Machine learning-based treatment outcome prediction in head and neck cancer using integrated noninvasive diagnostics

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Accurate prediction of treatment outcomes is vital for personalized care in head and neck squamous cell carcinoma (HNSCC).
• Assessing long-term outcomes like enteral nutrition dependence is crucial for patient counseling and resource management.
• Machine learning offers potential for predicting complex oncological outcomes.

Purpose of the Study:

• To predict one-year survival and feeding tube dependence in surgically treated HNSCC patients.
• To evaluate the predictive performance of clinical data, blood markers, and radiomic features.
• To explore the utility of machine learning models for outcome prediction in HNSCC.

Main Methods:

• Retrospective analysis of 558 surgically treated HNSCC patients.
• Collection of baseline clinical data, blood markers, and MRI-based radiomic features.
• Training random forest classifiers to predict one-year survival and feeding tube dependence, with SHAP values for explainability.

Main Results:

• Clinical data demonstrated the highest predictive performance for one-year survival (AUC=0.75).
• Blood markers and radiomic features showed moderate predictive power for survival, with no improvement upon multimodal integration.
• Postsurgical treatment information was the strongest predictor for feeding tube dependence, though it had low predictive value for survival.

Conclusions:

• Clinical data is the most significant predictor of one-year survival in surgically treated HNSCC.
• Postsurgical treatment information is critical for predicting feeding tube dependence.
• Further research is needed to explore the potential complementarity of multimodal data integration for improving predictive models.