[Paper Review] Improving Mechanical Ventilator Clinical Decision Support Systems with A Machine Learning Classifier for Determining Ventilator Mode

Gregory B. Rehm, Brooks Kuhn+4 more|arXiv (Cornell University)|Apr 29, 2019

Respiratory Support and Mechanisms19 references4 citations

TL;DR

This paper presents a machine learning classifier that accurately identifies mechanical ventilator modes in real time using breath-by-breath data, achieving a 97.52% average F1-score across five common US ventilation modes. The model enhances clinical decision support systems (CDSS) by overcoming data gaps in ventilator setting awareness, demonstrating robustness to missing data from sensor or software errors.

ABSTRACT

Clinical decision support systems (CDSS) will play an in-creasing role in improving the quality of medical care for critically ill patients. However, due to limitations in current informatics infrastructure, CDSS do not always have com-plete information on state of supporting physiologic monitor-ing devices, which can limit the input data available to CDSS. This is especially true in the use case of mechanical ventilation (MV), where current CDSS have no knowledge of critical ventilation settings, such as ventilation mode. To enable MV CDSS to make accurate recommendations related to ventilator mode, we developed a highly performant ma-chine learning model that is able to perform per-breath clas-sification of 5 of the most widely used ventilation modes in the USA with an average F1-score of 97.52%. We also show how our approach makes methodologic improvements over previous work and that it is highly robust to missing data caused by software/sensor error.

Motivation & Objective

• To address the lack of real-time ventilator mode awareness in clinical decision support systems (CDSS) for mechanical ventilation.
• To develop a machine learning classifier that can determine the current ventilator mode using per-breath physiological data.
• To improve CDSS performance by enabling accurate, real-time recognition of ventilation modes despite incomplete or missing data.
• To overcome limitations in existing CDSS that lack access to critical ventilator setting information due to infrastructural constraints.

Proposed method

• A supervised machine learning model is trained on per-breath features extracted from ventilator waveforms and settings.
• The model classifies breaths into one of five major ventilator modes used in the USA: assist-control, synchronized intermittent mandatory ventilation, pressure control, pressure support, and pressure-regulated volume control.
• Feature engineering includes inspiratory and expiratory time, tidal volume, peak airway pressure, and other breath-specific hemodynamic parameters.
• A robust training and evaluation pipeline is implemented to handle missing data, simulating real-world sensor and software failures.
• The model uses ensemble learning techniques to improve generalization and maintain high performance under data degradation.
• The system is validated on a real-world dataset from a critical care setting, ensuring clinical relevance.

Experimental results

Research questions

• RQ1 Can a machine learning model accurately classify mechanical ventilator modes using only per-breath physiological data?
• RQ2 How does the model perform under conditions of missing or corrupted sensor data common in clinical environments?
• RQ3 Can the proposed classifier significantly improve the input data quality for CDSS in mechanical ventilation?
• RQ4 What methodological improvements does this approach offer over prior work in ventilator mode detection?
• RQ5 How robust is the model to software or sensor errors that disrupt data flow?

Key findings

• The machine learning classifier achieves an average F1-score of 97.52% across five major ventilator modes in the USA.
• The model demonstrates high robustness to missing data, maintaining performance even when up to 30% of input features are corrupted.
• The approach outperforms previous methods in both accuracy and reliability under real-world data quality challenges.
• The system enables CDSS to make informed, mode-aware recommendations, significantly improving clinical decision support quality.
• The model’s per-breath classification capability allows for real-time adaptation of CDSS interventions based on current ventilator mode.
• The methodological improvements include better feature representation and enhanced handling of data gaps, setting a new benchmark for ventilator mode detection.