Dose to the left ventricle is associated to the risk of acute cardiac toxicity in patients with esophageal cancer undergoing trimodality treatment.

Abstract

Esophageal cancer (EC) patients treated with neoadjuvant chemoradiotherapy (nCRT) followed by surgery (trimodality treatment) are at risk for developing cardiac toxicity (CarTox). The interaction between radiotherapy and surgery warrants specific CarTox prediction models in this population. EC-patients treated with photon-based nCRT at UZ Leuven between 01-2015 and 12-2023 were eligible for inclusion. A subset scanned with a standardized imaging protocol was used to perform voxel-based analysis (VBA) comparing per-voxel dose to the heart between patients with versus without CarTox. Statistically significant voxels were matched with cardiac substructures. Relevant substructure dose metrics were included in subsequent logistic regression-based toxicity prediction modelling, with a temporal development/validation split. Final external validation was performed on a mixed photon and proton-treated cohort of MD Anderson Cancer Center. For VBA, 50 patients were used to identify a cluster of voxels located mainly in the left ventricle (LV) that was higher dosed in patients with adenocarcinoma who developed CarTox. Subsequent model development (n = 77) and validation (n = 29) identified a model containing age, the absolute volume of the LV that received at least 10 Gy (V10Gy) and the interaction term. Model performance, depicted by AUC, was 0.76 in the validation population. In external validation (n = 50) AUC reduced to 0.66. After recalibration, in agreement with recommended closed-loop testing procedure, good LV absolute V10Gy dependent model performance was observed in the calibration curve. Model-based identification of high-risk patients and minimizing their low to intermediate LV-dose may help reduce CarTox in EC-patients undergoing trimodality treatment.