S749
Clinical - Lung
ESTRO 2026
Purpose/Objective: Survival in stage III NSCLC has improved with adjuvant durvalumab after chemoradiotherapy (CRT), but occurrence of CRT-related toxicities often hinders initiation of immunotherapy. To optimize treatment planning and minimize toxicities, we aimed to develop a pre-CRT prediction model to identify patients likely to qualify for adjuvant immunotherapy. Material/Methods: Data from the real-world prospective Standard Follow- up Program (SFP) were used, focusing on stage III NSCLC patients treated with CRT between Jan-2017 and Jan-2024, who were eligible for adjuvant durvalumab a priori. A logistic regression-based model was developed to predict immunotherapy eligibility, evaluating gender, age, smoking (current vs previous/never), cardiac/pulmonary comorbidities, tumour morphology, T- and N-stage, concurrent chemotherapy, WHO performance score, radiotherapy technique (Figure 1) and lung/heart dose-volume metrics as candidate predictors. To address multicollinearity, the DVH metric with the lowest AIC was retained before backward selection (Wald P < 0.05). Model performance was assessed via AUC, calibration plots, and internal bootstrap validation.Figure 1. VMAT and IMPT plan for a representative patient
CCRT. Lacunarity showed significant increases in both RT alone group and CCRT group, post-RT compared to pre-RT, with median differences of 0.009 (p=0.024) and 0.016 (p < 0.001), respectively, as determined by Wilcoxon signed rank tests. In the best-performing RFM, the features with the highest importance were pre-RT MSTFD, and lung V10 in RT alone group and lung V20, BoxFD difference, lacunarity difference, lung V5, and pre-RT MSTFD in CCRT group. The prediction performances of the decision tree models were as follows: accuracy 0.87, area under the receiver operating curve (AUROC) 0.83, and F1 score 0.92 in RT alone group, and accuracy 0.80, AUROC 0.85, and F1 score 0.76 in CCRT group, respectively. The models also showed p-values < 0.05 for the hypothesis that the accuracy is greater than the no information rate. Conclusion: Fractal imaging biomarkers showed promising prognostic value in predicting grade ≥ 2 RP in patients with NSCLC. The proposed decision tree model may serve as a practical tool for early identification of high- risk patients, offering potential guidance for personalized treatment strategies and future research. References: Hwang, J.; Kim, H. Preliminary Results of Developing Imaging Complexity Biomarkers for the Incidence of Severe Radiation Pneumonitis Following Radiotherapy in Non-Small Cell Lung Cancer Patients with Underlying Idiopathic Pulmonary Fibrosis. Life (Basel) 2024, 14, doi:10.3390/life14070897.Hwang, J.; Kim, H. Development of Imaging Complexity Biomarkers for Prediction of Symptomatic Radiation Pneumonitis in Patients with Non-Small Cell Lung Cancer, Focusing on Underlying Lung Disease. Life (Basel) 2024, 14, doi:10.3390/life14111497.Wang, S.; Xu, D.; Xiao, L.; Liu, B.; Yuan, X. Radiation-induced lung injury: from mechanism to prognosis and drug therapy. Radiat Oncol 2025, 20, 39, doi:10.1186/s13014-025-02617-8. Keywords: radiation pneumonitis, imaging biomarker Predicting eligibility for adjuvant immunotherapy following chemoradiotherapy in stage III non- small-cell lung cancer Arno C. Hessels 1 , Gibson C. Ugwu 2 , T. J.N. Hiltermann 3 , G. H. de Bock 2 , Martijn W. Heymans 2 , Johannes A. Langendijk 1 , Behrooz Z. Alizadeh 2 , Robin Wijsman 1 1 Radiation Oncology, University Medical Center Groningen, Groningen, Netherlands. 2 Epidemiology, University Medical Center Groningen, Groningen, Netherlands. 3 Pulmonary Medicine, University Medical Center Groningen, Groningen, Netherlands Digital Poster Highlight 61
For this 68-year-old male ex-smoker, predicted immunotherapy probability was 36% with the VMAT plan (A), and 74% with the IMPT plan (B). Green: 95% isodose line. Orange: 3 Gy isodose line.
Made with FlippingBook - Share PDF online