Answers are grounded in this paper and cite their sources.
Acute Respiratory Distress Syndrome (ARDS) and sepsis are critical conditions with high mortality rates. Predicting extubation failure and 30-day mortality in patients with ARDS complicated by sepsis remains challenging due to overlapping pathophysiology and dynamic clinical variables. This study aimed to develop and validate parallel prediction models for extubation failure and 30-day mortality. We extracted 934 ventilated adults with ARDS and sepsis from MIMIC-IV (2017–2019). Feature selection used LassoCV, XGBoost-RFECV, and SVM-RFECV; their intersection defined endpoint-specific features. Five models (Logistic Regression, AdaBoost, Gradient Boosting Decision Tree [GBDT], Multilayer Perceptron [MLP], and Support Vector Classification [SVC]) were developed. SMOTE and grid search were restricted to the training set, and performance was evaluated in one held-out test set with bootstrap confidence intervals. Metrics included AUC, Brier score, F1 score, and graphical calibration. Lactate and body temperature were the only predictors common to the two endpoint-specific feature sets. For extubation failure, the GBDT AUC was 0.985 in training and 0.873 in the held-out test set; the test-set AUCs for Logistic Regression and SVC were 0.887 and 0.870, respectively. For 30-day mortality, the AdaBoost AUC was 0.863 in training and 0.792 in the test set. These estimates reflect internal evaluation using a single train–test split. The two independently developed models showed preliminary internal discrimination for extubation failure and 30-day mortality. Their combined clinical value was not assessed. External validation and clinical-impact evaluation are needed before either model can be recommended for clinical use.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: