Machine-learning models may improve postoperative risk stratification and help guide decisions about adjuvant therapy in oral cavity squamous cell carcinoma (OCSCC), according to a study published in JCO Precision Oncology.
Researchers developed and internally validated several machine-learning survival models using data from the National Cancer Database, including 35,625 patients with OCSCC treated with primary surgery. A surgery-alone cohort of 18,543 patients was used to train and test the models, which generated postoperative risk scores to stratify patients into low-, intermediate-, and high-risk groups.
CLINICAL SUMMARY
What was examined
Researchers developed and internally validated machine-learning survival models to improve postoperative risk stratification in oral cavity squamous cell carcinoma and examined associations between predicted risk groups and outcomes following adjuvant therapy.
Key findings
-
Machine-learning models were developed using data from 35,625 patients in the National Cancer Database.
-
A surgery-alone cohort of 18,543 patients was used to train and test the models, which stratified patients into low-, intermediate-, and high-risk groups.
-
Associations between adjuvant radiotherapy or chemoradiotherapy and overall survival differed across machine-learning–derived risk categories.
Clinical implications
-
Machine-learning models may help refine postoperative risk assessment in oral cavity squamous cell carcinoma.
-
AI-based risk stratification could support more personalised decisions regarding adjuvant therapy.
-
Prospective validation will be required before these tools can be incorporated into routine clinical practice.
Three algorithms—DeepSurv, neural multitask logistic regression,n, and random survival forest—were evaluated for their ability to predict survival outcomes and classify patients according to postoperative risk.
The investigators then examined whether these machine-learning–derived risk groups were associated with different outcomes following adjuvant therapy. Associations between adjuvant radiotherapy or chemoradiotherapy and overall survival differed across the predicted risk categories.
Featured Event
These findings suggest that machine-learning–based risk stratification may help identify patients with differing risk profiles following surgery and could potentially inform postoperative treatment decision-making.
The authors note that although the models demonstrated promising prognostic performance, prospective validation and integration into clinical workflows will be necessary before routine clinical use.
Paper: Costantino, A, et al. Machine Learning–Driven Risk Stratification and Adjuvant Treatment Guidance in Oral Cavity Cancer. JCO Precis Oncol 10, e2500914 (2026). Access online here.
