S999
Clinical – Paediatric tumours
ESTRO 2026
Keywords: Ototoxicity, Pediatric RT, Multicenter
Digital Poster Highlight 68 Hearing toxicity after pediatric cranial radiotherapy: results from a French multicenter cohort William Gehin 1 , Luc Ollivier 2 , Emmanuel Jouglar 3 , Claire Dossun 4 , Maria Jolnerovski 1 , Valérie Bernier 1 1 Radiation therapy, Institut de Cancérologie de Lorraine, Vandoeuvre-lès-Nancy, France. 2 Radiation therapy, Institut de Cancérologie de l'Ouest, Saint- Herblain, France. 3 Radiation therapy, Institut Curie, Paris, France. 4 Radiation therapy, Institut de Cancérologie Strasbourg-Europe, Strasbourg, France Purpose/Objective: No multicenter study has validated cochlear dose constraints for hearing preservation in pediatric brain tumor patients in a real-world setting. We aimed to provide the first national, multicenter validation of the 35 Gy mean cochlear dose threshold proposed by the PENTEC review [1] and assess its applicability for treatment planning and survivorship guidelines. Material/Methods: We retrospectively analyzed 88 children treated with cranial radiotherapy (2014–2024) across four French pediatric RT centers in the national PediaRT registry. All had baseline and ≥ 2 post-RT audiograms graded per the Chang scale. Mean (Dmean) and minimum (Dmin) cochlear doses were extracted, normalized to EQD2, and evaluated using Kaplan–Meier and Cox regression models. Sensorineural hearing loss (SNHL) was defined as Chang grade ≥ 2. Results: Over a median follow-up of 36 months, 17 patients (19.3%) developed SNHL. In multivariate analysis, both Dmean ≥ 35 Gy (HR = 3.90; 95% CI: 1.24–12.23; p = 0.020) and cisplatin exposure (HR = 3.85; 95% CI: 1.40– 10.64; p = 0.009) independently predicted SNHL. The 5-year cumulative incidence reached 43.4% for Dmean ≥ 35 Gy vs. 13.1% for < 35 Gy (p = 0.0014) and 51.6% with cisplatin vs. 12.4% without (p < 0.001). A monotonic dose–severity relationship was observed (p = 0.0011). Conclusion: This study provides the first multicenter, national validation of the 35 Gy mean cochlear dose threshold within a real-world setting and highlights cisplatin exposure as a dominant risk factor. These findings support incorporating strict cochlear dose minimization into RT planning and systematic long- term audiological follow-up in survivorship care. References: 1. Murphy B, Jackson A, Bass JK, et al. Modeling the risk of hearing loss from radiation therapy in childhood cancer survivors: a PENTEC comprehensive review. Int
Digital Poster 215
Risk of radiation-induced brain and brainstem injury after proton beam therapy for pediatric ependymoma: Prognostic modeling based on DVH parameters SungChul Park 1 , Yusuke Demizu 1 , Tianyuan Wang 2 , Tomohiro Yamashita 2 , Ryosuke Bessho 1 , Nor Shazrina 1 , Nobuyoshi Fukumitsu 1 , Takeshi Suzuki 3 , Suguru Uemura 4 , Daiichiro Hasegawa 4 , Junji Koyama 5 , Atsufumi Kawamura 5 , Toshinori Soejima 6 , Sunao Tokumaru 1 1 Radiation oncology, Kobe Proton Center, Kobe, Japan. 2 Radiation Physics, Kobe Proton Center, Kobe, Japan. 3 Anesthesiology, Kobe Proton Center, Kobe, Japan. 4 Hematology and Oncology, Kobe Children's Hospital, Kobe, Japan. 5 Neurosurgery, Kobe Children's Hospital, Kobe, Japan. 6 Radiation oncology, Kobe Minimally Invasive Cancer Center, Kobe, Japan Purpose/Objective: To clarify dose–response relationships and explore the feasibility of predictive modelling for radiation-induced brain and brainstem injury (RIBI) in children with intracranial ependymoma after proton beam therapy (PBT), focusing on dose–volume histogram (DVH) parameters. Material/Methods: We retrospectively analyzed 24 patients aged ≤ 18 years with newly diagnosed intracranial ependymoma treated with adjuvant PBT at Kobe Proton Center between 2018–2024. Inclusion required ≥ 6 months’ follow-up and no salvage therapy. RIBI (brain, brainstem) was graded according to CTCAE v5.0 (CNS necrosis) and dichotomized as grade 0 vs ≥ 1. To address DVH collinearity, features were clustered, and stability selection with Elastic-net (1,000 bootstraps; threshold ≥ 0.70) was performed. Candidate clinical factors were pre-screened by univariate logistic regression (p<0.20). Multivariable models were then fitted with clinical and DVH predictors. Generalizability was tested using nested cross-validation. Results: The cohort included 24 patients (median age 4.5 years, range 1–17; 10 male/14 female; WHO grade II/III = 8/16; supratentorial/infratentorial = 13/11). Brain injury occurred in 7/24 (29.2%) and brainstem injury in 4/24 (16.7%). **Most events were low grade: brain—6 grade 1 and 1 grade 2 (no grade ≥ 3); brainstem—all 4 grade 1.** Nested cross-validation showed no generalizable predictive performance (AUC ≈ 0.50 brain, ≈ 0.00 brainstem). Stability selection identified V50 (0.755) and V30 (0.718) as robust brain predictors, whereas V60 (0.273) and D2% (0.170) were below
J Radiat Oncol Biol Phys. 2023; doi:10.1016/j.ijrobp.2023.08.016
Made with FlippingBook - Share PDF online