Machine learning for survival outcome in head and neck squamous cell carcinoma: a multicenter validation study

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Head and neck squamous cell carcinoma (HNSCC) often presents at late stages, leading to poor prognoses.
• Machine learning (ML) models offer potential for personalized treatment planning but require robust external validation.

Purpose of the Study:

• To develop and externally validate a machine learning model for predicting overall survival (OS) in HNSCC patients.
• To integrate clinicopathological and treatment-related factors for enhanced prognostic accuracy.
• To identify key prognostic parameters using permutation feature importance (PFI).

Main Methods:

• A voting ensemble ML model was developed using data from the US SEER program (n=40,164).
• The model was externally validated using multicenter data from Germany (n=3950) and Sweden (n=323).
• Permutation feature importance (PFI) was employed to assess the prognostic significance of input variables.

Main Results:

• The ML model achieved an AUC of 0.76 and 70.0% accuracy in the SEER cohort.
• External validation yielded AUCs of 0.68 (Germany) and 0.76 (Sweden), demonstrating generalizability with performance variations.
• PFI identified age at diagnosis, T stage, tumor site, marital status, and surgical treatment as crucial predictors of OS.

Conclusions:

• The developed ML model shows promise for predicting HNSCC patient survival and can aid in risk-based therapeutic decisions.
• External geographic validation is crucial for assessing model reproducibility and generalizability, even if performance metrics vary.
• While independent validation is ideal, data privacy concerns can pose challenges in integrating it into the ML development pipeline.