GARD is a Pan-Cancer Predictor of Outcome and Radiation Treatment Benefit

Abstract

Background: Radiation therapy (RT) is prescribed using an empiric one-size-fits-all paradigm. We hypothesize that the Genomic Adjusted Radiation Dose (GARD), a quantification of personalized RT dose effect, will reveal interpatient heterogeneity associated with opportunities to improve outcomes compared to physical RT dose alone. To test this hypothesis, we performed a pan-cancer, pooled analysis of all available cohorts with information available to calculate GARD. Methods: Using 9 previously-published datasets, we defined two clinical endpoints: (i) time to first recurrence and (ii) overall survival, comprising 1,257 (972 +RT, 285 -RT) and 636 patients (414 +RT, 222 -RT), respectively. We used Cox regression stratified by disease site to test association between GARD and outcome with separate models using RT dose and sham-GARD for comparison. Interaction tests between GARD and treatment (+/- RT) were performed using the Wald statistic. Results: Pooled analysis reveals GARD is a continuous predictor of recurrence (HR = 0.984, CI [0.982,0.996], p=0.004) and survival (HR = 0.975, CI [0.958, 0.993], p=0.005). Interaction term was significant for OS (p =0.026) and for recurrence for patients who achieve GARD>19.2 (p=0.04). Physical RT dose and sham-GARD were not significantly associated with either outcome. Conclusions: Biologic effect, as quantified by GARD, is significantly associated with recurrence and survival: it is predictive of RT benefit; and physical RT dose is not. We propose integration of genomics into radiation dosing decisions, using a GARD-based framework, as the new paradigm for personalizing RT treatment. Funding: There was no funding for this work Declaration of Interest: JGS, SAE and JFTR hold IP on RSI, GARD and RxRSI, in addition to equity in Cvergenx, a company which seeks to commercialize these methods. All other authors have nothing to declare.