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.
Purpose:
Dysphagia is a common yet often overlooked complication in Chronic Obstructive Pulmonary Disease (COPD) patients, who are prone due to abnormal breathing patterns, impaired airway protection, and generalized frailty. This not only predisposes them to aspiration pneumonia but also serves as a key trigger for acute exacerbations of COPD, substantially increasing the risk of adverse outcomes. Early identification of dysphagia is essential for improving prognosis in COPD. However, research on early detection is limited, particularly regarding machine learning prediction. This study aimed to develop and validate a machine learning based risk assessment model for dysphagia in COPD and deploy it as a user-friendly web-based clinical tool to assist clinicians in risk identification and early intervention.
Patients And Methods:
Retrospective medical records from 710 COPD patients admitted between February 2025 and January 2026 were analyzed. Swallowing function was assessed using the Water-Swallowing Test. Univariate and multivariate logistic regression identified independent risk factors, which were used to develop and compare eight machine learning models. Model performance was evaluated using ROC, calibration, and decision curves with Bootstrap internal validation. Key variables were interpreted via Shapley Additive Explanations, and the final model was deployed online.
Results:
Dysphagia prevalence was 29.3%. Multivariate regression identified five key risk factors: disease duration, BMI, history of tracheal intubation, muscle strength, and the modified Medical Research Council (mMRC) score for dyspnea severity. Among eight machine learning models, the XGBoost model showed the best performance in the training set (AUC 0.921, 95% CI 0.901-0.940) and demonstrated good calibration and highest clinical net benefit. The model was deployed online (https://dysphagiamodel.shinyapps.io/COPD-DP/).
Conclusion:
We developed and validated an online machine learning-based dysphagia risk assessment tool for COPD, demonstrating discrimination, calibration, and clinical utility for risk stratification and clinical decision-making.