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

OBJECTIVES: 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). DESIGN: A development, testing, and external validation study using clinical data from four prospective, multicenter, observational cohorts. SETTING: A network of multidisciplinary ICUs. PATIENTS: A total of 1,303 patients with moderate-to-severe ARDS managed with lung-protective ventilation. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: 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). CONCLUSIONS: 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.

Stratification for Identification of Prognostic Categories In the Acute RESpiratory Distress Syndrome (SPIRES) Score.

OBJECTIVES: To develop a scoring model for stratifying patients with acute respiratory distress syndrome into risk categories (Stratification for identification of Prognostic categories In the acute RESpiratory distress syndrome score) for early prediction of death in the ICU, independent of the underlying disease and cause of death. DESIGN: A development and validation study using clinical data from four prospective, multicenter, observational cohorts. SETTING: A network of multidisciplinary ICUs. PATIENTS: One-thousand three-hundred one patients with moderate-to-severe acute respiratory distress syndrome managed with lung-protective ventilation. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: The study followed Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis guidelines for prediction models. We performed logistic regression analysis, bootstrapping, and internal-external validation of prediction models with variables collected within 24 hours of acute respiratory distress syndrome diagnosis in 1,000 patients for model development. Primary outcome was ICU death. The Stratification for identification of Prognostic categories In the acute RESpiratory distress syndrome score was based on patient's age, number of extrapulmonary organ failures, values of end-inspiratory plateau pressure, and ratio of Pao2 to Fio2 assessed at 24 hours of acute respiratory distress syndrome diagnosis. The pooled area under the receiver operating characteristic curve across internal-external validations was 0.860 (95% CI, 0.831-0.890). External validation in a new cohort of 301 acute respiratory distress syndrome patients confirmed the accuracy and robustness of the scoring model (area under the receiver operating characteristic curve = 0.870; 95% CI, 0.829-0.911). The Stratification for identification of Prognostic categories In the acute RESpiratory distress syndrome score stratified patients in three distinct prognostic classes and achieved better prediction of ICU death than ratio of Pao2 to Fio2 at acute respiratory distress syndrome onset or at 24 hours, Acute Physiology and Chronic Health Evaluation II score, or Sequential Organ Failure Assessment scale. CONCLUSIONS: The Stratification for identification of Prognostic categories In the acute RESpiratory distress syndrome score represents a novel strategy for early stratification of acute respiratory distress syndrome patients into prognostic categories and for selecting patients for therapeutic trials.

Is Overall Mortality the Right Composite Endpoint in Clinical Trials of Acute Respiratory Distress Syndrome?

OBJECTIVES: Overall mortality in patients with acute respiratory distress syndrome is a composite endpoint because it includes death from multiple causes. In most acute respiratory distress syndrome trials, it is unknown whether reported deaths are due to acute respiratory distress syndrome or the underlying disease, unrelated to the specific intervention tested. We investigated the causes of death after contracting acute respiratory distress syndrome in a large cohort. DESIGN: A secondary analysis from three prospective, multicenter, observational studies. SETTING: A network of multidisciplinary ICUs. PATIENTS: We studied 778 patients with moderate-to-severe acute respiratory distress syndrome treated with lung-protective ventilation. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We examined death in the ICU from individual causes. Overall ICU mortality was 38.8% (95% CI, 35.4-42.3). Causes of acute respiratory distress syndrome modified the risk of death. Twenty-three percent of deaths occurred from refractory hypoxemia due to nonresolving acute respiratory distress syndrome. Most patients died from causes unrelated to acute respiratory distress syndrome: 48.7% of nonsurvivors died from multisystem organ failure, and cancer or brain injury was involved in 37.1% of deaths. When quantifying the true burden of acute respiratory distress syndrome outcome, we identified 506 patients (65.0%) with one or more exclusion criteria for enrollment into current interventional trials. Overall ICU mortality of the "trial cohort" (21.3%) was markedly lower than the parent cohort (relative risk, 0.55; 95% CI, 0.43-0.70; p < 0.000001). CONCLUSIONS: Most deaths in acute respiratory distress syndrome patients are not directly related to lung damage but to extrapulmonary multisystem organ failure. It would be challenging to prove that specific lung-directed therapies have an effect on overall survival.

Age, PaO2/FIO2, and Plateau Pressure Score: A Proposal for a Simple Outcome Score in Patients With the Acute Respiratory Distress Syndrome.

OBJECTIVES: Although there is general agreement on the characteristic features of the acute respiratory distress syndrome, we lack a scoring system that predicts acute respiratory distress syndrome outcome with high probability. Our objective was to develop an outcome score that clinicians could easily calculate at the bedside to predict the risk of death of acute respiratory distress syndrome patients 24 hours after diagnosis. DESIGN: A prospective, multicenter, observational, descriptive, and validation study. SETTING: A network of multidisciplinary ICUs. PATIENTS: Six-hundred patients meeting Berlin criteria for moderate and severe acute respiratory distress syndrome enrolled in two independent cohorts treated with lung-protective ventilation. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Using individual demographic, pulmonary, and systemic data at 24 hours after acute respiratory distress syndrome diagnosis, we derived our prediction score in 300 acute respiratory distress syndrome patients based on stratification of variable values into tertiles, and validated in an independent cohort of 300 acute respiratory distress syndrome patients. Primary outcome was in-hospital mortality. We found that a 9-point score based on patient's age, PaO2/FIO2 ratio, and plateau pressure at 24 hours after acute respiratory distress syndrome diagnosis was associated with death. Patients with a score greater than 7 had a mortality of 83.3% (relative risk, 5.7; 95% CI, 3.0-11.0), whereas patients with scores less than 5 had a mortality of 14.5% (p < 0.0000001). We confirmed the predictive validity of the score in a validation cohort. CONCLUSIONS: A simple 9-point score based on the values of age, PaO2/FIO2 ratio, and plateau pressure calculated at 24 hours on protective ventilation after acute respiratory distress syndrome diagnosis could be used in real time for rating prognosis of acute respiratory distress syndrome patients with high probability.

Clinical and biological markers for predicting ARDS and outcome in septic patients.

Sepsis is a common cause of acute respiratory distress syndrome (ARDS) associated with a high mortality. A panel of biomarkers (BMs) to identify septic patients at risk for developing ARDS, or at high risk of death, would be of interest for selecting patients for therapeutic trials, which could improve ARDS diagnosis and treatment, and survival chances in sepsis and ARDS. We measured nine protein BMs by ELISA in serum from 232 adult septic patients at diagnosis (152 required invasive mechanical ventilation and 72 had ARDS). A panel including the BMs RAGE, CXCL16 and Ang-2, plus PaO2/FiO2, was good in predicting ARDS (area under the curve = 0.88 in total septic patients). Best performing panels for ICU death are related to the presence of ARDS, need for invasive mechanical ventilation, and pulmonary/extrapulmonary origin of sepsis. In all cases, the use of BMs improved the prediction by clinical markers. Our study confirms the relevance of RAGE, Ang-2, IL-1RA and SP-D, and is novel supporting the inclusion of CXCL16, in BMs panels for predicting ARDS diagnosis and ARDS and sepsis outcome.

Neurally adjusted ventilatory assist in acute respiratory failure: a randomized controlled trial.

PURPOSE: We hypothesized that neurally adjusted ventilatory assist (NAVA) compared to conventional lung-protective mechanical ventilation (MV) decreases duration of MV and mortality in patients with acute respiratory failure (ARF). METHODS: We carried out a multicenter, randomized, controlled trial in patients with ARF from several etiologies. Intubated patients ventilated for ≤ 5 days expected to require MV for ≥ 72 h and able to breathe spontaneously were eligible for enrollment. Eligible patients were randomly assigned based on balanced treatment assignments with a computerized randomization allocation sequence to two ventilatory strategies: (1) lung-protective MV (control group), and (2) lung-protective MV with NAVA (NAVA group). Allocation concealment was maintained at all sites during the trial. Primary outcome was the number of ventilator-free days (VFDs) at 28 days. Secondary outcome was all-cause hospital mortality. All analyses were done according to the intention-to-treat principle. RESULTS: Between March 2014 and October 2019, we enrolled 306 patients and randomly assigned 153 patients to the NAVA group and 153 to the control group. Median VFDs were higher in the NAVA than in the control group (22 vs. 18 days; between-group difference 4 days; 95% confidence interval [CI] 0 to 8 days; p = 0.016). At hospital discharge, 39 (25.5%) patients in the NAVA group and 47 (30.7%) patients in the control group had died (between-group difference - 5.2%, 95% CI - 15.2 to 4.8, p = 0.31). Other clinical, physiological or safety outcomes did not differ significantly between the trial groups. CONCLUSION: NAVA decreased duration of MV although it did not improve survival in ventilated patients with ARF.

Validation of a score for predicting fatal bleeding in patients receiving anticoagulation for venous thromboembolism.

BACKGROUND: The only available score to assess the risk for fatal bleeding in patients with venous thromboembolism (VTE) has not been validated yet. METHODS: We used the RIETE database to validate the risk-score for fatal bleeding within the first 3 months of anticoagulation in a new cohort of patients recruited after the end of the former study. Accuracy was measured using the ROC curve analysis. RESULTS: As of December 2011, 39,284 patients were recruited in RIETE. Of these, 15,206 had not been included in the former study, and were considered to validate the score. Within the first 3 months of anticoagulation, 52 patients (0.34%; 95% CI: 0.27-0.45) died of bleeding. Patients with a risk score of <1.5 points (64.1% of the cohort) had a 0.10% rate of fatal bleeding, those with a score of 1.5-4.0 (33.6%) a rate of 0.72%, and those with a score of >4 points had a rate of 1.44%. The c-statistic for fatal bleeding was 0.775 (95% CI 0.720-0.830). The score performed better for predicting gastrointestinal (c-statistic, 0.869; 95% CI: 0.810-0.928) than intracranial (c-statistic, 0.687; 95% CI: 0.568-0.806) fatal bleeding. The score value with highest combined sensitivity and specificity was 1.75. The risk for fatal bleeding was significantly increased (odds ratio: 7.6; 95% CI 3.7-16.2) above this cut-off value. CONCLUSIONS: The accuracy of the score in this validation cohort was similar to the accuracy found in the index study. Interestingly, it performed better for predicting gastrointestinal than intracranial fatal bleeding.