Radiogenomic studies have linked oncogenic alterations (OAs) to imaging patterns in lung cancers. As tobacco influences mutational profiles, we aimed to evaluate the impact of smoking status on radiological and radiomic features (RFs) in metastatic lung adenocarcinoma (MLUAD) on pre-treatment CT, and its effect on prognostication.

This IRB-approved retrospective study included adults with newly-diagnosed MLUAD (October 2016–January 2024) with contrast-enhanced CT, known smoking status, and molecular profiling (classified as smoker-related OA [sOA: BRAF, KRAS, STK11, MET], non-smoker OA [nsOA: EGFR, ALK, ROS1], other OA or wild-type). Two radiologists recorded metastatic sites. RFs were extracted from all lesions ≥ 1 cm³ and standardized into single-site RFs (ssRFs). Patient-level multi-site RFs (msRFs) were computed as averaged RFs and spatial dispersion metrics. Associations between smoking status and radiological, ssRFs, and msRFs were assessed using ANOVAs adjusted for lesion volume, location, number, tumor burden, and mutational status. Prognostic msRFs for objective response rate (ORR) and overall survival (OS) were identified via multivariable logistic and Cox regression. We evaluated whether smoking stratification improved LASSO-penalized radiomics survival models using 10-fold cross-validated concordance index (C-index).

357 patients were included (41.2% women, median age: 63.2 years, 11.5% never-smoker, 48.2% with sOA and 13.2% with nsOA). In exploratory multivariable interaction analyses, the miliary pattern, four ssRFs, and five msRFs were significantly associated with the smoking–mutation interaction. Among msRFs associated with ORR and/or OS (15.6% and 27.5% of them, respectively), two were also influenced by this smoking–mutation interaction. Smoking-based stratification modestly improved OS model performance (cross-validated c-index: from 0.573 to 0.627 without, and from 0.620 to 0.650 with mutation status).

Smoking status alone has limited direct impact on MLUAD radiophenotypes when concomitantly accounting for OAs, but may act as a surrogate for underlying molecular and treatment-related factors and thereby contribute to the prognostic performance of radiomics-based survival models, rather than representing an independent prognostic marker. The observed smoking–mutation interactions should be considered exploratory and warrant external validation.