Early Prediction of Acute Esophagitis Using Accumulated Dose and Local Volume Change of Esophagus

Abstract

Acute esophagitis (AE) is a common dose-limiting toxicity with significant impact on quality of life in patients with locally advanced non-small cell lung cancer (LA-NSCLC). In this study, we performed early AE prediction using weekly accumulated esophagus dose and its associated local volumetric change. The study included 60 LA-NSCLC patients consecutively treated (2017 to 2019) via IMRT to 50-72Gy in 2 Gy-fractions with concurrent chemotherapy and had weekly cone beam CTs (CBCTs). In addition, nine of the patients also had weekly respiratory triggered T2-weighted MRIs (3T) during the RT course. A total of 35 patients (58%) developed grade 2 AE (AE2; requiring medication cf. CTCAE) at a median of 4 weeks post-RT start. The weekly esophagus dose was accumulated on the contours delineated by a radiation oncologist on CBCTs and MRIs. Week to week esophagus changes were calculated using determinant of the gradient of the deformation vector field as Jacobian map obtained via a regularized diffeomorphic registration. The Jacobian at each voxel measures the percentage of local esophagus volume expansion/shrinkage. Parameters quantifying esophagus volume changes were the net volume change, maximum percentage of expansion (ExMax%) and volumes with ≥5%,10% and 15% local expansions (VE5%, VE10%, VE15%) at the current week with respect to the previous week. For the 51 patients with CBCTs only, a multivariate logistic regression model was built with 5-fold cross-validation to perform early prediction of AE2 using a combination of the accumulated Mean Esophagus Dose at week two (MEDW2) and VE10% from week one to two (w1→2). The model was externally validated on the nine patients who had MRI; and compared to a previously published model using only planned mean dose (MEDPlan). Performance was assessed using AUC and the Hosmer-Lemeshow test (PHL). Univariate analysis showed VE10% (p = 0.003), VE5% (p = 0.006) and ExMax% (p = 0.02) significantly predicted AE2 (p<0.05 after Bonferroni correction). A model combining MEDW2 and w1→2 VE10% from CBCT data had the highest accuracy (AUC = 0.74) with good calibration (PHL = 0.27) and was considerably better than the MEDPlan model (AUC = 0.64; PHL = 0.09). When externally validated on the MRI data, the combined model showed high accuracy (AUC = 0.79). The population average w1→2 VE10% and MEDW2 for AE2 vs. non-AE2 patients were 2.5cm3 vs. 0.6cm3 and 5.4Gy vs. 4Gy, respectively. According to the model, w1→2 VE10% >2.8cm3 and MEDW2 >8.1Gy correspond to a 50% complication probability of AE2. A cross-validated model with a combination of local volume change and accumulated dose of esophagus from the first two weeks of RT significantly improved the prediction of AE compared to previously published model using only the planned dose. This model could inform plan adaptation early during the treatment to lower the risk of esophagitis.