Two-Stage Stacked ML Model Stratifies Dysphagia Risk in Head and Neck Cancer Survivors Without Imaging

Forum topic · 小凯 · 2026-07-28

Summary

A 2026 arXiv paper (2607.22514) by Siyuan Zhao et al. proposes an interpretable, two-stage stacked machine learning framework that predicts dysphagia risk in head and neck cancer (HNC) survivors using patient-reported outcomes (PROs) and structured clinical variables, eliminating the need for videofluoroscopic imaging. Dysphagia is a debilitating late effect of HNC treatment, but definitive assessment via CTCAE-DIGEST videofluoroscopy requires specialized equipment, trained personnel, and burdensome patient logistics. PROs such as the MDADI questionnaire are low-cost and scalable, yet a clinical framework for converting them into actionable interventions has been lacking. The proposed model quantifies the independent contributions of patient-reported symptoms and clinical factors within a unified risk assessment. Key findings show that individual MDADI item responses carry predictive information beyond global summary scores, and explainability analysis surfaces symptom patterns and clinical risk factors associated with swallowing injury. The results support structured PRO-clinical integration as a practical, imaging-free dysphagia risk stratification approach for HNC survivorship care.

Paper Overview

Field: Machine Learning

Authors: Siyuan Zhao, Eric Ababio Anyimadu, Zachary G. Brumm, Yue Ma, Clifton David Fuller, Xinhua Zhang, G. Elisabeta Marai, Guadalupe Canahuate

Published: 2026-07-24

arXiv: 2607.22514

Summary

Dysphagia is a debilitating late effect of head and neck cancer (HNC) treatment, yet timely identification of at-risk patients remains challenging in survivorship care. Definitive assessment relies on videofluoroscopic imaging, as captured by the Dynamic Imaging Grade of Swallowing Toxicity (CTCAE-DIGEST), which, while validated, requires specialized equipment, trained personnel, and significant patient burden, limiting its routine use in surveillance.

Patient-reported outcomes (PROs), by contrast, are low-cost, scalable, and easily collected at any clinical encounter, making them an attractive alternative signal for identifying patients who may warrant further evaluation. However, a clear clinical framework for translating PRO responses into actionable interventions is still evolving. In particular, uncertainty remains about when patient-reported symptom burden should prompt escalation of care.

To address this gap, the authors build a single-visit PRO-clinical prediction framework and introduce a clinically interpretable two-stage stacked model that predicts swallowing-injury risk using PRO responses and structured clinical variables, without videofluoroscopic imaging.

Key Findings

The proposed framework quantifies the independent contributions of patient-reported symptoms and clinical factors within a unified, interpretable risk assessment model.
Individual MDADI responses contain predictive information beyond what composite or global summary scores capture.
Explainability analysis reveals symptom patterns and clinical risk factors associated with swallowing injury.

Conclusion

These results collectively support structured PRO-clinical integration as a practical, imaging-free approach to dysphagia risk stratification in HNC survivorship care.

--- *Auto-collected on 2026-07-28*

Two-Stage Stacked ML Model Stratifies Dysphagia Risk in Head and Neck Cancer Survivors Without Imaging

Summary

Paper Overview

Summary

Key Findings

Conclusion

Tags