Drivers and patterns of disease progression as a novel schema for risk stratification in metastatic non-small cell lung cancer

Background: Targeted therapy (TT) and immunotherapy (IO) have transformed the systemic therapy (ST) landscape for non-small cell lung cancer (NSCLC), but overall prognosis remains poor. Select organ metastases may be predictive of decreased survival; however, the fundamental drivers and patterns of metastatic progression are not well established, and the site-specific effects of ST are poorly understood. We leveraged a large-scale, real-world database containing metastatic site, genomic, and treatment information to characterize and assess the prognostic value of patterns of metastasis and genomic correlates on real-world overall survival (rwOS) in patients with metastatic NSCLC receiving ST. Methods: This retrospective study utilized the Flatiron Health-Foundation Medicine Clinico-Genomic Database of patients with metastatic NSCLC treated with first-line (1L) ST. Patient, tumor, and treatment variables were compared with chi-squared tests. rwOS was estimated via Kaplan Meier method and compared with logrank test; adjusted hazard ratios were computed with multivariable Cox regression. Bernoulli mixture models were used to cluster patients by metastatic sites at 1L therapy start using the R package “flexmix”. Results: Sites of disease were evaluated for 10,571 patients. Patients with spleen, skin, liver, kidney, and bone metastases had worse OS. Data from 18 metastatic sites revealed 8 clusters: high metastatic burden (HMB, ≥3 sites with frequency ≥0.5, n = 933), bone (n = 2,889), pleura (n = 2,469), lung (n = 1,710), liver (n = 1,345), brain (n = 1,163), lymph node (LN, n = 748), and adrenal avid (n = 659). While HMB (8.4 months) and liver mets (8.9 months) are known to be associated with poor median OS, we also found that bone (11.9 months), adrenal (13.4 months), pleura (14.0 months), and LN (14.4 months) clusters had worse median OS compared to lung (17.7 months) and brain (16.5 months, p<0.001). Certain mutations had increased odds (OR>1.5 and p < 0.05) in site-specific clusters, including: adrenal (SRC, ARID1A, BCL2L1, ATR, EPHA3, CCND3, KEAP1), liver (AXL, AKT1, AKT2, EPHB1), brain (MAP2K4, GLI1, PIK3CB, KDR), and bone (ERBB3). Interestingly, EP300, EPHB4, and PBRM1 mutations were associated with HMB, but no individual sites. Adrenal and brain clusters had the highest tissue tumor mutational burden of the clusters. TT was associated with greater OS compared to other ST in lung, pleura, and bone clusters (HR 0.68-0.80, p < 0.01). Conclusions: Definable patterns of metastasis predict survival and treatment response in lung cancer. Genomic patterns of mutation predict site-specific metastatic spread. This study is the most comprehensive evaluation of the impact of sites of metastasis and genetic markers on survival outcomes in patients with NSCLC to date. Our proposed clusters may represent a novel framework for mechanistic inquiry and risk stratification in advanced NSCLC.