Association of Dose to Cardio-vascular Structures with Treatment-induced Electrocardiogram changes in Stage III Non-Small Cell Lung Cancer

Abstract

To explore whether dose to cardiac substructures is associated with electrocardiogram (ECG) changes in stage III Non-Small Cell Lung Cancer (NSCLC) radiation therapy (RT). All our stage III NSCLC patients treated with conventionally fractionated intensity-modulated RT in one phase combined with concurrent/sequential chemotherapy in 2004-2014 were initially considered. Out of a total of 241 identified patients treated to a median of 64Gy (concurrent/sequential chemotherapy: 64%/36%), ECG before and after RT was available for 155 patients. A new ECG event after compared to before RT (ΔECG) was classified as Arrhythmic, Ischemic/Pericardial, or Non-specific (AΔECG, I/PΔECG, or NSΔECG). A total of 132 fractionation-corrected (α/β=3Gy) dose-volume histogram (DVH) metrics were extracted from the four heart chambers, and from the inferior and superior vena cava (IVC, SVC). Modeling of each ECG change endpoint was based on Cox regression, and a candidate predictor was suggested by a p-value≤0.05 adjusted for false discovery rate. One DVH metric was considered for each structure (lowest p-value). Multivariate regression was conducted with forward-stepwise selection (retention criterion: p≤0.05 from a likelihood ratio test), and both univariate and multivariate analyses were conducted on 1000 bootstrapped replicates. Model discrimination of the most frequently selected multivariate models (≥10% of bootstrap replicates) was assessed using the C-index. AΔECG, I/PΔECG, and NSΔECG was observed in 66%, 35%, and 67% of the patients within a median follow-up time of 7, 9 and 8 months, respectively. Four candidate predictors (p=0.01-0.03) were identified for AΔECG: the minimum IVC dose to the hottest 95% (IVCD95), left atrium D10 (LAD10), right atrium maximum dose (RADmax), and SVC minimum dose (SVCDmin). Both LAD10 and SVCDmin (p=0.002, 0.01) were also candidate predictors for NSΔECG, while no DVH metric predicted I/PΔECG. The three suggested candidate multivariate models for AΔECG were: IVCD95, RADmax, and IVCD95+RADmax. The corresponding models for NSΔECG were: LAD10, SVCDmin, and LAD10+SVCDmin. The discriminatory ability of the three AΔECG and the three NSΔECG models was of a similar magnitude (C-index: AΔECG=0.65-0.66; NSΔECG=0.65-0.71). Our identified DVH models indicated that high atrial dose (AΔECG: right; NSΔECG: left), as well as low dose to the vena cava (AΔECG: inferior; NSΔECG: superior) is associated with treatment-induced Arrhythmic and Non-specific ECG changes. The rate of Ischemic/Pericardial ECG changes within the studied time period was considerably lower, and dose to other cardiac substructures is likely to shed light on its radiation-induced etiology. Reducing high atrial doses and ‘spill’ doses to the vena cava may be considered in patients for whom it will still be possible to deliver the therapeutic tumor dose.