Biomarker predictors of outcome from a randomized trial of nivolumab +/- stereotactic body radiotherapy (SBRT) in metastatic (M1) head and neck squamous cell carcinoma (HNSCC).

Abstract

6063 Background: A minority of patients with M1 HNSCC respond to the anti-programmed death (PD-1) monoclonal antibody, Nivolumab (Nivo). We sought to determine biomarker predictors of outcome to Nivo in M1 HNSCC. Methods: Patients with M1 HNSCC were randomized (n = 60) with stratification by virus status (EBV/HPV+ vs not) to either Nivo alone or Nivo+SBRT to a single lesion. The primary end-point was objective response rate (ORR) in non-irradiated lesions with overall survival (OS) as a secondary end-point. PD-L1 staining in ≥ 1% of tumor cells was regarded as positive. Tumor mutation burden (TMB) was determined using a next generation sequencing assay (MSK IMPACT). Predictive model selection was done to minimize the Akaike Information Criterion (AIC). Results: There was no difference in ORR comparing the two treatment arms (p = 0.86). On univariate analysis, virus status trended towards predicting both ORR (Positive 41.9% vs Negative 20.7%, (p = 0.14)) and OS (1-yr OS for Positive, 65.9% vs 1-yr OS for Negative 40.6%, p = 0.08). In the sub-group of patients for whom PDL1 staining was available (n = 56), there was a trend towards association with ORR: PDL1+ 50% vs PDL1- 23.5% (p = 0.08). PDL1+ patients demonstrated significantly longer OS (1-yr OS 63% vs 47%, p = 0.02). There was no association between virus status and PDL1 staining. IMPACT data was available in 46 patients. TMB was significantly higher in virus negative (mean 8.1 mut/mB) vs. virus positive (mean 4.6 mut/mB) (p = 0.01); TMB was similar comparing PDL1+ to PDL1- (p = 0.47). TMB was higher in responders than non-responders (p = 0.02); TMB was significantly higher in virus positive responders (p = 0.01) and trended towards being higher in virus negative responders (p = 0.10). A model to predict ORR that included PDL1, virus status, and TMB minimized AIC with a C-index of 0.72. A model to predict OS that included PDL1 and virus status minimized AIC with a C-index of 0.62. Conclusions: A multi-variate model including viral status, PDL1 status, and TMB predicts well response to Nivolumab in M1 HNSCC. A model containing PDL1 and virus had moderate predictive value for OS. Clinical trial information: NCT02684253.