Evaluation of efficacy of nivolumab by baseline factors from ATTRACTION-2.

Abstract

8 Background: Nivolumab (Nivo) has shown superior efficacy with manageable safety for G/GEJ cancer refractory to, or intolerant of, standard chemotherapy in phase 3 study (ATTRACTION-2). Subgroup analysis of OS according to baseline demographics and disease characteristics favored Nivo over placebo in all subgroups. However, about half of patients had early disease progression even after administration of Nivo. Here, we explored factors of patients who had early progression after administration of Nivo. Methods: A statistical random forest method, a type of machine learning method, was used to explore clinical factors that could contribute to early progression after Nivo. In this analysis, the outcome variable was 56-day PD rate and covariates were patients’ clinical background factors. Factors were explored by comparing Nivo and placebo arms in all subgroups, constructed with 1) every single factor; 2) every pair of factors coming from the covariates. Results: In a single factor extraction, hyponatremia was identified as the most highly contributing factor to early progression. Of 330 patients in the Nivo arm, 59 patients showed hyponatremia at the baseline. Importantly, in a pair factor extraction, the patient population with hyponatremia and any of high NLR (N = 31, 9.4%, cut off; second tertile), high neutrophil (N = 28, 8.5%, cut off; second tertile), PS1 (N = 50, 15.2%) or no prior use of ramucirumab (N = 55, 16.7%), were found as risk factors of early progression in Nivo. PFS-KM curves were almost overlapped between Nivo and placebo arms in these patients with any pair of these risk factors. Conclusions: Our exploratory analysis suggested that a patient population with both ‘poor general condition’ represented by hyponatremia and PS 1, and ‘inflammatory conditions’ represented by high NLR and/or high neutrophil count, would be potential risk factors associated with early progression. This needs further validation with other potential biomarkers in tumor tissue and additional studies. Additional analysis with biomarkers such as PD-L1 will be presented at the meeting. Clinical trial information: NCT02267343.