Survival and data-driven phenotypes in head and neck cancer

Area of Science:

• Oncology
• Bioinformatics
• Data Science

Background:

• Head and neck cancer (HNC) is a major global health concern with a 5-year survival rate of 50-60%.
• Existing survival prediction models for HNC require refinement due to unexpected mortality patterns.
• Identifying distinct patient subgroups is crucial for improving prognostic accuracy.

Purpose of the Study:

• To discover novel, data-driven phenotypes in head and neck cancer patients.
• To identify clinical and demographic features associated with these phenotypes.
• To predict overall survival using a deep survival clustering model.

Main Methods:

• Utilized a retrospective cohort of 1341 HNC patients from Helsinki University Hospital.
• Employed the deep survival clustering model VaDeSC for phenotype identification and survival prediction.
• Analyzed pre-treatment clinical and demographic data from electronic health records.

Main Results:

• Identified six previously unrecognized HNC patient phenotypes with distinct survival trajectories.
• Key associated features include BMI, sleep apnoea, TNM stage, tumour site, treatment intention, gender, and age.
• VaDeSC achieved high predictive accuracy (C-index 0.895 training, 0.782 test) and clustering performance.

Conclusions:

• Clustering pre-treatment clinical data reveals interpretable HNC phenotypes and enables accurate individualized survival predictions.
• This data-driven approach offers significant potential for discovering novel phenotypes in various diseases.
• The findings support personalized medicine strategies for head and neck cancer.