Integrating lymph node ratio into personalized radiotherapy for oral cavity squamous cell carcinoma.

Abstract

Purpose: The use of postoperative radiotherapy (PORT) in patients with oral squamous cell carcinoma (OCSCC) lacks clear boundaries due to the non-negligible toxicity accompanying its remarkable cancer-killing effect. This study aims at validating the ability of deep learning models to develop individualized PORT recommendations for patients with OCSCC and quantifying the impact of patient characteristics on treatment selection.

Methods: Participants were categorized into two groups based on alignment between model-recommended and actual treatment regimens, with their overall survival compared. Inverse probability treatment weighting was used to reduce bias, and a mixed-effects multivariate linear regression illustrated how baseline characteristics influenced PORT selection.

Results: 4990 patients with OCSCC met the inclusion criteria. Deep Survival regression with Mixture Effects (DSME) demonstrated the best performance among all the models and National Comprehensive Cancer Network guidelines. The efficacy of PORT is enhanced as the lymph node ratio (LNR) increases. Similar enhancements in efficacy are observed in patients with advanced age, large tumors, multiple positive lymph nodes, tongue involvement, and stage IVA. Early-stage (stage 0-II) OCSCC may safely omit PORT.

Conclusions: This is the first study to incorporate LNR as a tumor character to make personalized recommendations for patients. DSME can effectively identify potential beneficiaries of PORT and provide quantifiable survival benefits.