S2441
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
intercepts. Predictors that retained in most folds formed the final model. Model performance and generalizability were evaluated using AUC, calibration, and I2. Results: AF occurred in 15% (n=250) of the patients. The pulmonary veins and left atrium were summarized into four dose patterns using PCA, since they were identified in multivariable analysis. The first dose pattern (PC1), representing overall radiation intensity to both structures, predicted AF. Other predictors included sex, age, pulmonary veins size, and surgery (Table 1). For a median patient (male, 67 years, pulmonary vein=22.15 cm3), predicted AF risk across the observed PC1 ranged from 13.0–23.5% with surgery and 3.0–6.0% without. Across centers, the model showed fair discrimination (pooled AUC:0.68 (95%CI 0.60–0.77; I2=77%)). Calibration across centers was good, with a pooled slope of 0.95 (95%CI 0.48– 1.43; I2=51%) and calibration-in-the-large of –0.07 (95%CI –0.27–0.14; I2=10%), close to the ideal values of 1 and 0, respectively. Figure 1 shows center-specific results.
Center Utrecht, Utrecht University, Utrecht, Netherlands. 4 Department of Radiation Oncology, Amsterdam UMC, Amsterdam, Netherlands. 5 Cancer Center Amsterdam, Cancer treatment and Quality of life, Amsterdam, Netherlands. 6 Department of Medical Oncology, Amsterdam UMC at the University of Amsterdam, Amsterdam, Netherlands. 7 Department of Radiation Oncology, University Medical Center Utrecht, Utrecht, Netherlands. 8 Department of Radiation Oncology, Erasmus MC Cancer Institute, Rotterdam, Netherlands. 9 Department of Radiation Oncology, Radboud University Nijmegen Medical Centre, Nijmegen, Netherlands. 10 Department of Surgery, Radboud University Nijmegen Medical Centre, Nijmegen, Netherlands. 11 Department of Radiation Oncology, Radiotherapiegroep, Arnhem, Netherlands. 12 Department of Surgery, Zuyderland Medical Center, Heerlen, Netherlands. 13 Department of Surgery, University Medical Center Groningen, Groningen, Netherlands. 14 Department of Research and Development, Netherlands Comprehensive Cancer Organisation, Utrecht, Netherlands. 15 Department of Radiation Oncology (Maastro), GROW School for Oncology and Reproduction, Maastricht University Medical Centre+, Maastricht, Netherlands Purpose/Objective: Atrial fibrillation (AF) occurs in 10-20% of patients following neoadjuvant chemoradiotherapy (nCRT) plus surgery. While radiation dose has been significantly associated with AF1, validated prediction models are limited. Normal Tissue Complication Probability (NTCP) models provide a robust quantitative framework for guiding radiotherapy treatment planning to limit the risk of treatment-related complications. This study aimed to develop and validate an NTCP model for AF across multiple centers. Material/Methods: Data from 1,704 esophageal cancer patients treated with nCRT and planned for esophagectomy (2015- 2021) were collected in collaboration with the Netherlands Cancer Registry, Dutch Upper GI Cancer Audit, and six Dutch treatment centers. This data was used to develop and validate a multivariable logistic regression model for AF grade ≥ 2 (CTCAE v5) within six months after nCRT. Missing data was handled using multiple imputation. Dose-volume histogram (DVH) data for cardiac substructures (atria, ventricles, pulmonary veins, and sinus node) were analyzed. Structures with p<0.1 in multivariable analysis, corrected for confounders, entered model development. To reduce multicollinearity, principal component analysis (PCA) derived independent dose patterns summarizing the complete dose distribution. Predictor selection used backward selection, optimizing predictive performance (Brier score), in 10- fold internal-external cross-validation with stratified
Made with FlippingBook - Share PDF online