ESTRO meets Asia 2024 - Abstract Book

S161

Interdisciplinary – Head & neck

ESTRO meets Asia 2024

Material/Methods:

RAIRI was tested in 2,148 patients across training, internal validation, external validation, and randomized controlled trial (RCT) cohorts from three academic cancer centers. Bayesian joint model was employed for integrative prediction. Prognostic accuracy was evaluated using calibration, concordance indices (C-indices), and areas under the curve (AUCs). RAIRI’s performances of predicting AC benefit were examined in patients from two RCTs designed to assess AC’s benefit in high-risk stage III/IVA NPC.

Results:

RAIRI incorporates six pretreatment characteristics (age, T-stage, N-stage, cfEBV-DNA, lactate dehydrogenase, and central-nodal-necrosis), longitudinal cfEBV-DNA, and tumor regression measurements. In the training cohort, RAIRI demonstrated accurate calibration and high prognostic accuracy (5-year: C-index 0.849, AUC 0.895), significantly superior to the conventional models. The internal validation, external validation, and RCT cohorts confirmed RAIR’s prognostic accuracy with 5-year C-indices and AUCs all over 0.8. In the RCT cohort, RAIRI identified approximately 70% of low-risk patients that did not benefit from AC; however, the high-risk patients experiencing substantial benefits from AC versus observation (5-year PFS, 58.7% vs. 22.8%; HR=0.40; 95% CI, 0.22– 0.73). Predictive performance of RAIRI for AC benefit was consistent across various clinical subsets and beyond post-radiotherapy cfEBV-DNA.