Abstract Book

S59

ESTRO 37

Material and Methods All structures were re-defined in a total of 179 stage III NSCLC patients receiving CRT in 2004-2014 with prescribed doses of 50-80Gy@1.8-2.0Gy/fraction. Cardiac substructure definition followed that of RTOG 1106, and a semi-automatic atlas-based approach was used for segmentation. For each structure, equivalent dose fractionation correction (α/β=3Gy) was performed, and the mean, max, and minimum dose, as well as the minimum dose to the hottest x% volume (Dx; x=5-95% in 5% steps), and structure volume were extracted. Each of these and a total of 23 disease- (chemo timing, histology, tumor inferiority/laterality/stage/volume), and patient characteristics (age, cardiovascular function, ethnicity, diabetes, hypertension, hyperlipidemia, lung function, performance status, sex, and smoking history) were subject to Cox proportional hazard regression modeling. A candidate variable was suggested if presenting with p≤0.05 (note: Bonferroni-adjusted to p≤0.002 for all dose-volume variables given 23 comparisons/structure), and at most a weak-modest correlation with any other candidate variable (Spearman’s rank correlation coefficient, R s <0.80). Ultimately, a forward-stepwise approach was applied in which a candidate variable was considered final if p≤0.05 of the log-likelihood ratio statistics (p LL ). The number of deaths was 132 and 40 for OS and NCS, respectively. Robustness of p-values for identified final multivariate variables was investigated using Bootstrap resampling (1000 sample populations). Results A total of 102 and 126 dose-volume variables presented with p≤0.05 for OS, and NCS, respectively, but taking into account the R s criterion, the number of candidate dose- volume variables was reduced to five and six (OS, NCS) together with four disease- and patient characteristics for both endpoints. The final models included a total of 2-3 variables: increased left atrium (LA) D min , and worse performance status (OS); increased LA D 95 , electrocardiogram (ECG) abnormalities, and cardiovascular disease (NCS) all being robust on Bootstrap resampling (Figure; Table). Overall, dose-volume variables of cardiac substructures presented with lower p-values compared to those of the whole heart, or the tumor-subtracted total lung.

Conclusion The minimum dose to the left atrium finding suggests that limiting the low dose-bath to the cardiovascular system is important to prolong survival after chemo- radiation for NSCLC. In addition to performance status assessments, cardiovascular functional test acquired prior to treatment could further guide in treatment protocol stratification.

Symposium: Blood borne biomarkers

SP-0113 Circulating tumour DNA as a therapeutic biomarker for radiotherapy S. Bratman 1 1 Princess Margaret Cancer Centre, Department of Radiation Oncology, Toronto, Canada Abstract text 'Precision radiation medicine' requires robust biomarkers to help refine the delivery of radiotherapy. Radiation oncologists currently have few molecular tools at their disposal for predicting or rapidly assessing treatment efficacy. Circulating tumour DNA (ctDNA) is an attractive source of cancer biomarkers because it contains many of the same genetic and epigenetic aberrations that are present in tumours and can be non-invasively and repeatedly accessed from peripheral blood. While most of the clinical impact of ctDNA analysis has thus far been observed among patients with metastatic incurable cancer, emerging ctDNA detection technologies are now providing an opportunity to influence patients with localized potentially curable cancer. Possible uses of ctDNA analysis for such patients include prognostication and risk stratification, prediction of treatment response, longitudinal monitoring for adaptive treatment, and evaluating minimal residual disease. In this talk, I will describe recent advances in ctDNA technologies and potential clinical applications of ctDNA analysis throughout the therapeutic course. Furthermore, I will illustrate how ctDNA analysis could someday guide radiotherapy delivery by revealing differences in tumour radiophenotype. Because ctDNA is released into the bloodstream during the process of tumour cell death, a transient elevation in ctDNA levels could reflect rapid response and treatment sensitivity. Thus, by displaying dynamic changes with treatment, ctDNA as a biomarker for radiation response could enable new personalized treatment approaches. The path toward implementation