Automatic Detection of Dyspnea in Mechanically Ventilated, Critically Ill Patients

Descripción detallada

Inclusion criteria: * Adults of either sex (≥18 years) * Non-c0mmunicative, endotracheally intubated in the past 24 hours. * Expected to require ≥24 hours of mechanical ventilation. Exclusion criteria: * Continuous neuromuscular blockade. * Ventilated on Airway Pressure Release Ventilation mode. Ventilator measured airway pressure and flow signals will be streamed at 31.25 Hz into 131.1-s time windows (epochs) for up to 5 days. A recursive algorithm will compute individual breath's inspiratory effort (Pmus) and calculate the inspiratory effort time-pressure product (PmusPTP cmH₂O·s·min-¹). Trained staff will assess the Mechanical Ventilation-Respiratory Distress Observation Scale (MV-RDOS), an intermittent and subjective assessment of dyspnea, at least twice daily. These will be time-matched to the airway signal epochs to determine their degree of correlation by Pearsons R2. Using receiver operating characteristic (ROC) area under the curve (AUC) analysis, the investigators will determine a dyspnea threshold for PmusPTP equivalent to the dyspnea cut-off for MV-RDOS \> 2.6, and report sensitivity/specificity. To address possible label misclassification, the investigators will train six supervised classifiers on concordant data pairs and apply a unanimity ensemble to label discordant pairs as "dyspneic", not dyspneic" and undetermined". Ensemble determined labels conflicting with those classified by MV-RDOS will be excluded from a secondary analysis. Primary outcome will be Pearson's R2 (PmusPTP vs MV-RDOS). Secondary outcome will be 28-day ventilator-free days.