A Machine Learning Model for Recurrence Prediction in Head and Neck Cancer Based on Quality-of-Life

Sunday, October 12, 2025

9:54 AM - 10:00 AM EDT

Location: 127

Disclosure(s):

Naomi C. Wang, MD:

faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

Introduction: Recurrence after head and neck cancer (HNC) treatment is common and associated with poor survival and diminished quality of life (QoL). This study aimed to develop a machine learning model using only patient-reported QoL data to predict recurrence within two years post-treatment.

Methods: This retrospective cohort study analyzed prospectively collected QoL data from HNC patients at an academic referral center between 2018-2024. Eligible patients had newly diagnosed squamous cell carcinoma of the oral cavity, oropharynx, hypopharynx, larynx, or cervical lymph nodes with an unknown primary site. QoL was assessed at baseline as well as three months and six months post-treatment using the University of Washington (UW)-QoL questionnaire. Recurrence was defined as either locoregional or distant recurrence. Patients who recurred before six months or had unknown recurrence status at two years were excluded. We developed a binary classification model using XGBoost, with feature selection based on Shapley values. Model performance was evaluated using five-fold cross-validation, with accuracy and area under the receiver operating characteristic curve (AUC) as primary metrics.

Results: The final cohort included 248 patients, with 66 (26.6%) experiencing recurrence between six months and two years post-treatment. The model achieved a mean cross-validation accuracy of 77.0% ± 2.9% and an AUC of 0.80 ± 0.05. Key predictive features from the UW-QoL questionnaire included the swallowing score three months after treatment, rate of change in recreation score from baseline to three months post-treatment, and both the change and rate of change in saliva score over the same period.

Conclusions: Pre-treatment and early post-treatment patient-reported QoL data can predict recurrence risk in HNC, which supports its potential as a patient-centered, non-invasive tool for identifying high-risk patients and highlights the importance of ongoing symptom monitoring in risk assessment.

Presenting Author(s)

Maria Feucht, MD

Resident
Indiana University School of Medicine
Indianapolis, IN, United States

Disclosure information not submitted.

Co-Author(s)

AN

Afaf Nabeeha, MS

Kansas City University College of Osteopathic Medicine, United States

Disclosure information not submitted.
Naomi C. Wang, MD

Resident
University of Kansas Medical Center Department of Otolaryngology– Head & Neck Surgery
Kansas City, MO, United States

Disclosure(s):
faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.
Uma R. Ramesh, MD

Clinical Research Fellow
University of Kansas Medical Center
kansas city, MO, United States

Disclosure information not submitted.
Rahul Alapati, MD

Otolaryngology-Head and Neck Surgery Resident
University of Kansas Medical Center
Kansas City, KS, United States

Disclosure information not submitted.
Ameen Amanian, MD FRCSC MSE

Clinical Instructor
University of Kansas Medical Center
Kelowna, BC, Canada

Disclosure information not submitted.

Senior Author(s)

Andres M. Bur, MD

Associate Professor
Kansas University Physicians Inc
Kansas City, Kansas, United States

Disclosure information not submitted.