A Quantitative CT Imaging Signature Predicts Survival and Complements Established Prognosticators in Stage I...

Prognostic biomarkers are needed to guide the management of early-stage non-small cell lung cancer (NSCLC). This work aims to develop an image-based prognostic signature and assess its complementary value to existing biomarkers.

We retrospectively analyzed data of stage I NSCLC in 8 cohorts. On the basis of an analysis of 39 computed tomography (CT) features characterizing tumor and its relation to neighboring pleura, we developed a prognostic signature in an institutional cohort (n = 117) and tested it in an external cohort (n = 88). A third cohort of 89 patients with CT and gene expression data was used to create a surrogate genomic signature of the imaging signature. We conducted further validation using data from 5 gene expression cohorts (n = 639) and built a composite signature by integrating with the cell-cycle progression (CCP) score and clinical variables.

An imaging signature consisting of a pleural contact index and normalized inverse difference was significantly associated with overall survival in both imaging cohorts (P = .0005 and P = .0009). Functional enrichment analysis revealed that genes highly correlated with the imaging signature were related to immune response, such as lymphocyte activation and chemotaxis (false discovery rate < 0.05). A genomic surrogate of the imaging signature remained a significant predictor of survival when we adjusted for known prognostic factors (hazard ratio, 1.81; 95% confidence interval, 1.34-2.44; P < .0001) and stratified patients within subgroups as defined by stage, histology, or CCP score. A composite signature outperformed the genomic surrogate, CCP score, and clinical model alone (P < .01) regarding concordance index (0.70 vs 0.62-0.63).

The proposed CT imaging signature reflects fundamental biological differences in tumors and predicts overall survival in patients with stage I NSCLC. When combined with established prognosticators, the imaging signature improves survival prediction.