EPID-33RISK MODELING FOR VASCULAR TOXICITY IN PATIENTS WITH GLIOBLASTOMA (GBM) TREATED ON NRG ONCOLOGY/RTOG 0825

Abstract

BACKGROUND: Tumor response and treatment-related toxicities vary greatly among patients despite careful selection criteria and accurate drug dosing. This projects was to evaluate risk and validate prediction models for vascular toxicity in patients with newly diagnosed GBM treated with bevacizumab. METHODS: We tabulated the incidence of cardiac and vascular toxicity in RTOG0525 (standard vs. dose-dense temozolomide) and RTOG0825 (bevacizumab (BEV) vs placebo). Based on the data collected from RTOG0825 BEV containing arm, a risk prediction model for hypertension (Grade 2-4) and thrombosis (Grade 3-4) was built based on known risk factors. Candidate risk factors were screened using univariable logistic regression (significance 0.15), then included in a multivariable model (significance 0.05) to determine the final risk model. RESULTS: In RTOG0825, overall incidence of hypertension was 11% (67/591) and thrombosis (73/591) 12%. Hypertension was higher with BEV (15 vs 8%) and no difference found for thrombosis (12 vs 13%). Men experienced more toxicity than women (12 vs 10% hypertension; 14 vs 9% thrombosis) but this was not significant at 0.05. None of the known risk factors for hypertension and thrombosis with BEV were significant in the final risk prediction model. For comparison, in RTOG0525, the incidence of cardiac toxicity was 4% (27/720) and thrombosis was 6%. CONCLUSIONS: Patients enrolled on these clinical trials had a lower overall incidence of both adverse events than would be expected in this demographic (middle aged or older with underlying glioblastoma where the incidence of hypertension is 30% and thromboembolism of 15%). Previously described risk factors were not confirmed, but may reflect selection by eligibility criteria of a healthier patient cohort compared with a population-based cohort. While clinical risk factors were unrevealing, planned genomic phenotyping focusing on polymorphisms may elucidate informative risk profiles. Supported by NCI grants U10CA21661, U10CA180868, U10CA180822, U10CA37422, U24CA114734, NIH grant 1R01NR013707, and Genentech.