Asynchrony Detection model Based on Frequency-Domain Analysis of Airway Pressure

BackgroundPatient–ventilator asynchrony (PVA) is frequent in mechanically ventilated patients and can negatively affect outcomes. Despite detection methods proposed in the literature, most rely on invasive measurements or complex interpretation, limiting implementation in ICU practice.ObjectiveThis study aimed to develop and validate a non-invasive method for PVA detection based on frequency-domain filtering and elastance modelling, to define an Asynchrony Index (AI%) through Receiver Operating Characteristic (ROC) analysis.MethodsAirway pressure signals were interpreted as the sum of a synchronous baseline and an asynchronous component. A low-pass filter with a 0.8 Hz cut-off was applied to isolate the baseline signal, while time-varying elastance was computed from original and filtered signals. This deviation yielded the Asynchrony Event Percentage (AE%), quantifying mismatches between patient effort and ventilator support. Using two training datasets, ROC analysis was performed to define the optimal AE% tolerance band and decision threshold for AI%. A third dataset was used for validation.ResultsROC analysis identified ±5% as the optimal AE% tolerance band and established a classification threshold of AI% > 70%. At this point, the model achieved 75% sensitivity and 100% specificity (AUC = 0.87). The AI% differentiated synchronous from asynchronous ventilation in cases, particularly for flow mismatch and premature cycling, but showed reduced reliability for isolated ineffective efforts and double triggering.ConclusionThe AI% provides a promising, non-invasive tool for PVA detection using routinely available ventilator signals. Further studies on larger, more heterogeneous populations are needed to improve sensitivity and confirm clinical applicability.