ESTRO 2024 - Abstract Book

S2216

Clinical - Upper GI

ESTRO 2024

Pieter Populaire 1,2 , Gilles Defraene 2 , Lieven Depypere 3,4 , Philippe Nafteux 3,4 , Karin Haustermans 1,2

1 UZ Leuven, Department of Radiation Oncology, Leuven, Belgium. 2 KU Leuven, Laboratory of Experimental Radiotherapy, Leuven, Belgium. 3 UZ Leuven, Department of Thoracic Surgery, Leuven, Belgium. 4 KU Leuven, department of chronic diseases and metabolism (CHROMETA), Leuven, Belgium

Purpose/Objective:

Trimodality therapy is the gold standard for locally advanced esophageal cancer. While improving survival, it also increases the risk for postoperative complications. Approximately one in three patients develop postoperative cardiac complications after trimodality therapy. Here we aim to identify the attribution of radiotherapy dose to the heart in developing these complications.

Material/Methods:

All patients planned with IMRT on a 4D-CT-based average intensity projection image in our institution between 2018 and 2022 were selected for voxel-based complication prediction analysis. Herein, deformable registration of all scans to a reference patient was performed in MIM using a heart contour mask. Dose maps were transferred accordingly to the reference anatomy. Next, voxel-based dose differences between the groups with and without complication were calculated. Only clinically relevant complications requiring treatment were considered, as defined by corresponding comprehensive complication index (CCI) > 300. Voxel-based analysis was performed in the whole group and in histology-driven subgroups. Significance level (alpha) was defined as 0.05, after event label permutation. The mean dose to the identified area of interest was then incorporated into a normal tissue complication probability (NTCP) model predicting postoperative cardiac complications. NTCP model building was performed based on input factors age, gender, BMI, mean heart dose (MHD), mean lung dose (MLD), PTV and histology (squamous cell vs other) in addition to dose to areas of interest as defined during the voxel-based complication prediction analysis. Stepwise logistic regression based on Akaike information criterion was used. Interactions between variables were allowed during model building. Subsequent validation of the developed NTCP model was executed in an independent, temporally split, validation dataset of patients treated with IMRT in 2016 and 2017. Fifty patients (21 postoperative cardiac complications events), were identified for voxel-based analysis. No statistically significant dose difference for any heart voxel was observed for the whole group. However, for the adenocarcinoma subgroup (15 complications in 34 patients) there were statistically significant differences in per voxel dose within the heart contour. 86% of these voxels were located within the left ventricle. Subsequently, NTCP model building was performed using predefined parameters and the mean dose to the left ventricle (LVD). The optimal logistic regression model contained the factors age, LVD and BMI with linear combination - 92.30+1.20*age+1.89*BMI+1.69*LVD-0.024*age*BMI-0.021*age*LVD . Good model discrimination was obtained with an apparent AUC=0.85 (CI: 0.71-0.93) in the development dataset and an observed AUC=0.72 in the validation dataset (10 complications in 33 patients). The model indicates that larger LVD is associated with greater risk of postoperative cardiac complications. Elderly patients have an increased risk of complications, even for negligible LVD. Model building without LVD yielded only a poorly validating model with an observed AUC=0.56. Results: