ESTRO 36 Abstract Book

S119 ESTRO 36 2017 _______________________________________________________________________________________________

We retrospectively analyzed outcomes in patients with a primary or recurrent non-small cell lung cancer measuring ≥5 cm, who were treated with 5 or 8 fractions of SABR at a single center, between 2003-2014. Patients who had prior thoracic radiotherapy were excluded. The maximum tumor diameter in the axial, transversal, or sagittal planes was measured on lung window-level settings in the end- inspiratory phase of the 10-phase free-breathing four- dimensional planning CT scan. Between 2003-2008, SABR was delivered using 8-12 non-coplanar static conformal beams and stereoscopic X-ray image-guidance, and after 2008, Volumetric Modulated Arc Therapy was used with online cone-beam CT based positioning on the tumor. All cases with potential severe toxicity (i.e. grade 3 or higher, ≥G3) were evaluated by a clinical panel consisting of three clinicians using the Common Terminology Criteria for Adverse Events version 4.03, and was consensus based. Results 63 consecutive patients with a median tumor diameter of 5.8 cm (range 5.1-10.4) were identified; 81% had T2N0 disease, and 18% T3N0. Median Charlson Comorbidity Index was 2 (range 0-6). After a median follow-up of 54.7 months, median survival was 28.3 months (95% CI 18.3- 38.2). For T2b tumors, median OS was 28.7 months (95% CI 12.2-45.3), and for T3 tumors this was 21.5 months (95% CI 16.4-26.6). Disease free survival at 2 years was 82.1%, and local, regional, and distant control rates at 2 years were 95.8%, 93.7%, and 83.6%, respectively. Distant metastases only were the commonest pattern of failure (10%). ≥G3 toxicity was recorded in 30% of patients, with radiation pneumonitis the most common toxicity (19%). A likely (n = 4) or possible (n = 8) treatment-related death was scored in 19% of patients. Retrospective review of CT scans revealed pre-existing interstitial lung disease in 8 patients (13%), with fatal toxicity developing in 5 of them (63%). Conclusion Lung SABR in tumors ≥5 cm resulted in high local control rates and acceptable survival outcomes, except in patients with co-existing interstitial lung disease. Our findings indicate that a more systematic screening for interstitial lung disease should take place prior to referral for SABR. PV-0239 Validation of lung cancer survival models in a clinical routine SBRT population J. Van Soest 1 , T. Purdie 2 , M. Giuliani 2 , P. Lindsay 2 , A. Hope 2 , D. Jaffray 2 , A. Dekker 1 1 Maastricht University Medical Centre+, Department of Radiotherapy MAASTRO - GROW School for Oncology & Developmental Biology, Maastricht, The Netherlands 2 University Health Network, Radiation Medicine Program, Toronto, Canada Purpose or Objective In recent years, many different prediction models have been developed based on clinical trial data. Although some of these models have been validated in external datasets, most of these validations did not report whether this validation tested reproducibility (same cohort characteristics for training and validation set), or transferability (different cohort characteristics). In this work, we performed an external validation of a survival prediction model learned on Stage III NSCLC patients in a routine clinical dataset of lung SBRT patients. Material and Methods Inclusion criteria were all patients with clinical T1-2N0M0 treated with SBRT from January 2005 to March 2014. The survival prediction model under validation was published before (PMID 25936599), including its dataset. Cohorts (original training and current validation) were compared using a KM curve and by calculating the cohort differences AUC (PMID 25179855). This cohort differences model predicts whether patients belong to the training or validation dataset. If this model has a high AUC, it means

that the model is able to predict whether a patient belongs to the training/validation cohort; indicating large cohort differences and testing for transferability of the model. If the AUC is low (close to 0.5), the cohort differences are small, and therefore validation tests reproducibility in a similar patient cohort. For validation, we applied the model to predict survival at several endpoints (6 months, 1, 2 and 3 years), and put the results into context with the cohort differences AUC. Finally, we learned a new logistic regression model for 2- year survival, using stepwise AIC as variable selection method. Results When investigating cohort differences, the KM curve in figure 1 already shows a difference between the Stage III training cohort, and the current clinical routine validation cohort of SBRT patients. Both in terms of survival and follow-up. The cohort differences AUC was 0.97; indicating an almost perfect prediction whether a patient belonged to the training or validation cohort. This indicates a large difference between patients in the training and validation cohort.

Table 1 shows the model performance on the validation dataset, indicating a decline in the current validation dataset (original model training & validation AUC: 0.64 and 0.58-0.60, respectively). Learning a model for 2-year survival on this SBRT cohort increased the AUC (0.73, n=267, events=97).

Conclusion Our current validation shows that the prediction model under investigation, learned using Stage III NSCLC patients, shows an equal performance in a routine clinical cohort of SBRT patients. Based on the cohort differences AUC, we conclude that the previous model is transferable to another patient population. Preliminary investigations show that a specific model for SBRT patients could increase model performance. Therefore future work is to further refine training, and externally validate a new model for NSCLC patients treated with SBRT. PV-0240 A logistic regression model to predict 30-day mortality: difference between routine and trial data A. Jochems 1 , I. El-Naqa 2 , M. Kessler 2 , C. Mayo 2 , J. Reeves 2 , J. Shruti 2 , M. Matuszak 2 , R. Ten Haken 2 , C. Faive-Fin 3 , G. Price 3 , L. Holloway 4 , S. Vinod 5 , M. Field 4 , M. Samir Barakat 4 , D. Thwaites 6 , A. Dekker 1 , P. Lambin 1 1 MAASTRO Clinic, Radiotherapy, Maastricht, The Netherlands 2 University of Michigan, Radiation oncology, Ann-Arbor,