S68 ESTRO 36 _______________________________________________________________________________________________ compared with the recalibrated QP-model. This suggested the need to add different factors to improve discrimination. Using restrictive (BIC) analysis, the final model contained smoking status (current vs former&never) and MLD (AUC=0.78; R2=0.23). At less restrictive analysis (AIC), age, total-lung-volume, V5 and V30 of the heart, sequential chemotherapy, and MLD might be useful; in addition, MLD may be replaced by ipsilateral-lung V20 and total-lung V5. At internal validation, this latter model rendered AUC=0.80 and R2=0.28, however with much higher correction for optimism, implying potentially decreased generalizability to other cohorts. Conclusion Intending external validation, both the QP and the AQPmodels needed recalibration (of slope and intercept, and of intercept only, respectively), which might be explained by employment of modern RT techniques and 90% administration of chemoradiotherapy in our cohort. A conservatively improved pneumonitis model employing modern chemoradiotherapy-techniques includes MLD and current-smoking status (Figure).
OC-0140 Updating QUANTEC and clinically adjusted QUANTEC models for pneumonitis at external validation A. Van Der Schaaf1, J. Lodeweges1, A. Niezink1, J. Langendijk1, J. Widder1 1 UMCG University Medical Center Groningen, Radiation Oncology, Groningen, The Netherlands Purpose or Objective To externally validate and eventually recalibrate and update the original QUANTEC pneumonitis (QP) model (Marks et al, IJROBP 2010) and the QUANTEC model adjusted for clinical risk factors (AQP; Appelt et al, Acta Oncol 2014) in a cohort treated with 3D-CRT, IMRT, or VMAT, combined in 90% with chemotherapy. Material and Methods The external validation cohort was composed of n=220 patients with lung cancer (NSCLC, SCLC) stages (II-)III with complete dosimetric and prospectively scored pneumonitis data (G2 or higher), treated from 2013 to 2016 within the framework of a prospective data registration program (clinicaltrials.gov NCT02421718). Model performance was tested for discrimination (area under the curve, AUC), (pseudo-)explained variance (Nagelkerke’s R2), and calibration (Hosmer-Lemeshow test, HL-test), before and after intercept and slope recalibration. Then, updating was performed by first refitting the coefficients from the AQP-model to our data, then stepwise manually removing unnecessary variables, followed by adding new potential variables. The procedure was then repeated automatically using Akaike and Bayes Information Criteria (AIC, BIC), respectively. Resulting models were in turn internally validated to correct AUC and R2 for optimism using bootstrapping with backward elimination based on AIC. Results After recalibration of intercept and slope, the QP-model predicting pneumonitis based exclusively on mean lung dose (MLD) performed well (AUC=0.77; R2=0.21; HL-test: p=0.38), while without recalibration the model would not fit our data (HL-test: p<0.001). The AQP-model needed recalibration of the intercept only, but discriminated worse and explained less variance (AUC=0.72; R2=0.16)
OC-0141 Validation of dose-sensitive heart regions affecting survival in SABR lung cancer patients A. McWilliam1, J. Kennedy2, C. Faivre-Finn1, M. Van Herk1 1 The University of Manchester, Division of Molecular and Clinical Cancer Science- Faculty of Biology- Medicine and Health, Manchester, United Kingdom 2 The Christie NHS Foundation Trust, Department of Informatics, Manchester, United Kingdom Purpose or Objective Recent advances in radiotherapy allow an increasing proportion of lung cancer patients to be treated with curative intent. However, evidence is emerging that dose to critical organs may be influencing patient survival. The authors recently presented their work identify a dose sensitive sub-region located in the base of the heart where excess dose resulted in worse patient survival (McWilliam IJROBP 96(2S):S48-S49). This work aims to determine whether the same effect was observed in patients treated with Stereotactic Ablative Radiotherapy (SABR), thereby validating our previous results. Material and Methods The previous work used 1101 non-small cell lung cancer patients treated with 55Gy in 20 fractions. Validation was performed in 89 SABR patients treated with 60Gy in 5 fractions. For both groups, CT scans and dose distributions
S69 ESTRO 36 _______________________________________________________________________________________________ were deformable registered to a reference patient, focusing on the lungs with bone masked. Mean dose distributions were created for patients alive or dead at a set time-point, censored for follow-up. Dose differences were tested for significance with permutation testing. The most significant area defined an anatomical region of interest and individual patient doses collected. A multivariate analysis investigated the importance of this region in patient survival, including tumour size. Coxregression survival curves were plotted with patients split to those receiving less than or more than the same biologically equivalent dose that optimally split survival in the 20 fraction patients (α/β = 2). Results For 20 fraction patients, from 6 months, a significant difference was seen in the dose difference between patients alive and dead (p<<0.001). The most significant area was in the base of the heart near the origin of the coronary arteries, median dose of 16.3Gy (BED 10.3Gy). Multivariate analysis showed that tumour size was highly significant for patient survival (p<0.001) as was dose received by the anatomical region (p=0.029), HR 1.21 (1.02–1.44), highlighting the importance of dose received by this region. Cox-regression survival curves were plotted with patients split by those receiving more than or less than 8.5Gy, log-rank p<0.001, figure 1A, controlled for tumour size (p<0.001) and age (p=0.11). A cox-regression with the SABR patients split at 6.3Gy (translated BED from the 20 fraction patients) was plotted, figure 1B. A highly significant difference in survival (log-rank p=0.016) was seen where patients receiving more than 6.3Gy showed worse survival. Tumour size was not significant in the SABR group.
Conclusion Dose to a specific region in the base of the base heart predicts for early death in lung cancer patients treated with 55Gy in 20 fractions, as well as for SABR patients treated to 60Gy in 5 fractions. The effect was seen for the same BED (a/b = 2Gy). In the future, we will extend the SABR group and initialise cardiac imaging studies to identify a clinical cause for this effect. OC-0142 Incidental dose to cardiac subvolumes does not improve prediction of radiation pneumonitis in NSCLC R. Wijsman1, F. Dankers1, E. Troost2, A. Hoffmann2, J. Bussink1 1 Radboud University Medical Center, Radiation oncology, Nijmegen, The Netherlands 2 Institute of radiooncology, Helmholtz-Zentrum DresdenRossendorf, Dresden, Germany
Purpose or Objective Conflicting results have been reported for the combined effect of heart and lung irradiation on the development of radiation pneumonitis (RP). The reported studies based on 3D-conformal radiotherapy considered the whole heart as an organ-at-risk, thereby not distinguishing between dose to the cardiac ventricles and atria. We assessed whether inclusion of incidental dose to these cardiac subvolumes improved the prediction of Grade ≥3 RP. Material and Methods We retrospectively assessed 188 consecutive patients with stage III non-small cell lung cancer (NSCLC) having undergone (chemo-)radiotherapy (≥60 Gy) using intensitymodulated radiation therapy (until 2011) or volumetricmodulated arc therapy (starting in 2011). Most patients (n=182) received 66 Gy in 33 (once-daily) fractions to the primary tumour and involved hilar/mediastinal lymph nodes based on FDG-PET/CT. The lungs and heart (ventricles and atria separately in 156 patients that received a contrast enhanced planning CT) were recontoured to generate accurate dose-volume histogram (DVH) data. RP was assessed using the Radiation Therapy Oncology Group scoring criteria for pulmonary toxicity. Since high multicollinearity was observed between the DVH parameters, those with the highest Spearman correlation coefficient (Rs) were selected for the modelling procedure. Using a bootstrap approach, clinical parameters [age, gender, performance, smoking status, forced expiratory volume in 1 second, and cardiac comorbidity (i.e., medical history of myocardial infarction, heart failure, valvular heart disease, cardiac arrhythmias and/or hypertension)] and DVH parameters of lungs and heart (assessing atria and ventricles separately and combined) were evaluated for RP prediction. Results Twenty-six patients (13.8%) developed RP (median followup 18.4 months). Only the median mean lung dose (MLD) differed between groups (15.3 Gy vs 13.7 Gy for the RP and non-RP group, respectively; p=0.004). Most Rs of the lung DVH parameters exceeded those of the heart DVH parameters and only some lung DVH parameters were significantly correlated with RP [See Figure 1; highest Rs for MLD (0.21; p<0.01)]. Only cardiac comorbidity was borderline associated with RP (p=0.066) on univariate logistic regression analysis. After bootstrap modelling, heart DVH parameters were seldom included in the model predicting Grade ≥3 RP. The optimal model consisted of: MLD (Odds ratio (OR) 1.28 per Gy increase; p=0.03) and cardiac comorbidity (OR 2.45 in case of cardiac comorbidity; p=0.04). The area under the receiver operator characteristic curve was 0.71, with good calibration of the model.