Predicting ICU Mortality in ARDS Patients

Acute Respiratory Distress Syndrome

outcome, mechanical ventilation, intensive care unit, machine learning

The investigators are planning to perform a secondary analysis of an academic dataset of 1,303 patients with moderate-to-severe acute respiratory distress syndrome (ARDS) included in several published cohorts (NCT00736892, NCT02288949, NCT02836444, NCT03145974), aimed to characterize the best early model to predict duration of mechanical ventilation and mortality in the intensive care unit (ICU) after ARDS diagnosis using machine learning approaches.

The acute respiratory distress syndrome (ARDS) is a severe form of acute hypoxemic respiratory failure in Critical Care Units worldwide. Most ARDS patients requiere mechanical ventilation (MV). Few studies have investigated the prediction of MV duration and mortality of ARDS. For model description, the investigators will extract data from the first two ICU days after diagnosis of moderate-to-severe ARDS from patients included in the de-identified database, which includes 1,303 mechanically ventilated patients enrolled in several observational cohorts in Spain, coordinated by the principal investigator (JV), and funded by the Instituto de Salud Carlos III (ISCIII). The investigators will follow the TRIPOD guidelines and machine learning tecniques will be implemented (Random Forest, XGBoost, Logistic regression analysis, and/or neural networks) for development of the prediction model, and the accuracy will be compared to those of existing scoring systems for assessing ICU severity (APACHE II, SOFA) and the PaO2/FiO2 ratio. For external validation, the investigators will use 303 patients enrolled in a contemporary observational study (NCT03145974). The investigators will evaluate the accuracy of prediction models by calculating the respective confusion matrices and several statistics such as sensitivity, specificity, positive predictive value, and negative predictive value for mortality and duration of MV. Investigators will select the best probabilistic model with a minimum number of clinical variables.

No related papers found yet.

Spain