ESTRO 2021 Abstract Book

S415

ESTRO 2021

Ten advanced stage cancer patients were included in an in-silico planning study comparing VMAT, PBSPT and dose-escalation-by-contour proton treatments (PBSPT-DE). Plans were normalized to equivalent target coverage and the magnitude of dose escalation calibrated to patient-specific oxygenation levels obtained from HX4 PET imaging. A conventional plan comparison at a dosimetric level has been expanded with the estimation of normal tissue complication (NTCP) and tumour control (TCP) probabilities accounting for the variable hypoxia-induced radioresistance. Results Patient-specific loss of tumour control caused by hypoxia could be recovered by the use of PBSPT-DE with a simultaneous integrated boost paradigm (up to 22% in the hypoxic tumour sub-volume – mean 12%). Despite this dose-escalation, proton therapy allowed for statistically significant reductions of normal tissue complications compared to conventional VMAT. Decrease of heart and lung dose resulted in significantly lower risk of radiation-induced pneumonitis (mean -11.3%) and mortality due to heart failure (mean -7.4%). Conclusion Compared with homogeneous dose prescriptions, targeted hypoxia dose escalation results in higher probability of tumour control. Furthermore, PBSPT-DE plans showed clinically relevant reductions in NTCPs even compared to conventional non-boosted VMAT plans.

OC-0529 Actuarial NTCP modeling of contrast-enhancing brain lesions in proton-treated glioma patients E. Bahn 1,2,3,4 , J. Bauer 1,3,4 , S. Harrabi 1,4,3,5 , K. Herfarth 1,5,3,4 , J. Debus 1,2,3,4,5 , M. Alber 1,3,4 1 Heidelberg University Hospital, Department of Radiation Oncology, Heidelberg, Germany; 2 German Cancer Research Center (DKFZ), Clinical Cooperation Unit Radiation Oncology, Heidelberg, Germany; 3 Heidelberg Institute of Radiation Oncology, (HIRO), Heidelberg, Germany; 4 National Center for Tumor Diseases NCT, Integrative Radiation Oncology, Heidelberg, Germany; 5 Heidelberg Ion-Beam Therapy Center (HIT), Department of Radiation Oncology, Heidelberg, Germany Purpose or Objective Late contrast-enhancing brain lesions (CEBL) are a relatively frequent side effect in the brain with unexpectedly high incidence in proton therapy. Time-to-event and censoring of complication data are rarely taken into account in NTCP modelling, although this may lead to bias. Here, we present an actuarial NTCP model and investigate the effect of censoring on model predictions and variable selection. Materials and Methods We analyzed 110 low-grade glioma patients treated with proton therapy. In 23 patients, CEBLs appeared on T1 MR images during follow-up. Regions of interest were contoured and verified by two clinicians. Dose and dose- averaged LET (LETd) were recalculated by Monte Carlo (FLUKA) on the original planning CT images. The serial cure model (SCM) can be regarded as a combination of a survival model and an NTCP model. Its functional form is given by where H(t) denotes a cumulative hazard function and X the linear predictor. We analyzed the data with the SCM and with univariate and multivariate logistic regression (LR) (all cross- validated) and calculated the predictive power for each covariate as p-values with the likelihood-ratio test. The dosimetric covariate derives from a recently developed voxel-level model that accurately predicts the CEBL locations [1]. For the purposes here, this model can be reduced to an EUD-like formula that considers a variable RBE and an increased periventricular sensitivity. In addition, we tested four clinical covariates: patient age (AGE), chemotherapy yes/no (CHE), tumor resection yes/no (OP) and WHO tumor grade 1/2 (TG). Results P-values for the clinical covariates range from 0.004 to 0.06 by univariate analysis and from 0.24 0.86 by multivariate analysis, respectively (Tab. 1). For AGE and OP, p-values are much higher in the SCM compared to LR.