Analyzing biomarkers of cancer immunotherapy (CIT) response using a real-world clinico-genomic database

Abstract

Background: Highly discriminating biomarkers of response to cancer immunotherapies (CIT) remain elusive. Characterization of large real-world populations treated with CIT as part of routine care may enable better stratification. Methods: Patients in the Flatiron Health Analytic Database with non-small cell lung cancer (NSCLC) who underwent comprehensive genomic profiling (CGP) by Foundation Medicine were included (n=2139). CGP included >300 genes and tumor mutation burden (TMB), stratified into low (TMB-L; <6 mut/MB), intermediate (TMB-I; 6-20 mut/MB), and high (TMB-H;>=20 mut/MB) tertiles (Johnson, CIR 2016). PD-L1 expression was obtained from results reported to clinicians from multiple labs (using varying antibodies). Genomic data was linked to de-identified electronic health record (EHR) data, from which nivolumab response was measured as overall response rate (ORR= SD, PR, or CR), median duration of therapy (mDOT), and median overall survival (mOS) from advanced diagnosis and from nivolumab initiation. Results: In patients treated with nivolumab (n=444, 20.8%), TMB-H predicted longer mDOT than TMB-L/I (7.5 vs 4.6 months, p=0.001), mOS from start of nivolumab treatment (median not reached vs 10 months, p<0.01), and mOS from advanced diagnosis (65 vs 29 months, p=0.10). In contrast, PD-L1 status (n=282) was not associated with ORR, DOT, or OS. Among patients negative for PD-L1, TMB-H predicted longer DOT (mDOT 391 vs 166 days, p=0.08) and higher ORR (100% in TMB-H [n=5] vs 62% in TMB-I/L [n=28], p=0.03). TMB remained predictive of DOT and OS from nivolumab start when controlled for histology, age, stage, smoking, gender, and race in multivariate analysis. Multivariate analysis of TMB-L patients identified two additional genomic predictors of duration on nivolumab: BRAF (HR 0.12, p=0.04), and BRCA 1/2 (HR 0.05, p=0.01). Conclusions: Real-world datasets combining clinical outcomes with genomic profiling may enable biomarker discovery in CIT. These data demonstrate the predictive power of TMB, which can augment and significantly improve on the currently approved PDL1 expression as a predictor of CIT response. They may also enable discovery of novel biomarkers that can identify potential CIT responders among TMB-L populations.