database were utilized. Patients with ischemic stroke admitted to the Intensive Care Unit (ICU) for the first time were identified. The exposure variable was the hydration state represented by the UCR. The study outcome measure was in-hospital mortality. The primary analytical approach involved multivariate Cox regression analysis. Kaplan-Meier curves were constructed, and subgroup analyses with interaction were performed.

RESULTS

A total of 1539 patients, with a mean age of 69.9 years, were included in the study. Kaplan-Meier curves illustrated that patients in higher UCR tertiles exhibited increased in-hospital mortality. Accordingly, the risk of in-hospital mortality significantly rose by 29 % with every 10 units increase in UCR. Subgroup analysis indicated a robust association between UCR and in-hospital mortality in each subgroup, with no statistically significant interactions observed.

CONCLUSION

Hydration status is significantly associated with in-hospital all-cause mortality in critically ill patients with ischemic stroke. This finding underscores the importance of closely monitoring critically ill patients for adequate hydration and implementing appropriate rehydration strategies.

Comparative Predictive Accuracies of the Simplified Mortality Score for the Intensive Care Unit, Sepsis Severity Score, and Standard Severity Scores for 90-day Mortality in Sepsis Patients

Background

The standard severity scores were used for predicting hospital mortality of intensive care unit (ICU) patients. Recently, the new predictive score, Simplified Mortality Score for the ICU (SMS-ICU), was developed for predicting 90-day mortality.

Objective

To validate the ability of the SMS-ICU and compare with sepsis severity score (SSS) and original severity scores for predicting 90-day mortality in sepsis patients.

Method

An analysis of retrospective data was conducted in the ICU of a university teaching hospital. Also, 90-day mortality was used for the primary outcome.

Results

A total of 1,161 patients with sepsis were included. The 90-day mortality was 42.4%. The SMS-ICU presented the area under the receiver operating characteristic curve (AUROC) of 0.71, whereas the SSS had significantly higher AUROC than that of the SMS-ICU (AUROC 0.876, p < 0.001). The acute physiology and chronic health evaluation (APACHE) II and IV, and the simplified acute physiology scores (SAPS) II demonstrated good discrimination, with an AUROC above 0.90. The SMS-ICU provides poor calibration for 90-day mortality prediction, similar to the SSS and other standard severity scores. Furthermore, 90-day mortality was underestimated by the SMS-ICU, which had a standardized mortality ratio (SMR) of 1.36. The overall performance by Brier score demonstrated that the SMS-ICU was inferior to the SSS (0.222 and 0.169, respectively). Also, SAPS II presented the best overall performance with a Brier score of 0.092.

Conclusion

The SMS-ICU indicated lower performance compared to the SSS, standard severity scores. Consequently, modifications are required to enhance the performance of the SMS-ICU.

How to cite this article

Sathaporn N, Khwannimit B. Comparative Predictive Accuracies of the Simplified Mortality Score for the Intensive Care Unit, Sepsis Severity Score, and Standard Severity Scores for 90-day Mortality in Sepsis Patients. Indian J Crit Care Med 2024;28(4):343-348.

A novel prognostic model to predict mortality in patients with acute-on-chronic liver failure in intensive care unit

Acute-on-chronic liver failure (ACLF) implies high short-term mortality rates and usually requires intensive care unit (ICU) admission. Proper prognosis for these patients is crucial for early referral for liver transplantation. The superiority of CLIF-C ACLF score in Asian patients with ACLF admitted to an ICU remains inconclusive when compared to other scoring systems. The purpose of the study is (i) to compare the predictive performance of original MELD, MELD-Lactate, CLIF-C ACLF, CLIF-C ACLF-Lactate, and APACHE-II scores for short-term mortality assessment. (ii) to build and validate a novel scoring system and to compare its predictive performance to that of the original five scores. Two hundred sixty-five consecutive cirrhotic patients with ACLF who were admitted to our ICU were enrolled. The prognostic values for mortality were assessed by ROC analysis. A novel model was developed and internally validated using fivefold cross-validation. Alcohol abuse was identified as the primary etiology of cirrhosis. The AUROC of the five prognostic scores were not significantly superior to each other in predicting 1-month and 3-month mortality. The newly developed prognostic model, incorporating age, alveolar-arterial gradient (A-a gradient), BUN, total bilirubin level, INR, and HE grades, exhibited significantly improved performance in predicting 1-month and 3-month mortality with AUROC of 0.863 and 0.829, respectively, as compared to the original five prognostic scores. The novel ACLF model seems to be superior to the original five scores in predicting short-term mortality in ACLF patients admitted to an ICU. Further rigorous validation is required.

Advancing prognostic precision in pulmonary embolism: A clinical and laboratory-based artificial intelligence approach for enhanced early mortality risk stratification

BACKGROUND

Acute pulmonary embolism (PE) is a critical medical emergency that necessitates prompt identification and intervention. Accurate prognostication of early mortality is vital for recognizing patients at elevated risk for unfavourable outcomes and administering suitable therapy. Machine learning (ML) algorithms hold promise for enhancing the precision of early mortality prediction in PE patients.

OBJECTIVE

To devise an ML algorithm for early mortality prediction in PE patients by employing clinical and laboratory variables.

METHODS

This study utilized diverse oversampling techniques to improve the performance of various machine learning models including ANN, SVM, DT, RF, and AdaBoost for early mortality prediction. Appropriate oversampling methods were chosen for each model based on algorithm characteristics and dataset properties. Predictor variables included four lab tests, eight physiological time series indicators, and two general descriptors. Evaluation used metrics like accuracy, F1_score, precision, recall, Area Under the Curve (AUC) and Receiver Operating Characteristic (ROC) curves, providing a comprehensive view of models' predictive abilities.

RESULTS

The findings indicated that the RF model with random oversampling exhibited superior performance among the five models assessed, achieving elevated accuracy and precision alongside high recall for predicting the death class. The oversampling approaches effectively equalized the sample distribution among the classes and enhanced the models' performance.

CONCLUSIONS

The suggested ML technique can efficiently prognosticate mortality in patients afflicted with acute PE. The RF model with random oversampling can aid healthcare professionals in making well-informed decisions regarding the treatment of patients with acute PE. The study underscores the significance of oversampling methods in managing imbalanced data and emphasizes the potential of ML algorithms in refining early mortality prediction for PE patients.

Predictive Modeling Using Artificial Intelligence and Machine Learning Algorithms on Electronic Health Record Data: Advantages and Challenges

The rapid adoption of electronic health record (EHR) systems in US hospitals from 2008 to 2014 produced novel data elements for analysis. Concurrent innovations in computing architecture and machine learning (ML) algorithms have made rapid consumption of health data feasible and a powerful engine for clinical innovation. In critical care research, the net convergence of these trends has resulted in an exponential increase in outcome prediction research. In the following article, we explore the history of outcome prediction in the intensive care unit (ICU), the growing use of EHR data, and the rise of artificial intelligence and ML (AI) in critical care.

Association between anion gap and all-cause mortality of critically ill surgical patients: a retrospective cohort study

BACKGROUND

There are few widely accepted and operationally feasible models for predicting the mortality risk of patients in surgical intensive care unit (SICU). Although serum anion gap (AG) is known to be correlated with severe metabolic acidosis, no investigations have been reported about the association between AG level and the outcome during hospitalization in SICU. This study aimed to explore the predictive power of AG for 90-day all-cause mortality in SICU.

METHODS

Data of the eligible patients in SICU from 2008 to 2019 was obtained from the Medical Information Mart for Intensive Care IV version 2.0 (MIMIC-IV v2.0) database. Baseline clinical data of the selected patients was compared in different groups stratified by the outcome during their admission via univariate analysis. Restricted cubic spline (RCS) was drawn to confirm the relationship of AG and the short-term mortality. Kaplan-Meier survival curve was plotted in different AG level groups. Univariate and multivariate Cox analyses were performed, and Cox proportional-hazards models were built to investigate an independent role of AG to predict 90-day all-cause mortality risk in SICU. Receiver operating characteristics (ROC) curves analysis was performed to evaluate the predictive value of AG on the 90-day prognosis of patients.

RESULTS

A total of 6,395 patients were enrolled in this study and the 90-day all-cause mortality rate was 18.17%. Univariate analysis showed that elevated serum AG was associated with higher mortality (P < 0.001). RCS analysis indicated a positively linear relationship between serum AG and the risk of 90-day all-cause mortality in SICU (χ2 = 4.730, P = 0.193). Kaplan-Meier survival analysis demonstrated that low-AG group (with a cutoff value of 14.10 mmol/L) had a significantly higher cumulative survival rate than the counterpart of high-AG group (χ2 = 96.370, P < 0.001). Cox proportional-hazards models were constructed and confirmed the independent predictive role of AG in 90-day all-cause mortality risk in SICU after adjusting for 23 confounding factors gradually (HR 1.423, 1.246-1.625, P < 0.001). In the further subgroup analyses, a significant interaction was confirmed between AG and sepsis as well as surgery on the risk for the 90-day mortality. The ROC curve showed that the optimal cut-off value of AG for predicting 90-day mortality was 14.89 with sensitivity of 60.7% and specificity of 54.8%. The area under curve (AUC) was 0.602. When combined with SOFA score, the AUC of AG for predicting 90-day prognosis was 0.710, with a sensitivity and specificity of 70% and 62.5% respectively.

CONCLUSIONS

Elevated AG (≥ 14.10 mmol/L) is an independent risk factor for predicting severe conditions and poor prognosis of critical ill surgical patients.

Improving Mortality Risk Prediction with Routine Clinical Data: A Practical Machine Learning Model Based on eICU Patients

Purpose

Mortality risk prediction helps clinicians make better decisions in patient healthcare. However, existing severity scoring systems or algorithms used in intensive care units (ICUs) often rely on laborious manual collection of complex variables and lack sufficient validation in diverse clinical environments, thus limiting their practical applicability. This study aims to evaluate the performance of machine learning models that utilize routinely collected clinical data for short-term mortality risk prediction.

Patients and Methods

Using the eICU Collaborative Research Database, we identified a cohort of 12,393 ICU patients, who were randomly divided into a training group and a validation group at a ratio of 9:1. The models utilized routine variables obtained from regular medical workflows, including age, gender, physiological measurements, and usage of vasoactive medications within a 24-hour period prior to patient discharge. Four different machine learning algorithms, namely logistic regression, random forest, extreme gradient boosting (XGboost), and artificial neural network were employed to develop the mortality risk prediction model. We compared the discrimination and calibration performance of these models in assessing mortality risk within 1-week time window.

Results

Among the tested models, the XGBoost algorithm demonstrated the highest performance, with an area under the receiver operating characteristic curve (AUROC) of 0.9702, an area under precision and recall curves (AUPRC) of 0.8517, and a favorable Brier score of 0.0259 for 24-hour mortality risk prediction. Although the model's performance decreased when considering larger time windows, it still achieved a comparable AUROC of 0.9184 and AUPRC of 0.5519 for 3-day mortality risk prediction.

Conclusion

The findings demonstrate the feasibility of developing a highly accurate and well-calibrated model based on the XGBoost algorithm for short-term mortality risk prediction with easily accessible and interpretative data. These results enhance confidence in the application of the machine learning model to clinical practice.

Evaluation of medication regimen complexity as a predictor for mortality

While medication regimen complexity, as measured by a novel medication regimen complexity-intensive care unit (MRC-ICU) score, correlates with baseline severity of illness and mortality, whether the MRC-ICU improves hospital mortality prediction is not known. After characterizing the association between MRC-ICU, severity of illness and hospital mortality we sought to evaluate the incremental benefit of adding MRC-ICU to illness severity-based hospital mortality prediction models. This was a single-center, observational cohort study of adult intensive care units (ICUs). A random sample of 991 adults admitted ≥ 24 h to the ICU from 10/2015 to 10/2020 were included. The logistic regression models for the primary outcome of mortality were assessed via area under the receiver operating characteristic (AUROC). Medication regimen complexity was evaluated daily using the MRC-ICU. This previously validated index is a weighted summation of medications prescribed in the first 24 h of ICU stay [e.g., a patient prescribed insulin (1 point) and vancomycin (3 points) has a MRC-ICU = 4 points]. Baseline demographic features (e.g., age, sex, ICU type) were collected and severity of illness (based on worst values within the first 24 h of ICU admission) was characterized using both the Acute Physiology and Chronic Health Evaluation (APACHE II) and the Sequential Organ Failure Assessment (SOFA) score. Univariate analysis of 991 patients revealed every one-point increase in the average 24-h MRC-ICU score was associated with a 5% increase in hospital mortality [Odds Ratio (OR) 1.05, 95% confidence interval 1.02-1.08, p = 0.002]. The model including MRC-ICU, APACHE II and SOFA had a AUROC for mortality of 0.81 whereas the model including only APACHE-II and SOFA had a AUROC for mortality of 0.76. Medication regimen complexity is associated with increased hospital mortality. A prediction model including medication regimen complexity only modestly improves hospital mortality prediction.

Using Culture Sensitivity Reports to Optimize Antimicrobial Therapy: Findings and Implications of Antimicrobial Stewardship Activity in a Hospital in Pakistan

Background: There are concerns with inappropriate prescribing of antibiotics in hospitals especially broad spectrum in Pakistan and the subsequent impact on antimicrobial resistance rates. One recognized way to reduce inappropriate prescribing is for empiric therapy to be adjusted according to the result of culture sensitivity reports. Objective: Using culture sensitivity reports to optimize antibiotic prescribing in a teaching hospital in Pakistan. Methods: A retrospective observational study was undertaken in Ghurki Trust Teaching Hospital. A total of 465 positive cultures were taken from patients during the study period (May 2018 and December 2018). The results of pathogen identification and susceptibility testing from patient-infected sites were assessed. Additional data was collected from the patient's medical file. This included demographic data, sample type, causative microbe, antimicrobial treatment, and whether empiric or definitive treatment as well as medicine costs. Antimicrobial data was assessed using World Health Organization's Defined Daily Dose methodology. Results: A total of 497 isolates were detected from the 465 patient samples as 32 patients had polymicrobes, which included 309 g-negative rods and 188 g-positive cocci. Out of 497 isolates, the most common Gram-positive pathogen isolated was Staphylococcus aureus (Methicillin-sensitive Staphylococcus aureus) (125) (25.1%) and the most common Gram-negative pathogen was Escherichia coli (140) (28.1%). Most of the gram-negative isolates were found to be resistant to ampicillin and co-amoxiclav. Most of the Acinetobacter baumannii isolates were resistant to carbapenems. Gram-positive bacteria showed the maximum sensitivity to linezolid and vancomycin. The most widely used antibiotics for empiric therapy were cefoperazone plus sulbactam, ceftriaxone, amikacin, vancomycin, and metronidazole whereas high use of linezolid, clindamycin, meropenem, and piperacillin + tazobactam was seen in definitive treatment. Empiric therapy was adjusted in 220 (71.1%) cases of Gram-negative infections and 134 (71.2%) cases of Gram-positive infections. Compared with empiric therapy, there was a 13.8% reduction in the number of antibiotics in definitive treatment. The average cost of antibiotics in definitive treatment was less than seen with empiric treatment (8.2%) and the length of hospitalization also decreased. Conclusions: Culture sensitivity reports helped reduced antibiotic utilization and costs as well as helped select the most appropriate treatment. We also found an urgent need for implementing antimicrobial stewardship programs in hospitals and the development of hospital antibiotic guidelines to reduce unnecessary prescribing of broad-spectrum antibiotics.

Predicting survival and neurological outcome in out-of-hospital cardiac arrest using machine learning: the SCARS model

BACKGROUND

A prediction model that estimates survival and neurological outcome in out-of-hospital cardiac arrest patients has the potential to improve clinical management in emergency rooms.

METHODS

We used the Swedish Registry for Cardiopulmonary Resuscitation to study all out-of-hospital cardiac arrest (OHCA) cases in Sweden from 2010 to 2020. We had 393 candidate predictors describing the circumstances at cardiac arrest, critical time intervals, patient demographics, initial presentation, spatiotemporal data, socioeconomic status, medications, and comorbidities before arrest. To develop, evaluate and test an array of prediction models, we created stratified (on the outcome measure) random samples of our study population. We created a training set (60% of data), evaluation set (20% of data), and test set (20% of data). We assessed the 30-day survival and cerebral performance category (CPC) score at discharge using several machine learning frameworks with hyperparameter tuning. Parsimonious models with the top 1 to 20 strongest predictors were tested. We calibrated the decision threshold to assess the cut-off yielding 95% sensitivity for survival. The final model was deployed as a web application.

FINDINGS

We included 55,615 cases of OHCA. Initial presentation, prehospital interventions, and critical time intervals variables were the most important. At a sensitivity of 95%, specificity was 89%, positive predictive value 52%, and negative predictive value 99% in test data to predict 30-day survival. The area under the receiver characteristic curve was 0.97 in test data using all 393 predictors or only the ten most important predictors. The final model showed excellent calibration. The web application allowed for near-instantaneous survival calculations.

INTERPRETATION

Thirty-day survival and neurological outcome in OHCA can rapidly and reliably be estimated during ongoing cardiopulmonary resuscitation in the emergency room using a machine learning model incorporating widely available variables.

FUNDING

Swedish Research Council (2019-02019); Swedish state under the agreement between the Swedish government, and the county councils (ALFGBG-971482); The Wallenberg Centre for Molecular and Translational Medicine.

Prediction of in-hospital mortality among intensive care unit patients using modified daily Laboratory-based Acute Physiology Scores, version 2 (LAPS2)

Background

Mortality prediction for intensive care unit (ICU) patients frequently relies on single acuity measures based on ICU admission physiology without accounting for subsequent clinical changes.

Objectives

Evaluate novel models incorporating modified admission and daily, time-updating Laboratory-based Acute Physiology Scores, version 2 (LAPS2) to predict in-hospital mortality among ICU patients.

Research design

Retrospective cohort study.

Subjects

All ICU patients in five hospitals from October 2017 through September 2019.

Measures

We used logistic regression, penalized logistic regression, and random forest models to predict in-hospital mortality within 30 days of ICU admission using admission LAPS2 alone in patient-level and patient-day-level models, or admission and daily LAPS2 at the patient-day level. Multivariable models included patient and admission characteristics. We performed internal-external validation using four hospitals for training and the fifth for validation, repeating analyses for each hospital as the validation set. We assessed performance using scaled Brier scores (SBS), c-statistics, and calibration plots.

Results

The cohort included 13,993 patients and 120,101 ICU days. The patient-level model including the modified admission LAPS2 without daily LAPS2 had an SBS of 0.175 (95% CI 0.148-0.201) and c-statistic of 0.824 (95% CI 0.808-0.840). Patient-day-level models including daily LAPS2 consistently outperformed models with modified admission LAPS2 alone. Among patients with <50% predicted mortality, daily models were better calibrated than models with modified admission LAPS2 alone.

Conclusions

Models incorporating daily, time-updating LAPS2 to predict mortality among an ICU population perform as well or better than models incorporating modified admission LAPS2 alone.

Sex-specific treatment characteristics and 30-day mortality outcomes of critically ill COVID-19 patients over 70 years of age-results from the prospective COVIP study

PURPOSE

Older critically ill patients with COVID-19 have been the most vulnerable during the ongoing pandemic, with men being more prone to hospitalization and severe disease than women. We aimed to explore sex-specific differences in treatment and outcome after intensive care unit (ICU) admission in this cohort.

METHODS

We performed a sex-specific analysis in critically ill patients ≥ 70 yr of age with COVID-19 who were included in the international prospective multicenter COVIP study. All patients were analyzed for ICU admission and treatment characteristics. We performed a multilevel adjusted regression analysis to elucidate associations of sex with 30-day mortality.

RESULTS

A total of 3,159 patients (69.8% male, 30.2% female; median age, 75 yr) were included. Male patients were significantly fitter than female patients as determined by the Clinical Frailty Scale (fit, 67% vs 54%; vulnerable, 14% vs 19%; frail, 19% vs 27%; P < 0.001). Male patients more often underwent tracheostomy (20% vs 14%; odds ratio [OR], 1.57; P < 0.001), vasopressor therapy (69% vs 62%; OR, 1.25; P = 0.02), and renal replacement therapy (17% vs 11%; OR, 1.96; P < 0.001). There was no difference in mechanical ventilation, life-sustaining treatment limitations, and crude 30-day mortality (50% male vs 49% female; OR, 1.11; P = 0.19), which remained true after adjustment for disease severity, frailty, age and treatment limitations (OR, 1.17; 95% confidence interval, 0.94 to 1.45; P = 0.16).

CONCLUSION

In this analysis of sex-specific treatment characteristics and 30-day mortality outcomes of critically ill patients with COVID-19 ≥ 70 yr of age, we found more tracheostomy and renal replacement therapy in male vs female patients, but no significant association of patient sex with 30-day mortality.

STUDY REGISTRATION

www.

CLINICALTRIALS

gov (NCT04321265); registered 25 March 2020).

Comparison of predictive scoring systems in patients hospitalized in the internal medicine intensive care unit

Aim: Various scoring systems have been developed to predict mortality, disease severity, and length of stay of patients in intensive care units. It is important to demonstrate the validity of these scores in the society in which they are used. This study aims to evaluate the effects of The Acute Physiologic and Chronic Evaluation (APACHE)-II, APACHE-IV, The Simplified Acute Physiologic Score (SAPS), and Mortality Prediction Model (MPM0) scores on mortality in the internal medicine intensive care unit. Material and Method: The patients who were followed up in an internal medicine intensive care unit between June 2021 and December 2021 in a tertiary hospital in Turkey were included in this study. The scores were calculated at the time they were admitted to the intensive care unit. 115 patients who were followed up in the internal medicine intensive care unit for 6 months were included. The patients were divided into two groups alive or deceased. 52 (45.2%) patients in the survivor group and 63 (54.8%) patients in the deceased group were included. Patients received no study-related medical intervention. Results: When all four prognostic scoring systems were analyzed according to the median cut-off values, rising values ​​were related to mortality with statistical significance (p

Predicting in-hospital mortality for MIMIC-III patients: A nomogram combined with SOFA score

Predicting the mortality of patients provides a reference for doctors to judge their physical condition. This study aimed to construct a nomogram to improve the prediction accuracy of patients' mortality. Patients with severe diseases were screened from the Medical Information Mart for Intensive Care (MIMIC) III database; 70% of patients were randomly selected as the training set for the model establishment, while 30% were used as the test set. The least absolute shrinkage and selection operator (LASSO) regression method was used to filter variables and select predictors. A multivariable logistic regression fit was used to determine the association between in-hospital mortality and risk factors and to construct a nomogram. A total of 9276 patients were included. The area under the curve (AUC) for the clinical nomogram based on risk factors selected by LASSO and multivariable logistic regressions were 0.849 (95% confidence interval [CI]: 0.835-0.863) and 0.821 (95% CI: 0.795-0.846) in the training and test sets, respectively. Therefore, this nomogram might help predict the in-hospital mortality of patients admitted to the intensive care unit (ICU).

Outcome of early cholecystectomy compared to percutaneous drainage of gallbladder and delayed cholecystectomy for patients with acute cholecystitis: systematic review and meta-analysis

BACKGROUND

Compare outcomes of early laparoscopic cholecystectomy (ELC) and percutaneous trans-hepatic drainage of gallbladder (PTGBD) as an initial intervention for AC and to compare operative outcomes of ELC and delayed laparoscopic cholecystectomy (DLC).

METHODS

English-language studies published until December 2020 were searched. Randomised controlled trials (RCTs) and observational studies compared EC and PTGBD with delayed cholecystectomy for patients presented with acute cholecystitis were considered. Main outcomes were mortality, conversion to open, complications and length of hospital stay.

RESULTS

Out of 1347 records, 14 studies were included. 205,361 (94.7%) patients had EC and 11,565 (5.3%) patients had PTGBD as an initial intervention for AC. Mortality was higher in PTGBD; HR, 95% CI: [3.68 (2.13, 6.38)]. In contrast, complication rate was significantly higher in EC group (47%) vs PTGBD group (8.7%) in patients admitted to ICU; P-value = 0.011. Patients who had ELC were at higher risk of post-operative complications compared to DLC; RR [95% CI]: 2.88 [1.78, 4.65]. Risk of bile duct injury was six folds more in ELC; RR [95% CI]: 6.07 [1.67, 21.99].

CONCLUSION

ELC may be a preferred treatment option over PTGBD in AC. However, patient and disease specific factors should be considered to avoid unfavourable outcomes with ELC.

Evaluation of the correlation of serum calcium, phosphorus levels and calcium phosphorus product with disease severity and ICU mortality in SARS-COV-2 pneumonia patients followed up in ICU

Background: Calcium and phosphorus are important elements in the body that have been shown to decrease in critical inflammatory diseases. The aim of this study was to evaluate serum levels of calcium and phosphorus and the calcium phosphate product (CPP) in patients followed up in intensive care unit (ICU) due to hypoxemic respiratory failure caused by coronavirus disease 2019 (COVID-19) pneumonia. The secondary endpoint of the study were respiratory support therapies used in the evaluation of independent mortality and disease severity in ICU that were divided into four groups depending on the time of administration: (i) first 24 hours, (ii) 48-72 hours, (iii) 72 hours, and (iv) 72 hours-28 days. Material and Method: The retrospective study included patients with critical and severe COVID-19 pneumonia followed up in ICU. Results: The study included 369 patients with a mean age of 64.3±14.8 years. ICU mortality was observed in 142 (38.5%) patients, among whom 17 (4.6%) patients died within 24 hours, 28 (7.6%) died between 48-72 hours, 50 (12.7%) died within 72 hours, and 47 (12.7%) died between 72 hours and 28 days. Serum calcium level established a significant relationship with ICU mortality at 28 days and 72 hours (p0.05). Conclusion: Serial assessment of serum calcium may be a new criterion in the prediction of independent mortality in critical and severe COVID-19 pneumonia, which has been recently identified and has numerous unknown features.

The Global Open Source Severity of Illness Score (GOSSIS)

OBJECTIVES

To develop and demonstrate the feasibility of a Global Open Source Severity of Illness Score (GOSSIS)-1 for critical care patients, which generalizes across healthcare systems and countries.

DESIGN

A merger of several critical care multicenter cohorts derived from registry and electronic health record data. Data were split into training (70%) and test (30%) sets, using each set exclusively for development and evaluation, respectively. Missing data were imputed when not available.

SETTING/PATIENTS

Two large multicenter datasets from Australia and New Zealand (Australian and New Zealand Intensive Care Society Adult Patient Database [ANZICS-APD]) and the United States (eICU Collaborative Research Database [eICU-CRD]) representing 249,229 and 131,051 patients, respectively. ANZICS-APD and eICU-CRD contributed data from 162 and 204 hospitals, respectively. The cohort included all ICU admissions discharged in 2014-2015, excluding patients less than 16 years old, admissions less than 6 hours, and those with a previous ICU stay.

INTERVENTIONS

Not applicable.

MEASUREMENTS AND MAIN RESULTS

GOSSIS-1 uses data collected during the ICU stay's first 24 hours, including extrema values for vital signs and laboratory results, admission diagnosis, the Glasgow Coma Scale, chronic comorbidities, and admission/demographic variables. The datasets showed significant variation in admission-related variables, case-mix, and average physiologic state. Despite this heterogeneity, test set discrimination of GOSSIS-1 was high (area under the receiver operator characteristic curve [AUROC], 0.918; 95% CI, 0.915-0.921) and calibration was excellent (standardized mortality ratio [SMR], 0.986; 95% CI, 0.966-1.005; Brier score, 0.050). Performance was held within ANZICS-APD (AUROC, 0.925; SMR, 0.982; Brier score, 0.047) and eICU-CRD (AUROC, 0.904; SMR, 0.992; Brier score, 0.055). Compared with GOSSIS-1, Acute Physiology and Chronic Health Evaluation (APACHE)-IIIj (ANZICS-APD) and APACHE-IVa (eICU-CRD), had worse discrimination with AUROCs of 0.904 and 0.869, and poorer calibration with SMRs of 0.594 and 0.770, and Brier scores of 0.059 and 0.063, respectively.

CONCLUSIONS

GOSSIS-1 is a modern, free, open-source inhospital mortality prediction algorithm for critical care patients, achieving excellent discrimination and calibration across three countries.

Sequential organ failure assessment score improves survival prediction for left ventricular assist device recipients in intensive care

BACKGROUND

Preoperative risk scores facilitate patient selection, but postoperative risk scores may offer valuable information for predicting outcomes. We hypothesized that the postoperative Sequential Organ Failure Assessment (SOFA) score would predict mortality after left ventricular assist device (LVAD) implantation.

METHODS

We retrospectively reviewed data from 294 continuous-flow LVAD implantations performed at Mayo Clinic Rochester during 2007 to 2015. We calculated the EuroSCORE, HeartMate-II Risk Score, and RV Failure Risk Score from preoperative data and the APACHE III and Post Cardiac Surgery (POCAS) risk scores from postoperative data. Daily, maximum, and mean SOFA scores were calculated for the first 5 postoperative days. The area under receiver-operator characteristic curves (AUC) was calculated to compare the scoring systems' ability to predict 30-day, 90-day, and 1-year mortality.

RESULTS

For the entire cohort, mortality was 5% at 30 days, 10% at 90 days, and 19% at 1 year. The Day 1 SOFA score had better discrimination for 30-day mortality (AUC 0.77) than the preoperative risk scores or the APACHE III and POCAS postoperative scores. The maximum SOFA score had the best discrimination for 30-day mortality (AUC 0.86), and the mean SOFA score had the best discrimination for 90-day mortality (AUC 0.82) and 1-year mortality (AUC 0.76).

CONCLUSIONS

We observed that postoperative mean and maximum SOFA scores in LVAD recipients predict short-term and intermediate-term mortality better than preoperative risk scores do. However, because preoperative and postoperative risk scores each contribute unique information, they are best used in concert to predict outcomes after LVAD implantation.

Severity of illness and organ dysfunction scoring systems in pediatric critical care: The impacts on clinician's practices and the future

Severity and organ dysfunction (OD) scores are increasingly used in pediatric intensive care units (PICU). Therefore, this review aims to provide 1/ an updated state-of-the-art of severity scoring systems and OD scores in pediatric critical care, which explains 2/ the performance measurement tools and the significance of each tool in clinical practice and provides 3/ the usefulness, limits, and impact on future scores in PICU. The following two pediatric systems have been proposed: the PRISMIV, is used to collect data between 2 h before PICU admission and the first 4 h after PICU admission; the PIM3, is used to collect data during the first hour after PICU admission. The PELOD-2 and SOFApediatric scores were the most common OD scores available. Scores used in the PICU should help clinicians answer the following three questions: 1/ Are the most severely ill patients dying in my service: a good discrimination allow us to interpret that there are the most severe patients who died in my service. 2/ Does the overall number of deaths observed in my department consistent with the severity of patients? The standard mortality ratio allow us to determine whether the total number of deaths observed in our service over a given period is in adequacy with the number of deaths predicted, by considering the severity of patients on admission? 3/ Does the number of deaths observed by severity level in my department consistent with the severity of patients? The calibration enabled us to determine whether the number of deaths observed according to the severity of patients at PICU admission in a department over a given period is in adequacy with the number of deaths predicted, according to the severity of the patients at PICU admission. These scoring systems are not interpretable at the patient level. Scoring systems are used to describe patients with PICU in research and evaluate the service's case mix and performance. Therefore, the prospect of automated data collection, which permits their calculation, facilitated by the computerization of services, is a necessity that manufacturers should consider.

Multicenter validation of PIM3 and PIM2 in Brazilian pediatric intensive care units

OBJECTIVE

To validate the PIM3 score in Brazilian PICUs and compare its performance with the PIM2.

METHODS

Observational, retrospective, multicenter study, including patients younger than 16 years old admitted consecutively from October 2013 to September 2019. We assessed the Standardized Mortality Ratio (SMR), the discrimination capability (using the area under the receiver operating characteristic curve - AUROC), and the calibration. To assess the calibration, we used the calibration belt, which is a curve that represents the correlation of predicted and observed values and their 95% Confidence Interval (CI) through all the risk ranges. We also analyzed the performance of both scores in three periods: 2013-2015, 2015-2017, and 2017-2019.

RESULTS

41,541 patients from 22 PICUs were included. Most patients aged less than 24 months (58.4%) and were admitted for medical conditions (88.6%) (respiratory conditions = 53.8%). Invasive mechanical ventilation was used in 5.8%. The median PICU length of stay was three days (IQR, 2-5), and the observed mortality was 1.8% (763 deaths). The predicted mortality by PIM3 was 1.8% (SMR 1.00; 95% CI 0.94-1.08) and by PIM2 was 2.1% (SMR 0.90; 95% CI 0.83-0.96). Both scores had good discrimination (PIM3 AUROC = 0.88 and PIM2 AUROC = 0.89). In calibration analysis, both scores overestimated mortality in the 0%-3% risk range, PIM3 tended to underestimate mortality in medium-risk patients (9%-46% risk range), and PIM2 also overestimated mortality in high-risk patients (70%-100% mortality risk).

CONCLUSIONS

Both scores had a good discrimination ability but poor calibration in different ranges, which deteriorated over time in the population studied.

Establishment and Effectiveness Evaluation of a Scoring System-RAAS (RDW, AGE, APACHE II, SOFA) for Sepsis by a Retrospective Analysis

Background

Methods

Results

Conclusion

Understanding the complexity of sepsis mortality prediction via rule discovery and analysis: a pilot study

BACKGROUND

Sepsis, defined as life-threatening organ dysfunction caused by a dysregulated host response to infection, has become one of the major causes of death in Intensive Care Units (ICUs). The heterogeneity and complexity of this syndrome lead to the absence of golden standards for its diagnosis, treatment, and prognosis. The early prediction of in-hospital mortality for sepsis patients is not only meaningful to medical decision making, but more importantly, relates to the well-being of patients.

METHODS

In this paper, a rule discovery and analysis (rule-based) method is used to predict the in-hospital death events of 2021 ICU patients diagnosed with sepsis using the MIMIC-III database. The method mainly includes two phases: rule discovery phase and rule analysis phase. In the rule discovery phase, the RuleFit method is employed to mine multiple hidden rules which are capable to predict individual in-hospital death events. In the rule analysis phase, survival analysis and decomposition analysis are carried out to test and justify the risk prediction ability of these rules. Then by leveraging a subset of these rules, we establish a prediction model that is both more accurate at the in-hospital death prediction task and more interpretable than most comparable methods.

RESULTS

In our experiment, RuleFit generates 77 risk prediction rules, and the average area under the curve (AUC) of the prediction model based on 62 of these rules reaches 0.781 ([Formula: see text]) which is comparable to or even better than the AUC of existing methods (i.e., commonly used medical scoring system and benchmark machine learning models). External validation of the prediction power of these 62 rules on another 1468 sepsis patients not included in MIMIC-III in ICU provides further supporting evidence for the superiority of the rule-based method. In addition, we discuss and explain in detail the rules with better risk prediction ability. Glasgow Coma Scale (GCS), serum potassium, and serum bilirubin are found to be the most important risk factors for predicting patient death.

CONCLUSION

Our study demonstrates that, with the rule-based method, we could not only make accurate prediction on in-hospital death events of sepsis patients, but also reveal the complex relationship between sepsis-related risk factors through the rules themselves, so as to improve our understanding of the complexity of sepsis as well as its population.

Performance in mortality prediction of SAPS 3 And MPM-III scores among adult patients admitted to the ICU of a private tertiary referral hospital in Tanzania: a retrospective cohort study

Background

Illness predictive scoring systems are significant and meaningful adjuncts of patient management in the Intensive Care Unit (ICU). They assist in predicting patient outcomes, improve clinical decision making and provide insight into the effectiveness of care and management of patients while optimizing the use of hospital resources. We evaluated mortality predictive performance of Simplified Acute Physiology Score (SAPS 3) and Mortality Probability Models (MPM₀-III) and compared their performance in predicting outcome as well as identifying disease pattern and factors associated with increased mortality.

Methods

This was a retrospective cohort study of adult patients admitted to the ICU of the Aga Khan Hospital, Dar- es- Salaam, Tanzania between August 2018 and April 2020. Demographics, clinical characteristics, outcomes, source of admission, primary admission category, length of stay and the support provided with the worst physiological data within the first hour of ICU admission were extracted. SAPS 3 and MPM₀-III scores were calculated using an online web-based calculator. The performance of each model was assessed by discrimination and calibration. Discrimination between survivors and non-survivors was assessed by the area under the receiver operator characteristic curve (ROC) and calibration was estimated using the Hosmer-Lemeshow goodness-of-fit test.

Results

A total of 331 patients were enrolled in the study with a median age of 58 years (IQR 43-71), most of whom were male (n = 208, 62.8%), of African origin (n = 178, 53.8%) and admitted from the emergency department (n = 306, 92.4%). In- hospital mortality of critically ill patients was 16.1%. Discrimination was very good for all models, the area under the receiver-operating characteristic (ROC) curve for SAPS 3 and MPM₀-III was 0.89 (95% CI [0.844-0.935]) and 0.90 (95% CI [0.864-0.944]) respectively. Calibration as calculated by Hosmer-Lemeshow goodness-of-fit test showed good calibration for SAPS 3 and MPM₀-III with Chi- square values of 4.61 and 5.08 respectively and P-Value > 0.05.

Conclusion

Both SAPS 3 and MPM₀-III performed well in predicting mortality and outcome in our cohort of patients admitted to the intensive care unit of a private tertiary hospital. The in-hospital mortality of critically ill patients was lower compared to studies done in other intensive care units in tertiary referral hospitals within Tanzania.

Brescia-COVID Respiratory Severity Scale (BRCSS) and Quick SOFA (qSOFA) score are most useful in showing severity in COVID-19 patients

In this study, we compare the predictive value of clinical scoring systems that are already in use in patients with Coronavirus disease 2019 (COVID-19), including the Brescia-COVID Respiratory Severity Scale (BCRSS), Quick SOFA (qSOFA), Sequential Organ Failure Assessment (SOFA), Multilobular infiltration, hypo-Lymphocytosis, Bacterial coinfection, Smoking history, hyper-Tension, and Age (MuLBSTA) and scoring system for reactive hemophagocytic syndrome (HScore), for determining the severity of the disease. Our aim in this study is to determine which scoring system is most useful in determining disease severity and to guide clinicians. We classified the patients into two groups according to the stage of the disease (severe and non-severe) and adopted interim guidance of the World Health Organization. Severe cases were divided into a group of surviving patients and a deceased group according to the prognosis. According to admission values, the BCRSS, qSOFA, SOFA, MuLBSTA, and HScore were evaluated at admission using the worst parameters available in the first 24 h. Of the 417 patients included in our study, 46 (11%) were in the severe group, while 371 (89%) were in the non-severe group. Of these 417 patients, 230 (55.2%) were men. The median (IQR) age of all patients was 44 (25) years. In multivariate logistic regression analyses, BRCSS in the highest tertile (HR 6.1, 95% CI 2.105–17.674, p = 0.001) was determined as an independent predictor of severe disease in cases of COVID-19. In multivariate analyses, qSOFA was also found to be an independent predictor of severe COVID-19 (HR 4.757, 95% CI 1.438–15.730, p = 0.011). The area under the curve (AUC) of the BRCSS, qSOFA, SOFA, MuLBSTA, and HScore was 0.977, 0.961, 0.958, 0.860, and 0.698, respectively. Calculation of the BRCSS and qSOFA at the time of hospital admission can predict critical clinical outcomes in patients with COVID-19, and their predictive value is superior to that of HScore, MuLBSTA, and SOFA. Our prediction is that early interventions for high-risk patients, with early identification of high-risk group using BRCSS and qSOFA, may improve clinical outcomes in COVID-19.

Machine learning based early mortality prediction in the emergency department

BACKGROUND

It is a great challenge for emergency physicians to early detect the patient's deterioration and prevent unexpected death through a large amount of clinical data, which requires sufficient experience and keen insight.

OBJECTIVE

To evaluate the performance of machine learning models in quantifying the severity of emergency department (ED) patients and identifying high-risk patients.

METHODS

Using routinely-available demographics, vital signs and laboratory tests extracted from electronic health records (EHRs), a framework based on machine learning and feature engineering was proposed for mortality prediction. Patients who had one complete record of vital signs and laboratory tests in ED were included. The following patients were excluded: pediatric patients aged < 18 years, pregnant woman, and patients died or were discharged or hospitalized within 12 h after admission. Based on 76 original features extracted, 9 machine learning models were adopted to validate our proposed framework. Their optimal hyper-parameters were fine-tuned using the grid search method. The prediction results were evaluated on performance metrics (i.e., accuracy, area under the curve (AUC), recall and precision) with repeated 5-fold cross-validation (CV). The time window from patient admission to the prediction was analyzed at 12 h, 24 h, 48 h, and entire stay.

RESULTS

We studied a total of 1114 ED patients with 71.54% (797/1114) survival and 28.46% (317/1114) death in the hospital. The results revealed a more complete time window leads to better prediction performance. Using the entire stay records, the LightGBM model with refined feature engineering demonstrated high discrimination and achieved 93.6% (±0.008) accuracy, 97.6% (±0.003) AUC, 97.1% (±0.008) recall, and 94.2% (±0.006) precision, even if no diagnostic information was utilized.

CONCLUSIONS

This study quantifies the criticality of ED patients and appears to have significant potential as a clinical decision support tool in assisting physicians in their clinical routine. While the model requires validation before use elsewhere, the same methodology could be used to create a strong model for the new hospital.

Evaluation of severity scoring systems in patients with severe community acquired pneumonia

BACKGROUND

The aim of this study was to evaluate the ability of severity scoring systems to predict 30-day mortality in patients with severe community-acquired pneumonia.

METHODS

The study included 98 patients aged ≥18 years with community acquired pneumonia hospitalized at the Intensive Care Unit of the University Clinic for Infectious Diseases in Skopje, Republic of North Macedonia, during a 3-year period. We recorded demographic, clinical and common biochemical parameters. Five severity scores were calculated at admission: CURB 65 (Confusion, Urea, Respiratory Rate, Blood pressure, Age ≥65 years), SCAP (Severe Community Acquired Pneumonia score), SAPS II (Simplified Acute Physiology Score), SOFA (Sequential Organ Failure Assessment Score) and MPM (Mortality Prediction Model). Primary outcome variable was 30-day in-hospital mortality.

RESULTS

The mean age of the patients was 59.08 ± 15.76 years, predominantly males (68%). The overall 30-day mortality was 52%. Charlson Comorbidity index was increased in non-survivors (3.72 ± 2.33) and was associated with the outcome. All severity indexes had higher values in patients who died, that showed statistical significance between the analysed groups. The areas under curve (AUC) values of the five scores for 30-day mortality were 0.670, 0.732, 0,726, 0.785 and 0.777, respectively.

CONCLUSION

Widely used severity scores accurately detected patients with pneumonia that had increased risk for poor outcome, but none of them individually demonstrated any advantage over the others.

Prospective validation of an 11-gene mRNA host response score for mortality risk stratification in the intensive care unit

Several clinical calculators predict intensive care unit (ICU) mortality, however these are cumbersome and often require 24 h of data to calculate. Retrospective studies have demonstrated the utility of whole blood transcriptomic analysis in predicting mortality. In this study, we tested prospective validation of an 11-gene messenger RNA (mRNA) score in an ICU population. Whole blood mRNA from 70 subjects in the Stanford ICU Biobank with samples collected within 24 h of Emergency Department presentation were used to calculate an 11-gene mRNA score. We found that the 11-gene score was highly associated with 60-day mortality, with an area under the receiver operating characteristic curve of 0.68 in all patients, 0.77 in shock patients, and 0.98 in patients whose primary determinant of prognosis was acute illness. Subjects with the highest quartile of mRNA scores were more likely to die in hospital (40% vs 7%, p < 0.01) and within 60 days (40% vs 15%, p = 0.06). The 11-gene score improved prognostication with a categorical Net Reclassification Improvement index of 0.37 (p = 0.03) and an Integrated Discrimination Improvement index of 0.07 (p = 0.02) when combined with Simplified Acute Physiology Score 3 or Acute Physiology and Chronic Health Evaluation II score. The test performed poorly in the 95 independent samples collected > 24 h after emergency department presentation. Tests will target a 30-min turnaround time, allowing for rapid results early in admission. Moving forward, this test may provide valuable real-time prognostic information to improve triage decisions and allow for enrichment of clinical trials.

Designing a bed-side system for predicting length of stay in a neonatal intensive care unit

Increased length of stay (LOS) in intensive care units is directly associated with the financial burden, anxiety, and increased mortality risks. In the current study, we have incorporated the association of day-to-day nutrition and medication data of the patient during its stay in hospital with its predicted LOS. To demonstrate the same, we developed a model to predict the LOS using risk factors (a) perinatal and antenatal details, (b) deviation of nutrition and medication dosage from guidelines, and (c) clinical diagnoses encountered during NICU stay. Data of 836 patient records (12 months) from two NICU sites were used and validated on 211 patient records (4 months). A bedside user interface integrated with EMR has been designed to display the model performance results on the validation dataset. The study shows that each gestation age group of patients has unique and independent risk factors associated with the LOS. The gestation is a significant risk factor for neonates < 34 weeks, nutrition deviation for < 32 weeks, and clinical diagnosis (sepsis) for ≥ 32 weeks. Patients on medications had considerable extra LOS for ≥ 32 weeks’ gestation. The presented LOS model is tailored for each patient, and deviations from the recommended nutrition and medication guidelines were significantly associated with the predicted LOS.

Comparing CLIF-C ACLF, CLIF-C ACLFlactate, and CLIF-C ACLF-D Prognostic Scores in Acute-on-Chronic Liver Failure Patients by a Single-Center ICU Experience

Patients with liver cirrhosis have a higher risk of developing acute-on-chronic liver failure (ACLF). Poor prognosis with a high rate of short-term mortality leads to limited opportunities for further liver transplantation. Thus, precise prognostic evaluation of patients with ACLF is necessary before transplant surgery. In this study, a total of one hundred and thirty-five patients with ACLF admitted to the hepato-gastroenterologic intensive care unit (ICU) for intensive monitoring and treatment at Chang-Gung Memorial Hospital (CGMH, Linkou, Taiwan) were screened from November 2012 to April 2015 and tracked until April 2017. Three new prognostic scores of ACLF, including CLIF-C ACLF (Chronic Liver Failure Consortium Acute-on-chronic Liver Failure score), CLIF-C ACLF-D (CLIF-C ACLF Development score), and CLLF-C ACLF_lactate (lactate-adjusted CLIF-C ACLF score) were compared. The primary outcome considered was overall mortality. Mortality predictions at 28, 90, 180, and 365 days were also calculated. By area under the receiver operating characteristic curve (AUROC) analysis, the CLIF-C ACLF and CLIF-C ACLF-D scores were superior to CLIF-C ACLF_lactate scores in predicting 28-day mortality. The CLIF-C ACLF-D score had the highest AUROC in predicting overall mortality as well as at 90, 180, and 365 days. In conclusion, our study demonstrates that CLIF-C ACLF and CLIF-C ACLF-D scores are significant predictors of outcome in critical patients with liver cirrhosis and ACLF. The CLIF-C ACLF-D score may have a superior predictive power for the prediction of 3-month, 6-month, and one-year mortality.

Comparison of SAPS 3 performance in patients with and without solid tumor admitted to an intensive care unit in Brazil: a retrospective cohort study

Objective

To compare the performance of the Simplified Acute Physiology Score 3 (SAPS 3) in patients with and without solid cancer who were admitted to the intensive care unit of a comprehensive oncological hospital in Brazil.

Methods

We performed a retrospective cohort analysis of our administrative database of the first admission of adult patients to the intensive care unit from 2012 to 2016. The patients were categorized according to the presence of solid cancer. We evaluated discrimination using the area under the Receiver Operating Characteristic curve (AUROC) and calibration using the calibration belt approach.

Results

We included 7,254 patients (41.5% had cancer, and 12.1% died during hospitalization). Oncological patients had higher hospital mortality than nononcological patients (14.1% versus 10.6%, respectively; p < 0.001). SAPS 3 discrimination was better for oncological patients (AUROC = 0.85) than for nononcological patients (AUROC = 0.79) (p < 0.001). After we applied the calibration belt in oncological patients, the SAPS 3 matched the average observed rates with a confidence level of 95%. In nononcological patients, the SAPS 3 overestimated mortality in those with a low-middle risk. Calibration was affected by the time period only for nononcological patients.

Conclusion

SAPS 3 performed differently between oncological and nononcological patients in our single-center cohort, and variation over time (mainly calibration) was observed. This finding should be taken into account when evaluating severity-of-illness score performance.

Prediction of Postoperative Mortality in Patients with Organ Failure After Gastric Cancer Surgery

Background

Scarce data are available on the characteristics of postoperative organ failure (POF) and mortality after gastrectomy. We aimed to describe the causes of organ failure and mortality related to gastrectomy for gastric cancer and to identify patients with POF who are at a risk of failure to rescue (FTR).

Methods

The study examined patients with POF or in-hospital mortality in Seoul National University Hospital between 2005 and 2014. We identified patients at a high risk of FTR by analyzing laboratory findings, complication data, intensive care unit records, and risk scoring including Acute Physiology and Chronic Health Evaluation (APACHE) IV, Sequential Organ Failure Assessment (SOFA) score, and Simplified Acute Physiology Score (SAPS) 3 at ICU admission.

Results

Among the 7304 patients who underwent gastrectomy, 80 (1.1%) were identified with Clavien–Dindo classification (CDC) grade ≥ IVa. The numbers of patients with CDC grade IVa, IVb, and V were 48 (0.66%), 11 (0.15%), and 21 (0.29%), respectively. Pulmonary failure (43.8%), surgical site complication (27.5%), and cardiac failure (13.8%) were the most common causes of POF and mortality. Cancer progression (100%) and cardiac events (45.5%) showed high FTR rates. In univariate analysis, acidosis, hypoalbuminemia, SOFA, APACHE IV, and SAPS 3 were identified as risk factors for FTR (P < 0.05). Finally, SAPS 3 was identified as an independent predictive factor for FTR.

Conclusions

Cancer progression and acute cardiac failure were the most lethal causes of FTR. SAPS 3 is an independent predictor of FTR among POF patients after gastrectomy.

The objective nursing workload and perceived nursing workload in Intensive Care Units: Analysis of association

BACKGROUND

A range of classification systems are in use for the measurement of nursing workload in Intensive Care Units. However, it is unknown to what extent the measured (objective) nursing workload, usually in terms of the amount of nursing activities, is related to the workload actually experienced (perceived) by nurses.

OBJECTIVES

The aim of this study was to assess the association between the objective nursing workload and the perceived nursing workload and to identify other factors associated with the perceived nursing workload.

METHODS

We measured the objective nursing workload with the Nursing Activities Score and the perceived nursing workload with the NASA-Task Load Index during 228 shifts in eight different Intensive Care Units. We used linear mixed-effect regression models to analyze the association between the objective and perceived nursing workload. Furthermore, we investigated the association of patient characteristics (severity of illness, comorbidities, age, body mass index, and planned or unplanned admission), education level of the nurse, and contextual factors (numbers of patients per nurse, the type of shift (day, evening, night) and day of admission or discharge) with perceived nursing workload. We adjusted for confounders.

RESULTS

We did not find a significant association between the observed workload per nurse and perceived nursing workload (p=0.06). The APACHE-IV Acute Physiology Score of a patient was significantly associated with the perceived nursing workload, also after adjustment for confounders (p=0.02). None of the other patient characteristics was significantly associated with perceived nursing workload. Being a certified nurse or a student nurse was the only nursing or contextual factor significantly associated with the perceived nursing workload, also after adjustment for confounders (p=0.03).

CONCLUSION

Workload is perceived differently by nurses compared to the objectively measured workload by the Nursing Activities Score. Both the severity of illness of the patient and being a student nurse are factors that increase the perceived nursing workload. To keep the workload of nurses in balance, planning nursing capacity should be based on the Nursing Activities Score, on the severity of patient illness and the graduation level of the nurse.

Triage and Prognostication of Cancer Patients Admitted to the Intensive Care Unit

Cancer remains a leading cause of morbidity and mortality. Advances in cancer screening, early detection, targeted therapies, and supportive care have led to improvements in outcomes and quality of life. The rapid increase in novel cancer therapies can cause life-threatening adverse events. The need for intensive care unit (ICU) care is projected to increase. Until 2 decades ago, cancer diagnosis often precluded ICU admission. Recently, substantial cancer survival has been achieved; therefore, ICU denial is not recommended. ICU resources are limited and expensive; hence, appropriate utilization is needed. This review focuses on triage and prognosis in critically ill cancer patients requiring ICU admission.

Overall and subgroup specific performance of the pediatric index of mortality 2 score in Switzerland: a national multicenter study

Pediatric Index of Mortality (PIM) 2 score is used in pediatric intensive care unit (PICU) to predict the patients' risk of death. The performance of this model has never been assessed in Switzerland. The aim of this study was to evaluate the performance of the PIM2 score in the whole cohort and in pre-specified diagnostic subgroups of patients admitted to PICUs in Switzerland. All children younger than 16 years admitted to any PICU in Switzerland between January 1, 2012 and December 31, 2017 were included in the study. A total of 22,382 patients were analyzed. Observed mortality was 2%, whereas mortality predicted by PIM2 was 4.2% (SMR = 0.47, 95% CI, 0.42-0.52). Calibration was also poor across the deciles of mortality risks (p < 0.001). The AUC-ROC for the entire cohort was 0.88 (95% CI, 0.87-0.90). Calibration varied significantly according to primary diagnosis.Conclusion: The performance of the PIM 2 score in a cohort of Swiss patients is poor with adequate discrimination and poor calibration. The PIM 2 score tends to under predict the number of deaths among septic patients and in patients admitted after a cardiorespiratory arrest. What is Known: •PIM2 score is a widely used mortality prediction model in PICU. •PIM2 performance among uncommon but clinically relevant diagnostic subgroups of patients is unknown. •The performance of PIM2 score has never been assessed in Switzerland. What is New: •The performance of the PIM 2 score in a cohort of Swiss patients is poor with adequate discrimination and poor calibration. •Calibration varies significantly according to primary diagnosis. The PIM 2 score under predict the number of deaths among septic patients and in patients admitted after a cardiorespiratory arrest.

Exploration of critical care data by using unsupervised machine learning

BACKGROUND AND OBJECTIVE

Identification of subgroups may be useful to understand the clinical characteristics of ICU patients. The purposes of this study were to apply an unsupervised machine learning method to ICU patient data to discover subgroups among them; and to examine their clinical characteristics, therapeutic procedures conducted during the ICU stay, and discharge dispositions.

METHODS

K-means clustering method was used with 1503 observations and 9 types of laboratory test results as features.

RESULTS

Three clusters were identified from this specific population. Blood urea nitrogen, creatinine, potassium, hemoglobin, and red blood cell were distinctive between the clusters. Cluster Three presented the highest blood products transfusion rate (19.8%), followed by Cluster One (15.5%) and cluster Two (9.3%), which was significantly different. Hemodialysis was more frequently provided to Cluster Three while bronchoscopy was done to Cluster One and Two. Cluster Three showed the highest mortality (30.4%), which was more than two-fold compared to Cluster One (14.1%) and Two (12.2%).

CONCLUSION

Three subgroups were identified and their clinical characteristics were compared. These findings may be useful to anticipate treatment strategies and probable outcomes of ICU patients. Unsupervised machine learning may enable ICU multi-dimensional data to be organized and to make sense of the data.

Serum Albumin Levels: Who Needs Them?

OBJECTIVES

The purpose of this critical narrative review is to discuss common indications for ordering serum albumin levels in adult critically ill patients, evaluate the literature supporting these indications, and provide recommendations for the appropriate ordering of serum albumin levels.

DATA SOURCES

PubMed (1966 to August 2020), Cochrane Library, and current clinical practice guidelines were used, and bibliographies of retrieved articles were searched for additional articles.

STUDY SELECTION AND DATA EXTRACTION

Current clinical practice guidelines were the preferred source of recommendations regarding serum albumin levels for guiding albumin administration and for nutritional monitoring. When current comprehensive reviews were available, they served as a baseline information with supplementation by subsequent studies.

DATA SYNTHESIS

Serum albumin is a general marker of severity of illness, and hypoalbuminemia is associated with poor patient outcome, but albumin is an acute phase protein, so levels vacillate in critically ill patients in conjunction with illness fluctuations. The most common reasons for ordering serum albumin levels in intensive care unit (ICU) settings are to guide albumin administration, to estimate free phenytoin or calcium levels, for nutritional monitoring, and for severity-of-illness assessment.

RELEVANCE TO PATIENT CARE AND CLINICAL PRACTICE

Because hypoalbuminemia is common in the ICU setting, inappropriate ordering of serum albumin levels may lead to unnecessary albumin administration or excessive macronutrient administration in nutritional regimens, leading to possible adverse effects and added costs.

CONCLUSIONS

With the exception of the need to order serum albumin levels as a component of selected severity-of-illness scoring systems, there is little evidence or justification for routinely ordering levels in critically ill patients.

Establishment and effectiveness evaluation of a scoring system for exertional heat stroke by retrospective analysis

BACKGROUND

Heat stroke (HS) is a serious, life-threatening disease. However, there is no scoring system for HS so far. This research is to establish a scoring system that can quantitatively assess the severity of exertional heat stroke (EHS).

METHODS

Data were collected from a total of 170 exertional heat stroke (EHS) patients between 2005 and 2016 from 52 hospitals in China. Univariate statistical methods and comparison of the area under the receiver operating characteristic (ROC) curve (AUC) were used to screen exertional heat stroke score (EHSS) parameters, including but not limited body temperature (T), Glasgow Coma Scale (GCS) and others. By comparing the sizes of the AUCs of the APACHE II, SOFA and EHSS assessments, the effectiveness of EHSS in evaluating the prognosis of EHS patients was verified.

RESULTS

Through screening with a series of methods, as described above, the present study determined 12 parameters - body temperature (T), GCS, pH, lactate (Lac), platelet count (PLT), prothrombin time (PT), fibrinogen (Fib), troponin I (TnI), aspartate aminotransferase (AST), total bilirubin (TBIL), creatinine (Cr) and acute gastrointestinal injury (AGI) classification - as EHSS parameters. It is a 0-47 point system designed to reflect increasing severity of heat stroke. Low (EHSS< 20) and high scores (EHSS> 35) showed 100% survival and 100% mortality, respectively. We found that AUCEHSS > AUCSOFA > AUCAPACHE II.

CONCLUSION

A total of 12 parameters - T, GCS, pH, Lac, PLT, PT, Fib, TnI, AST, TBIL, Cr and gastrointestinal AGI classification - are the EHSS parameters with the best effectiveness in evaluating the prognosis of EHS patients. As EHSS score increases, the mortality rate of EHS patients gradually increases.

What factors predict length of stay in the intensive care unit? Systematic review and meta-analysis

PURPOSE

Studies have shown that a small percentage of ICU patients have prolonged length of stay (LoS) and account for a large proportion of resource use. Therefore, the identification of prolonged stay patients can improve unit efficiency. In this study, we performed a systematic review and meta-analysis to understand the risk factors of ICU LoS.

MATERIALS AND METHODS

We searched MEDLINE, Embase and Scopus databases from inception to November 2018. The searching process focused on papers presenting risk factors of ICU LoS. A meta-analysis was performed for studies reporting appropriate statistics.

RESULTS

From 6906 citations, 113 met the eligibility criteria and were reviewed. A meta-analysis was performed for six factors from 28 papers and concluded that patients with mechanical ventilation, hypomagnesemia, delirium, and malnutrition tend to have longer stay, and that age and gender were not significant factors.

CONCLUSIONS

This work suggested a list of risk factors that should be considered in prediction models for ICU LoS, as follows: severity scores, mechanical ventilation, hypomagnesemia, delirium, malnutrition, infection, trauma, red blood cells, and PaO2:FiO2. Our findings can be used by prediction models to improve their predictive capacity of prolonged stay patients, assisting in resource allocation, quality improvement actions, and benchmarking analysis.

Investigating SOFA, delta-SOFA and MPM-III for mortality prediction among critically ill patients at a private tertiary hospital ICU in Kenya: A retrospective cohort study

Background

Outcomes in well-resourced, intensive care units (ICUs) in Kenya are thought to be comparable to those in high-income countries (HICs) but risk-adjusted mortality data is unavailable. We undertook an evaluation of the Aga Khan University Hospital, Nairobi ICU to analyze patient clinical-demographic characteristics, compare the performance of Sequential Organ Failure Assessment (SOFA), delta-SOFA at 48 hours and Mortality Prediction Model-III (MPM-III) mortality prediction systems, and identify factors associated with increased risk of mortality.

Methods

A retrospective cohort study was conducted of adult patients admitted to the ICU between January 2015 and September 2017. SOFA and MPM-III scores were determined at admission and SOFA repeated at 48 hours.

Results

Approximately 33% of patients did not meet ICU admission criteria. Mortality among the population of critically ill patients in the ICU was 31.7%, most of whom were male (61.4%) with a median age of 53.4 years. High adjusted odds of mortality were found among critically ill patients with leukemia (aOR 6.32, p<0.01), tuberculosis (aOR 3.96, p<0.01), post-cardiac arrest (aOR 3.57, p<0.01), admissions from the step-down unit (aOR 3.13, p<0.001), acute kidney injury (aOR 2.97, p<0.001) and metastatic cancer (aOR 2.45, p = 0.04). The area under the receiver-operating characteristic (ROC) curve of admission SOFA was 0.77 (95% CI, 0.73–0.81), MPM-III 0.74 (95% CI, 0.69–0.79), delta-SOFA 0.69 (95% CI, 0.63–0.75) and 48-hour SOFA 0.83 (95% CI, 0.79–0.87). The difference between SOFA at 48 hours and admission SOFA, MPM-III and delta-SOFA was statistically significant (chi² = 17.1, 24.2 and 26.5 respectively; p<0.001). Admission SOFA, MPM-III and 48-hour SOFA were well calibrated (p >0.05) while delta-SOFA was borderline (p = 0.05).

Conclusion

Mortality among the critically ill was higher than expected in this well-resourced ICU. 48-hour SOFA performed better than admission SOFA, MPM-III and delta-SOFA in our cohort. While a large proportion of patients did not meet admission criteria but were boarded in the ICU, critically ill patients stepped-up from the step-down unit were unlikely to survive. Patients admitted following a cardiac arrest, and those with advanced disease such as leukemia, stage-4 HIV and metastatic cancer, had particularly poor outcomes. Policies for fair allocation of beds, protocol-driven admission criteria and appropriate case selection could contribute to lowering the risk of mortality among the critically ill to a level on par with HICs.

Comparing Eight Prognostic Scores in Predicting Mortality of Patients with Acute-On-Chronic Liver Failure Who Were Admitted to an ICU: A Single-Center Experience

Limited data is available on long-term outcome predictions for patients with acute-on-chronic liver failure (ACLF) in an intensive care unit (ICU) setting. Assessing the reliability and accuracy of several mortality prediction models for these patients is helpful. Two hundred forty-nine consecutive patients with ACLF and admittance to the liver ICU in a single center in northern Taiwan between December 2012 and March 2015 were enrolled in the study and were tracked until February 2017. Ninety-one patients had chronic hepatitis B-related cirrhosis. Clinical features and laboratory data were collected at or within 24 h of the first ICU admission course. Eight commonly used clinical scores in chronic liver disease were calculated. The primary endpoint was overall survival. Acute physiology and chronic health evaluation (APACHE) III and chronic liver failure consortium (CLIF-C) ACLF scores were significantly superior to other models in predicting overall mortality as determined by time-dependent receiver operating characteristic (ROC) curve analysis (area under the ROC curve (AUROC): 0.817). Subgroup analysis of patients with chronic hepatitis B-related cirrhosis displayed similar results. CLIF-C organ function (OF), CLIF-C ACLF, and APACHE III scores were statistically superior to the mortality probability model III at zero hours (MPM0-III) and the simplified acute physiology (SAP) III scores in predicting 28-day mortality. In conclusion, for 28-day and overall mortality prediction of patients with ACLF admitted to the ICU, APACHE III, CLIF-OF, and CLIF-C ACLF scores might outperform other models. Further prospective study is warranted.

Ideal Severity of Illness Scoring System for Critically Ill Cancer Patients: A Dream

How to cite this article: Krishna B. Ideal Severity of Illness Scoring System for Critically Ill Cancer Patients: A Dream. Indian J Crit Care Med 2020;24(4):215.

Early noncardiovascular organ failure and mortality in the cardiac intensive care unit

Background

Noncardiac organ failure has been associated with worse outcomes among a cardiac intensive care unit (CICU) population.

Hypothesis

We hypothesized that early organ failure based on the sequential organ failure assessment (SOFA) score would be associated with mortality in CICU patients.

Methods

Adult CICU patients from 2007 to 2015 were reviewed. Organ failure was defined as any SOFA organ subscore ≥3 on the first CICU day. Organ failure was evaluated as a predictor of hospital mortality and postdischarge survival after adjustment for illness severity and comorbidities.

Results

We included 10 004 patients with a mean age of 67 ± 15 years (37% female). Admission diagnoses included acute coronary syndrome in 43%, heart failure in 46%, cardiac arrest in 12%, and cardiogenic shock in 11%. Organ failure was present in 31%, including multiorgan failure in 12%. Hospital mortality was higher in patients with organ failure (22% vs 3%, adjusted OR 3.0, 95% CI 2.5‐3.7, P < .001). After adjustment, each failing organ system predicted twofold higher odds of hospital mortality (adjusted OR 1.9, 95% CI 1.1‐2.1, P < .001). Mortality risk was highest with cardiovascular, coagulation and liver failure. Among hospital survivors, organ failure was associated with higher adjusted postdischarge mortality risk (P < .001); multiorgan failure did not confer added long‐term mortality risk.

Conclusions

Early noncardiovascular organ failure, especially multiorgan failure, is associated with increased hospital mortality in CICU patients, and this risk continues after hospital discharge, emphasizing the need to promote early recognition of organ failure in CICU patients.

Risk factors of in-hospital mortality in adult postcardiotomy cardiogenic shock patients successfully weaned from venoarterial extracorporeal membrane oxygenation

OBJECTIVE

Mortality of adult postcardiotomy cardiogenic shock patients after successfully weaned from venoarterial extracorporeal membrane oxygenation remains high. The objective of this study is to identify the risk factors associated with mortality after successfully weaning from venoarterial extracorporeal membrane oxygenation in adult postcardiotomy cardiogenic shock patients.

METHODS

All consecutive patients who were successfully weaned from venoarterial extracorporeal membrane oxygenation between January 2011 and December 2016 at the Beijing Anzhen Hospital were analyzed retrospectively. Multivariate logistic regression was performed to identify risk factors associated with in-hospital mortality after successfully weaning from venoarterial extracorporeal membrane oxygenation.

RESULTS

In total, 212 (58.4%) of 363 postcardiotomy cardiogenic shock patients were successfully weaned from venoarterial extracorporeal membrane oxygenation. The non-survivors had a longer duration of extracorporeal membrane oxygenation than the survivors (120.0 (98.0, 160.50) vs. 100.0 (77.0, 126.0), p = 0.000). Variables associated with mortality of patients successfully weaned from extracorporeal membrane oxygenation by univariable analysis were age, diabetes, vasoactive inotropic score pre-extracorporeal membrane oxygenation, vasoactive inotropic score at weaning, left ventricular ejection fraction at weaning, central venous pressure at weaning, sequential organ failure assessment score pre-extracorporeal membrane oxygenation, sequential organ failure assessment at weaning, survival after venoarterial ECMO pre-extracorporeal membrane oxygenation, and survival after venoarterial ECMO at weaning. In the multivariate analysis, sequential organ failure assessment score at weaning (odds ratio = 1.889, 95% confidence interval = 1.460-2.455, p < 0.001) was an independent risk factor for in-hospital mortality of patients successfully weaned from venoarterial extracorporeal membrane oxygenation. The cumulative 30-day survival rate in patients with a sequential organ failure assessment score < 7 was significantly (p < 0.001) higher than in patients with a sequential organ failure assessment score ⩾ 7 (87% vs. 56.7%, p < 0.001).

CONCLUSION

Vasoactive inotropic score, left ventricular ejection fraction, central venous pressure, and sequential organ failure assessment score at weaning were associated with in-hospital mortality for postcardiotomy cardiogenic shock patients successfully weaned from venoarterial extracorporeal membrane oxygenation. Sequential organ failure assessment score might help clinicians to predict in-hospital mortality for patients successfully weaned from venoarterial extracorporeal membrane oxygenation.

Intensive Care Outcomes and Mortality Prediction at a National Referral Hospital in Western Kenya

RATIONALE

The burden of critical care is greatest in resource-limited settings. Intensive care unit (ICU) outcomes at public hospitals in Kenya are unknown. The present study is timely, given the Kenyan Ministry of Health initiative to expand ICU capacity.

OBJECTIVES

To identify factors associated with mortality at Moi Teaching and Referral Hospital and validate the Mortality Probability Admission Model II (MPM₀-II).

METHODS

A retrospective cohort of 450 patients from January 1, 2013, to April 5, 2015, was evaluated using demographics, presenting diagnoses, interventions, mortality, and cost data.

RESULTS

ICU mortality was 53.6%, and 30-day mortality was 57.3%. Most patients were male (61%) and at least 18 years old (70%); the median age was 29 years. Factors associated with high adjusted odds of mortality were as follows: age younger than 10 years (adjusted odds ratio [aOR], 3.59; P ≤ 0.001), ages 35-49 years (aOR, 3.13; P = 0.002), and age above 50 years (aOR, 2.86; P = 0.004), with reference age range 10-24 years; sepsis (aOR, 3.39; P = 0.01); acute stroke (aOR, 8.14; P = 0.011); acute respiratory failure or mechanical ventilation (aOR, 6.37; P < 0.001); and vasopressor support (aOR, 7.98; P < 0.001). Drug/alcohol poisoning (aOR, 0.33; P = 0.005) was associated with lower adjusted odds of mortality. MPM₀-II discrimination showed an area under the receiver operating characteristic curve of 0.78 (95% confidence interval, 0.72-0.82). The result of the Hosmer-Lemeshow test for calibration was significant (P < 0.001).

CONCLUSIONS

In a Kenyan public ICU, high mortality was noted despite the use of advanced therapies. MPM₀-II has acceptable discrimination but poor calibration. Modification of MPM₀-II or development of a new model using a prospective multicenter global collaboration is needed. Standardized triage and treatment protocols for high-risk diagnoses are needed to improve ICU outcomes.

An Evaluation of the Influence of Body Mass Index on Severity Scoring

OBJECTIVES

Although one third or more of critically ill patients in the United States are obese, obesity is not incorporated as a contributing factor in any of the commonly used severity of illness scores. We hypothesize that selected severity of illness scores would perform differently if body mass index categorization was incorporated and that the performance of these score models would improve after consideration of body mass index as an additional model feature.

DESIGN

Retrospective cohort analysis from a multicenter ICU database which contains deidentified data for more than 200,000 ICU admissions from 208 distinct ICUs across the United States between 2014 and 2015.

SETTING

First ICU admission of patients with documented height and weight.

PATIENTS

One-hundred eight-thousand four-hundred two patients from 189 different ICUs across United States were included in the analyses, of whom 4,661 (4%) were classified as underweight, 32,134 (30%) as normal weight, 32,278 (30%) as overweight, 30,259 (28%) as obese, and 9,070 (8%) as morbidly obese.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

To assess the effect of adding body mass index as a risk adjustment element to the Acute Physiology and Chronic Health Evaluation IV and Oxford Acute Severity of Illness scoring systems, we examined the impact of this addition on both discrimination and calibration. We performed three assessments based upon 1) the original scoring systems, 2) a recalibrated version of the systems, and 3) a recalibrated version incorporating body mass index as a covariate. We also performed a subgroup analysis in groups defined using World Health Organization guidelines for obesity. Incorporating body mass index into the models provided a minor improvement in both discrimination and calibration. In a subgroup analysis, model discrimination was higher in groups with higher body mass index, but calibration worsened.

CONCLUSIONS

The performance of ICU prognostic models utilizing body mass index category as a scoring element was inconsistent across body mass index categories. Overall, adding body mass index as a risk adjustment variable led only to a minor improvement in scoring system performance.

Semi-quantitative visual assessment of chest radiography is associated with clinical outcomes in critically ill patients

Background

Respiratory pathology is a major driver of mortality in the intensive care unit (ICU), even in the absence of a primary respiratory diagnosis. Prior work has demonstrated that a visual scoring system applied to chest radiographs (CXR) is associated with adverse outcomes in ICU patients with Acute Respiratory Distress Syndrome (ARDS). We hypothesized that a simple, semi-quantitative CXR score would be associated with clinical outcomes for the general ICU population, regardless of underlying diagnosis.

Methods

All individuals enrolled in the Registry of Critical Illness at Brigham and Women’s Hospital between June 2008 and August 2018 who had a CXR within 24 h of admission were included. Each patient’s CXR was assigned an opacification score of 0–4 in each of four quadrants with the total score being the sum of all four quadrants. Multivariable negative binomial, logistic, and Cox regression, adjusted for age, sex, race, immunosuppression, a history of chronic obstructive pulmonary disease, a history of congestive heart failure, and APACHE II scores, were used to assess the total score’s association with ICU length of stay (LOS), duration of mechanical ventilation, in-hospital mortality, 60-day mortality, and overall mortality, respectively.

Results

A total of 560 patients were included. Higher CXR scores were associated with increased mortality; for every one-point increase in score, in-hospital mortality increased 10% (OR 1.10, CI 1.05–1.16, p < 0.001) and 60-day mortality increased by 12% (OR 1.12, CI 1.07–1.17, p < 0.001). CXR scores were also independently associated with both ICU length of stay (rate ratio 1.06, CI 1.04–1.07, p < 0.001) and duration of mechanical ventilation (rate ratio 1.05, CI 1.02–1.07, p < 0.001).

Conclusions

Higher values on a simple visual score of a patient’s CXR on admission to the medical ICU are associated with increased in-hospital mortality, 60-day mortality, overall mortality, length of ICU stay, and duration of mechanical ventilation.

Development and Evaluation of an Automated Machine Learning Algorithm for In-Hospital Mortality Risk Adjustment Among Critical Care Patients

OBJECTIVES

Risk adjustment algorithms for ICU mortality are necessary for measuring and improving ICU performance. Existing risk adjustment algorithms are not widely adopted. Key barriers to adoption include licensing and implementation costs as well as labor costs associated with human-intensive data collection. Widespread adoption of electronic health records makes automated risk adjustment feasible. Using modern machine learning methods and open source tools, we developed and evaluated a retrospective risk adjustment algorithm for in-hospital mortality among ICU patients. The Risk of Inpatient Death score can be fully automated and is reliant upon data elements that are generated in the course of usual hospital processes.

SETTING

One hundred thirty-one ICUs in 53 hospitals operated by Tenet Healthcare.

PATIENTS

A cohort of 237,173 ICU patients discharged between January 2014 and December 2016.

DESIGN

The data were randomly split into training (36 hospitals), and validation (17 hospitals) data sets. Feature selection and model training were carried out using the training set while the discrimination, calibration, and accuracy of the model were assessed in the validation data set.

MEASUREMENTS AND MAIN RESULTS

Model discrimination was evaluated based on the area under receiver operating characteristic curve; accuracy and calibration were assessed via adjusted Brier scores and visual analysis of calibration curves. Seventeen features, including a mix of clinical and administrative data elements, were retained in the final model. The Risk of Inpatient Death score demonstrated excellent discrimination (area under receiver operating characteristic curve = 0.94) and calibration (adjusted Brier score = 52.8%) in the validation dataset; these results compare favorably to the published performance statistics for the most commonly used mortality risk adjustment algorithms.

CONCLUSIONS

Low adoption of ICU mortality risk adjustment algorithms impedes progress toward increasing the value of the healthcare delivered in ICUs. The Risk of Inpatient Death score has many attractive attributes that address the key barriers to adoption of ICU risk adjustment algorithms and performs comparably to existing human-intensive algorithms. Automated risk adjustment algorithms have the potential to obviate known barriers to adoption such as cost-prohibitive licensing fees and significant direct labor costs. Further evaluation is needed to ensure that the level of performance observed in this study could be achieved at independent sites.