A XGBoost model, incorporating 12 radiomic features and four clinical variables (primary tumor site [oral cavity], TNM, age, and smoking), emerged as the most predictive for 5-year progression. This model yielded an AUC of 0.74, sensitivity of 0.53, specificity of 0.81, and accuracy of 0.66. Our efforts focus now on increasing the sample size to validate this model and assess progression-free survival outcomes, aiming to provide clinicians with a risk score for more precise patient categorization based on diagnostic CT scans.
Bruixola et al. J Clin Oncol 41, 2023 (suppl 16; abstr 6076)