The study included patients diagnosed with metastatic lung adenocarcinoma between October 2016 and January 2024. Participants needed pretreatment contrast-enhanced computed tomography, documented smoking status and molecular profiling. Women represented 41.2% of the cohort, and the median age was 63.2 years. Never-smokers accounted for 11.5% of patients. Molecular findings were grouped as smoker-related oncogenic alterations, including BRAF, KRAS, STK11 and MET; non-smoker-related alterations, including EGFR, ALK and ROS1; other alterations; or wild-type status. Smoker-related alterations were present in 48.2% of the cohort, while non-smoker-related alterations occurred in 13.2%.

Two radiologists recorded metastatic sites, and the investigators extracted radiomic features from every lesion measuring at least 1 cm³. These measurements became single-site radiomic features for individual lesions. At the patient level, the researchers also calculated multi-site features by averaging lesion-level features and deriving spatial-dispersion metrics that described differences among lesions. Associations between smoking status and radiological patterns, single-site features and multi-site features were evaluated with analysis of variance adjusted for lesion volume, lesion location, number of lesions, total tumor burden and mutational status. The adjustment was designed to assess smoking-related imaging differences after accounting for several tumor characteristics and molecular categories.

In exploratory multivariable interaction analyses, the miliary imaging pattern, four single-site radiomic features and five multi-site radiomic features were associated with the interaction between smoking status and mutation class. Among multi-site features associated with objective response rate or overall survival—15.6% and 27.5% of the multi-site features, respectively—two also showed influence from the smoking-by-mutation interaction. These results indicate that the relationships among smoking history, tumor molecular alterations and imaging phenotypes were not uniform across all patients. They do not show that smoking caused the observed differences, and the source study characterizes the interaction findings as exploratory rather than established clinical effects.

For prognostic analysis, the researchers used multivariable logistic regression to examine objective response rate and Cox regression to examine overall survival. They then tested whether stratifying LASSO-penalized survival models by smoking status improved their performance, using 10-fold cross-validation and the concordance index. The overall-survival model’s cross-validated concordance index rose from 0.573 to 0.627 without mutation status and from 0.620 to 0.650 when mutation status was included. The authors interpret smoking status as potentially acting as a surrogate for underlying molecular and treatment-related factors that contribute to radiomics model performance, rather than as an independent prognostic marker. Because the study was retrospective and the interaction analyses were exploratory, the findings require testing in independent cohorts before their prognostic implications can be established.

Subject of Research: Smoking-related radiomic and prognostic variation in metastatic lung adenocarcinoma

Article Title: Impact of smoking status on CT-derived radiological and single- and multi-site radiomic features in metastatic lung adenocarcinoma: implications for prognostication

Article References: Crombe, A., Sitruk, L. A., Masson–Grehaigne, C., Lafon, M., Palussiere, J., Bonhomme, B., Cousin, S., Lassau, N., & Italiano, A. (2026). Impact of smoking status on CT-derived radiological and single- and multi-site radiomic features in metastatic lung adenocarcinoma: implications for prognostication. BMC Cancer. https://doi.org/10.1186/s12885-026-16794-6

Image Credits: AI Generated

DOI: 10.1186/s12885-026-16794-6

Keywords: metastatic lung adenocarcinoma, smoking status, CT radiomics, oncogenic alterations, smoking-mutation interaction, radiological features, overall survival, objective response rate, LASSO modeling, prognostication, Impact, smoking