Abstract
Acute respiratory distress syndrome (ARDS) is a devastating critical care syndrome with significant morbidity and mortality. The objective of this study was to evaluate the predictive values of dynamic clinical indices by developing machine-learning (ML) models for early and accurate clinical assessment of the disease prognosis of ARDS. We conducted a retrospective observational study by applying dynamic clinical data collected in the ARDSNet FACTT Trial (n = 1000) to ML-based algorithms for predicting mortality. In order to compare the significance of clinical features dynamically, we further applied the random forest (RF) model to nine selected clinical parameters acquired at baseline and day 3 independently. An RF model trained using clinical data collected at day 3 showed improved performance and prognostication efficacy (area under the curve [AUC]: 0.84, 95% CI: 0.78-0.89) compared to baseline with an AUC value of 0.72 (95% CI: 0.65-0.78). Mean airway pressure (MAP), bicarbonate, age, platelet count, albumin, heart rate, and glucose were the most significant clinical indicators associated with mortality at day 3. Thus, clinical features collected early (day 3) improved performance of integrative ML models with better prognostication for mortality. Among these, MAP represented the most important feature for ARDS patients' early risk stratification.