Genomic Adjusted Radiation Dose (GARD) Predicts Local Recurrence and Overall Survival in Triple-Negative Breast Cancer

Abstract

<h3>Purpose/Objective(s)</h3> Although it is known that breast cancer is a heterogeneous disease with variable response to different treatments, radiation therapy (RT) is typically prescribed with a uniform approach. We have previously shown that the genomic adjusted radiation dose (GARD) is predictive of RT benefit in a pooled analysis which included hormone receptor positive and triple-negative breast cancer (TNBC) patients. Recent trials showed no benefit in overall survival (OS) with a randomized 16 Gy boost, but we hypothesize that GARD can predict outcomes in TNBC and could select patients who would benefit from dose escalation. <h3>Materials/Methods</h3> We used two IRB-approved, prospectively collected cohorts of TNBC patients treated with RT, 58 from the Netherlands Cancer Institute and 55 from Moffitt Cancer Center. Cox proportional hazards analysis was performed with GARD as a continuous variable for local recurrence (LR) and OS. We used the log-rank test to maximize the difference between groups above and below each GARD value and defined RxRSI as the physical radiation dose necessary to achieve that GARD cutpoint. Patients were stratified into 3 groups based on whether their RxRSI fell above, within, or below the standard-of-care dosing range of 50-66 Gy. To determine the minimum enrollment number needed to show a significant LR benefit in a clinical trial, we performed a power analysis by simulating an <i>in silico</i> trial with two arms. In one arm, all patients received 50 Gy, and in the other, dose was boosted to 66 Gy only for patients whose RxRSI fell within 50-66 Gy while the rest of the cohort received 50 Gy. Thus, selective dose escalation was only applied to patients who we predicted would benefit from it based on their individual RSI and GARD. The simulation was repeated with different enrollment numbers to find the smallest number needed to show a significant benefit in LR in a trial of this design. <h3>Results</h3> While the radiation dose delivered in the cohorts analyzed was between 43.5 to 75 Gy, GARD showed a much wider range (12 to 76). Cox analysis showed significant association between GARD and OS (HR=0.88, p=0.02) and strong association between GARD and LR (HR = 0.96, p=0.08). Receiving GARD of at least 22 maximized LR and OS benefit (p<0.01). By this measure, 50% of patients were optimized with 50 Gy, 22% would benefit from a boost, and 28% require higher radiation dose than is standard. Using our stochastic <i>in silico</i> method, we found that to see a significant difference in LR between boosted and standard-dosing groups in a trial of genomically stratified TNBC patients, only 30 patients would be needed. <h3>Conclusion</h3> Our analysis shows that for each unit GARD, a statistically significant increase in OS seen, and that using genomic stratification in trial design yields a small and quickly tractable trial to test the efficacy of this approach in tumor bed dose decision-making.