Related Experiment Video
Updated: Sep 11, 2025
Adapting Human Videofluoroscopic Swallow Study Methods to Detect and Characterize Dysphagia in Murine Disease Models
Published on: March 1, 2015
Exploring explainable machine learning techniques to aid dysphagia risk identification: A feasibility study.
Melanie L McIntyre1, Yuxi Liu2, Joanne Murray3
1Swallowing Neurorehabilitation Research Lab, Caring Futures Institute, College of Nursing and Health Sciences, Flinders University, GPO Box 2100, Adelaide, SA, 5001, Australia; Bendigo Health, Department of Speech Pathology, GPO Box 126, Bendigo, VIC, 3552, Australia.
Machine learning can identify dysphagia (swallowing difficulty) risk in intensive care unit (ICU) patients requiring mechanical ventilation. Key factors include ventilation duration, age, and admission type, enabling personalized risk assessment.
Area of Science:
- Medical Informatics
- Artificial Intelligence in Healthcare
- Clinical Decision Support
Background:
- Machine learning (ML) offers advanced capabilities for identifying intricate risk patterns within extensive healthcare datasets.
- This study investigates the feasibility and proof of concept for applying ML techniques to identify dysphagia risk in intensive care unit (ICU) patients requiring endotracheal intubation.
Purpose of the Study:
- To explore the methodological feasibility of developing ML models for dysphagia risk identification.
- To establish a proof of concept for ML-driven dysphagia risk assessment in critically ill adult patients.
Main Methods:
- A cohort study linking two large healthcare databases using deterministic logic.
- Exploration of various ML model candidates for dysphagia risk prediction.
- Utilized SHapley Additive exPlanation (SHAP) values to interpret model decision-making.
Main Results:
- Included 59,811 patients from 42 sites who received invasive mechanical ventilation in an ICU.
- Identified the top five risk factors for dysphagia: duration of mechanical ventilation, age, cardiac admission, neurological admission, and APACHE III score.
- Demonstrated ML's ability to discern complex risk profiles.
Conclusions:
- ML shows significant promise for dynamic, individualized dysphagia risk screening in ICUs.
- Proposed future clinical integration of ML models for more accurate patient-specific risk assessment.
- Emphasized the need to move beyond cohort means to individualized risk factor analysis.
