Authors:
Background:A central goal of precision oncology is to relate genomic alterations in a patient’s tumor to prognosis and therapy response. Today, variant interpretation is based on functional studies in model systems, retrospective case series, and prospective clinical trials. As more patients undergo tumor sequencing to guide care, ongoing real-world clinical annotation at scale using the electronic health record (EHR) presents an opportunity to identify new clinico-genomic associations.Methods:We retrospectively studied non-small cell lung cancer (NSCLC) patients treated in the Flatiron Health network ( > 265 US practices) from Jan 2011 to Dec 2016, who underwent FoundationOne tumor sequencing as part of routine care. Clinical data from EHRs (dates of diagnosis, death) were linked in a HIPAA-compliant manner with genomic findings (short variants, deletions, and amplifications across 395 genes). We calculated per-gene hazard ratios (HRs) using a Cox model to compare overall survival (OS) from advanced diagnosis in patients with and without alterations in each gene. Covariates included age, gender, smoking status, and histology. P-values were adjusted for multiple comparisons using false discovery rate (FDR) control.Results:The cohort included 1404 patients with NSCLC whose tumor samples were collected < 90 days after advanced diagnosis. When restricting to variants predicted to be functional, EGFR alterations were associated with lower mortality (HR = 0.7; 95% CI 0.5 - 0.8, adj-p = 0.03); this was expected given available targeted therapy. Variants in KEAP1 (HR = 2.0; CI 1.5 - 2.6, adj-p = 0.0004), SMARCA4 (HR = 1.7; CI 1.3 - 2.3, adj-p = 0.01), CTNNA1 (HR = 16; CI 3.8 - 63.8, adj-p = 0.01) and CREBBP (HR = 3.0; CI 1.6 - 5.6, adj-p = 0.03) were associated with higher mortality. In KEAP1, variants of unknown significance (VUS) were also associated with lower OS (HR = 1.6, CI 1.3-2.0, adj-p = 0.007).Conclusions:A continuously growing linked CGDB can reveal prognostic biomarkers for OS, and provide evidence to inform VUS annotation. Future work is needed to account for co-mutation patterns, divergent variant effects in the same gene, and unmeasured factors influencing treatment and outcome.
Sources:
ASCO Annual Meeting