Urban infrastructure and spatiotemporal environmental features for EGFR-mutant lung cancer

Background: Lung cancer in never-smokers is rising, with a substantial proportion harboring the EGFR mutation. While fine particulate matter (PM2.5) is a recognized risk factor, other intervenable pollutants and built environmental factors remain unknown. Objectives: To identify urban characteristics associated with EGFR-mutant (vs. wild-type) lung cancer using high-resolution spatiotemporal data. Methods: We analyzed 2,699 lung cancer patients with documented EGFR status treated at a high-volume academic medical center in New York City. Patient residential addresses were linked to high-resolution (300m x 300m) 5-year cumulative exposures to 3 air pollutants and 26 urban features. We developed Light Gradient Boosting Machine (LightGBM) models to classify EGFR status, comparing a basic clinical model with established predictors (Asian, female, never-smoking status, and adenocarcinoma histology) to an extended model with additional urban factors. Predictive performance was assessed based on discrimination (AUC). Results: We included 2,699 patients, of whom 54.1% were female and 25.8% self-identified as Asian, 11.2% as Black, and 7.4% as Hispanic; and 29% had EGFR-mutated cancer. The extended model showed modest improvements in discrimination (AUC: 0.775 [95% CI, 0.739-0.809] vs. 0.768 [0.723-0.811]), compared to the clinical model. Newly identified factors for EGFR-mutant status included black carbon (BC), nitrogen dioxide (NO2), proximity to airports, reduced access to public transportation, elevated noise levels, and lead exposure. Conclusions: Traffic-related pollutants (BC, NO2) from diesel engines and motor vehicles, and proximity to airports, were among the novel spatiotemporal features associated with EGFR-mutant lung cancer. These results may inform policy interventions.