Development and Validation of an Explainable Machine Learning Model for Identification of Dysphagia in Patients with
Chiteng Zhou1, Shuangwei Hong2, Jun Fang3
1School of Medicine, Jinhua University of Vocational Technology, Jinhua, Zhejiang, People's Republic of China.
Summary
Dysphagia is common in Chronic Obstructive Pulmonary Disease (COPD). This study developed a machine learning tool to predict dysphagia risk in COPD patients, aiding early detection and intervention.
Area of Science:
- Pulmonary Medicine
- Clinical Informatics
- Gerontology
Background:
- Dysphagia (swallowing difficulty) is a frequent complication in COPD patients.
- It increases risks of aspiration pneumonia and COPD exacerbations.
- Early dysphagia detection is crucial for improving COPD patient outcomes.
Purpose of the Study:
- To develop and validate a machine learning (ML) model for predicting dysphagia risk in COPD patients.
- To deploy the ML model as a web-based tool for clinical use.
- To facilitate early identification and intervention for dysphagia in COPD.
Main Methods:
- Retrospective analysis of 710 COPD patient records.
- Swallowing function assessed via Water-Swallowing Test.
- XGBoost ML model developed using logistic regression risk factors, validated with ROC, calibration, and decision curves.
Main Results:
- Dysphagia prevalence was 29.3% in the study cohort.
- Key risk factors identified: disease duration, BMI, tracheal intubation history, muscle strength, mMRC score.
- The XGBoost model achieved high performance (AUC 0.921) and was deployed online.
Conclusions:
- An online ML tool for dysphagia risk assessment in COPD was successfully developed and validated.
- The tool demonstrates discrimination, calibration, and clinical utility for risk stratification.
- This facilitates improved clinical decision-making and early intervention for dysphagia in COPD.