The following is a summary of “Flow starvation during square-flow assisted ventilation detected by supervised deep learning techniques,” published in the March 2024 issue of Critical Care by Haro et al.
Flow starvation, a mismatch between ventilator supply and patient demand during mechanical ventilation, is often missed by traditional methods but is detectable with artificial intelligence (AI).
Researchers started a retrospective study to develop a machine-learning algorithm for spotting airway pressure disruptions during square-flow ventilation (SFV) and patient-initiated breaths.
They undertook a multicenter observational study, enrolling adult critically ill patients who underwent mechanical ventilation for over 24 hours using SFV. Five intensive care experts served as references to evaluate the severity of airway pressure deformation. Utilizing convolutional neural network (CNN) and recurrent neural network (RNN) models.Trained and assessed for performance metrics. In esophageal pressure measurement (ΔPes) patients, the link between inspiratory effort and airway pressure deformation was studied.
The results showed 6,428 breaths from 28 patients. Among these, 42% showed normal-mild, 23% moderate, and 34% severe airway pressure deformation. The RNN algorithm achieved an accuracy of 87.9% [87.6–88.3], while the CNN attained 86.8% [86.6–87.4]. Double triggering occurred in 8.8% of breaths, consistently linked with severe airway pressure deformation. Subgroup analysis revealed that 74.4% of breaths with severe airway pressure deformation had ΔPes > 10 cmH2O, and 37.2% had ΔPes > 15 cmH2O.
In a retrospective analysis, investigators concluded that the developed algorithm effectively identified flow starvation events during SFV and patient-triggered breaths.
Source: ccforum.biomedcentral.com/articles/10.1186/s13054-024-04845-y
