Let’s get in sync

What this paper is

A review of automatic detection of patient–ventilator asynchrony (PVA) over the last 15 years. PVA is a mismatch between the patient’s respiratory drive or effort and the ventilator’s breath delivery. Visual inspection of ventilator waveforms is time-consuming; the paper surveys rule-based, machine-learning, and deep-learning approaches aimed at bedside monitoring.

The authors identified 19 studies. Reported average sensitivity, specificity, and accuracy across those studies were 0.80, 0.93, and 0.92. Most algorithms remain offline, detect only a subset of PVA types (often ineffective effort and double triggering), or are still in development or validation (16 of 19 studies). A reference method for breathing effort was available in 11 of 19 studies. Three licensed algorithms are reported.

The paper argues that widespread implementation still needs better data quality, algorithms that cover multiple PVA types, external validation with effort measurements as ground truth, and prospective testing in different ICUs.

What Deep Breath contributed

Anton Balakirev and Daria Kozlova are co-authors, affiliated with Deep Breath B.V. The author group forms the SIREN consortium (Synergy in Respiration: Improving Patient–Ventilator Interaction via AI-based Monitoring).