S556
Clinical – Head & neck
ESTRO 2026
Poster Discussion 1132 Predictive modeling of outcomes in salivary gland cancers using machine learning: simulated prospective validation with the RTOG 1008 trial Federico Lorenzo 1,2 , Agustin Rosich 1,2 , Jesica Lell 1,2 , Sergio Aguiar 1 , Valentina Ferreira 1 , Karina Ochandorena 1,2 , Eduardo Larrinaga 1 , Aldo Quarneti 1 1 Radiotherapy, RT International Institute, Montevideo, Uruguay. 2 Radiotherapy, Unidad Academica de Radioterapia, Montevideo, Uruguay Purpose/Objective: To estimate, in a simulated prospective framework mirroring RTOG 1008, whether concurrent chemoradiotherapy (CRT) versus PORT alone improves overall survival (OS) and progression-free survival (PFS), and to characterize toxicity and locoregional control, thereby identifying conditions under which CRT might be justified in high-risk SGC. Material/Methods: A Gradient Boosting Machine trained on 4,600 SEER patients with SGC generated a virtual high-risk cohort aligned with RTOG 1008 eligibility and contemporary risk distributions (age 30–70; margin status; nodal involvement; perineural invasion). Patients were randomized 1:1 to PORT or PORT + weekly cisplatin. OS and PFS were modeled with proportional hazards; an a priori HR of 0.74 for OS in favor of CRT guided power calculations. Two hundred iterations (100 patients/arm each) yielded Kaplan–Meier estimates, log-rank tests, and Cox HRs. Grade 3–4 mucositis, xerostomia, and neutropenia were assigned from published rates (5–15%). Sample size requirements for detecting the prespecified HR with 80% power and α =0.05 were computed. Simulated outcomes were contrasted with contemporary PORT/CRT series. Results: Across 252 simulated patients randomized 1:1, CRT did not significantly improve OS versus PORT: median OS 33.5 vs 28.7 months; HR 0.85 (95% CI 0.64–1.14), p=0.36. In contrast, PFS favored CRT: HR 0.74 (95% CI 0.56–0.96), p=0.034, accompanied by higher 2–5-year locoregional control (82.5% vs 66.7%), p=0.006. Acute and late grade 3–4 toxicities in the CRT arm were 37% and 15%, respectively, without a statistically significant excess versus PORT. The prevailing failure pattern was distant metastasis, consistent with observational data indicating that intensified local therapy improves control without translating into a survival advantage. Conclusion: In a machine-learning simulation of RTOG 1008, concurrent cisplatin with PORT improved PFS and locoregional control but did not significantly prolong OS, with acceptable toxicity. These findings support PORT as the adjuvant standard and suggest reserving CRT for selected subgroups or trial settings until
Conclusion: Objective MIO explained only limited portion of variability in PROs. Further investigation is suggested to explore additional contributors and/or combination effects from other complications on subjective experience. References: 1. Basch E, lasonos A, McDonough T, Barz A, Culkin A, Kris MG, et al. Patient versus clinician symptom reporting using the National Cancer Institute Common Terminology Criteria for Adverse Events: results of a questionnaire-based study. Lancet Oncol. 2006;7(11):P903-909.2. Teguh DN, Levendag PC, Voet P, van der Est H, Noever I, de Kruijf W, et al. Trismus in patients with oropharyngeal cancer: relationship with dose in structures of mastication apparatus. Head Neck. 2008;30(5):622–630.3. Saini J, Bakshi J, Panda NK, Sharma M, Vir D, Goyal AK. Cut-off points to classify numeric values of quality of life into normal, mild, moderate, and severe categories: an update for EORTC-QLQ-H&N35. Egypt. J. Otolaryngol. 2024;40(83):1-8. Keywords: Trismus, Patient-reported outcomes, NPC
Made with FlippingBook - Share PDF online