A feedback-driven ventilation model for assessing airway secretions in mechanically ventilated patients

TL;DR

A new model helps monitor airway secretions in ventilated ICU patients by simulating respiratory dynamics and identifying secretion levels non-invasively.

Contribution

A feedback-driven ventilation model and secretion index for real-time, non-invasive secretion monitoring in mechanically ventilated patients.

Findings

Simulations showed waveform changes like reduced inspiratory flow and prolonged expiration due to secretion accumulation.

Clustering of patient data identified three secretion levels: low, medium, and high.

A model-informed secretion index was developed for continuous bedside monitoring of secretion accumulation.

Abstract

A mechanistic compartmental model with a feedback-driven simulation framework was developed to investigate the impact of airway secretion accumulation and its removal on the respiratory dynamics of mechanically ventilated patients. Understanding these dynamics is essential for secretion management and improving respiratory care in the intensive care unit (ICU).

The model simulates pressure support ventilation by incorporating airway resistances, lung and chest wall compliances, and patient effort via a dynamic respiratory muscle pressure term, enabling realistic modelling of patient-ventilator interaction. To validate the model, simulated waveforms were compared against clinical waveform recordings. Waveform features sensitive to secretion-related changes, as indicated by the model, were then extracted from the patient waveform recordings. The Wasserstein distance metric was used to quantify shifts in pre- and post-suction feature distributions, and unsupervised clustering was applied to identify distinct patient groups corresponding to low, medium, and high secretion levels.

The simulations revealed characteristic changes in ventilator waveforms associated with secretion accumulation, including reduced inspiratory flow and prolonged expiration. Analysis of patient data using clustering methods identified distinct groups corresponding to low, medium, and high levels of secretion. Further, we introduce a model-informed secretion index derived from the simulations and patient data, enabling non-invasive and continuous monitoring of secretion accumulation at the bedside.

This study demonstrates the potential of physiology-informed, model-based approaches for real-time assessment of secretion accumulation in mechanically ventilated patients. The proposed framework supports personalized respiratory care by providing clinicians with data-driven insights into secretion accumulation, paving the way for more precise secretion management strategies in the ICU.

Figures