S1066
Clinical – Upper GI
ESTRO 2026
patients, it identified a low-risk cohort with a 2-year survival as high as 71% (median EDIC 4.2Gy), in contrast to the predicted high-risk group at only 44% (median EDIC 5.6Gy). The stratification was also pronounced for SCC, separating a low-risk group with 87% survival (median EDIC 3.0Gy) from a high-risk group with 48% survival (median EDIC 5.6Gy).
University Medical Center Utrecht, Utrecht, Netherlands. 7 Department of Radiation Oncology, Erasmus MC Cancer Institute, Rotterdam, Netherlands. 8 Department of Radiation Oncology, Radboud University Nijmegen Medical Centre, Nijmegen, Netherlands. 9 Department of Surgery, Radboud University Nijmegen Medical Centre, Nijmegen, Netherlands. 10 Department of Radiation Oncology, Radiotherapiegroep, Arnhem/Deventer, Netherlands. 11 Department of Surgery, Zuyderland Medical Center, Heerlen, Netherlands. 12 Department of Surgery, University Medical Center Groningen, Groningen, Netherlands. 13 Department of Surgery, Erasmus University Medical Centre, Rotterdam, Netherlands. 14 Department of Medical Oncology, Amsterdam UMC, Amsterdam, Netherlands. 15 Department of Radiation Oncology, Maastro Clinic, Maastricht, Netherlands Purpose/Objective: Treatment decisions in locally advanced oesophageal cancer, such as type of neoadjuvant therapy and radiotherapy technique, lack quantitative, personalized risk-assessment tools. To address this, we developed and validated a multicentre prediction model for 2-year mortality following neoadjuvant chemoradiotherapy (nCRT) to enable more informed, risk-adapted clinical decision-making. Material/Methods: A Firth logistic regression model, to minimize bias, was trained on a multicentre retrospective cohort of 1492 photon (23 x 1.8Gy) nCRT patients from six institutions (2015-2020). The model was developed and validated using an internal-external cross-validation framework to assess and achieve consistent model performance across institutions, ensuring generalizability [1]. Candidate predictors included patient and tumour factors, alongside the Effective Dose to Immune Cells (EDIC), a metric quantifying radiotherapy-induced lymphopenia, which is associated with worse survival [2]. The model’s risk-stratification ability was assessed using Cox regression on predicted risk tertiles, stratified by histology (adenocarcinoma [AC] and squamous cell carcinoma [SCC]) to account for differences in treatment protocols. Model performance was evaluated by calibration (slope [ideally 1) / intercept [ideally 0]), discrimination (AUC), and heterogeneity across these performance measures (I ² [ideally 0]). Results: The model, which includes the EDIC [2] as a statistically significant predictor (Figure 1), demonstrated excellent and consistent calibration across institutions (Slope: 0.90 [0.76–1.05]; Intercept: -0.01 [-0.13–0.10], I ² =0% for both). Discrimination was also robust across all institutions (AUC: 0.65 [0.63–0.66], I ² =0%). The model achieved statistically significant risk stratification (p<0.01) for both histological types (Figure 2). For AC
Made with FlippingBook - Share PDF online