ESTRO 36 Abstract Book

S120 ESTRO 36 2017 _______________________________________________________________________________________________

USA 3 University of Manchester, Manchester Academic Health Science Centre, Manchester, United Kingdom 4 Ingham Institute and Liverpool and Macarthur Cancer Therapy Centres, School of Physics, Liverpool, Australia 5 Ingham Institute and Liverpool and Macarthur Cancer Therapy Centres, School of medicine, Liverpool, Australia 6 University of Sydney, School of Physics, Sydney, Australia Purpose or Objective Early death after a treatment can be seen as a therapeutic failure. Wallington and colleagues reported that 8% of all non-small cell lung cancer (NSCLC) patients die within thirty days of systemic treatment initiation[1]. Identification of patient at risk for early mortality is crucial to avoid unnecessary harm and avoid costs. In this work, we validate the logistic regression model proposed by Wallington and colleagues in 2 independent datasets. Additionally, we develop our own model and validate it on the same datasets. Material and Methods Patients with NSCLC treated with concurrent chemoradiation were included in this study. The Institute 1 cohort consists of 411 patients treated in routine clinical practice. The Institute 3 cohort consists of 121 patients, treated in clinical trials. The Institute 4 cohort consists of 57 patients, treated in a clinical trial. The Institute 2 cohort consists of 355 patients, treated in routine clinical practice. A logistic regression model was learned on the Institute 1 cohort. This model used WHO performance status, age, nodal stage and prescribed tumor dose to make predictions. Results 11 out of 411 (3%) patients died within 30 day of start of treatment in the Institute 1 cohort and 22 out of 355 (6%) patients in the Institute 2 cohort. In both the Institute 4 and Institute 3 clinical trials, no patients died within 30 days. Death rates for the Institute 1 and Institute 2 cohorts combined are significantly higher than the death rates of the Institute 4 cohort and Institute 3 cohort combined (P<0.01) Survival curves for these cohorts are reported in figure 1. Based on the Institute 1 cohort, the AUC for the Wallington model was 0.69 (95% CI: 0.53-0.85) and 0.72 (95% CI: 0.49-0.94) with our own model. The AUCs were not significantly different (P=0.64) Based on the Institute 2 cohort, the AUC for the Wallington model was 0.58 (95% CI: 048-0.7), whereas it was 0.72 (95% CI: 0.64-0.81) with our own model. The difference was significant (P<0.001).

Circles identify deceased patients.

Figure 2: ROC curves of the models. Conclusion

Early mortality is more common in cohorts originating from routine clinical practice compared to clinical trials, indicating a selection bias for the trial patients. Development of accurate predictive tools for early mortality is important to inform patients about treatment options and optimize care. References [1] Wallington M,et al. Lancet Oncol 2016;17:1203–1216. PV-0241 Comparing endpoints of radiation induced lung injury for NSCLC: radiology vs. clinical symptoms U. Bernchou 1 , R.L. Christiansen 1 , J.T. Asmussen 2 , T. Schytte 2 , O. Hansen 2 , C. Brink 1 1 Odense University Hospital, Laboratory of Radiation physcis, Odense, Denmark 2 Odense University Hospital, Department of Radiology, Odense, Denmark Purpose or Objective Clinical symptoms is the gold standard endpoint in most studies of radiation induced lung injury for non-small cell lung cancer (NSCLC) patients even though the scoring often is challenged by confounding medical conditions. However, lung injuries frequently manifest radiologically; and radiologic injury could potentially be used in outcome modelling to disentangle effects of confounding factors. The purpose of the present study was to investigate the relation between clinically scored dyspnea and the extent and appearance of radiologic injury in the lung after radiotherapy for NSCLC patients. Material and Methods Eligible follow-up CT scans acquired within 6 months after commencement of radiotherapy were retrospectively evaluated in a cohort of 220 NSCLC patients treated to 60- 66 Gy in 30-33 fractions. The volume fraction of lung with radiologic injuries was estimated in each scan and was divided in three categories based on appearance: Interstitial changes, ground-glass opacities, or consolidation in the lung. Clinical symptoms of dyspnea was recorded retrospectively and scored according to the Common Terminology for Adverse Events scale. The scores were divided into the following groups: No (grade 0-1), mild (grade 2), or severe (grade 3+) symptoms. Differences in the fraction of injured lung between groups were analyzed using Mann-Whitney U tests with a Bonferroni correction used to adjust P-values to compensate for multiple comparisons. Results Of the patients included in the study, 127 (58%) did not develop symptoms, 82 (37%) developed mild symptoms, while 11 (5%) developed severe symptoms. Patients with severe dyspnea had a statistically significant higher fraction of injured lung (median fraction of injured lung =

Figure 1: Survival curves for each cohort investigated.