Authors
Elizabeth C Hyde, Luiz E V da Silva, Sanjiv Mehta, Hannah Huth, Vinay M Nadkarni, Natalie Napolitano, Cheryl Dominick, Amanda J Nickel, Lindsay N Shepard, Fuchiang Rich Tsui, Akira Nishisaki
Published in
Critical care explorations. Volume 8. Issue 11. Pages e1509. Nov 01, 2026. Epub Nov 17, 2026.
Abstract
Bag-mask ventilation (BMV) is a critical phase of pediatric airway management, yet objective feedback remains limited.
To develop and validate a two-stage, CO2-only framework for pediatric BMV assessment comprising automated breath segmentation and machine learning classification of breaths according to predefined low-quality criteria for insufficient ventilation. We hypothesized that our model would achieve area under the receiver operating characteristic curve (AUROC) greater than or equal to 0.75 and area under the precision-recall curve (AUPRC) greater than or equal to 0.75.
We retrospectively analyzed capnography waveforms from 85 critically ill children receiving peri-intubation BMV in a quaternary-care PICU. Flow and pressure data established reference-standard low-quality criteria: expiratory tidal volume (VTE) less than 4 mL/kg of ideal body weight, mask leak greater than 40%, or upper airway obstruction (VTE < 4 mL/kg with peak inspiratory pressure > 30 cm H2O and leak > 80%).
CO2-based segmentation was validated in 10 randomly selected patients comprising 605 manually adjudicated breaths. Classification performance was internally validated using repeated 10-fold cross-validation within the derivation cohort; no independent validation cohort was used.
A rule-based algorithm segmented continuous CO2 waveforms into breaths. Deep learning architectures then used CO2 waveforms alone to classify breaths according to whether they met predefined low-quality criteria.
The CO2-only algorithm identified 6942 breaths (median, 66 per patient [interquartile range, 47-100 per patient]); 49.5% met at least one low-quality criterion. Segmentation achieved positive predictive value 0.95 (95% CI, 0.93-0.96) and sensitivity 0.86 (95% CI, 0.83-0.89). Twenty-seven breaths were adjudicated as false-positive detections (0.05; 95% CI, 0.036-0.074). CO2-only segmentation may miss ventilatory attempts with minimal or absent exhaled CO2. The best-performing classification model achieved AUROC 0.79 (95% CI, 0.66-0.89) and AUPRC 0.75 (95% CI, 0.57-0.88).
A CO2-only framework showed proof-of-concept feasibility for breath segmentation and quality classification during pediatric BMV, supporting further development of objective bedside feedback systems.
PMID:
43049301
Bibliographic data and abstract were imported from PubMed on 18 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 5
- Comments 0