Image-Based Consensus Molecular Subtyping Refines Prognostication in Stage II/III Colorectal Cancer: A Retrospective Analysis of the SCOT Trial

Purpose Risk stratification in localized colorectal cancer (CRC) remains inadequate. Transcriptome-based classifiers such as consensus molecular subtyping (CMS) improve this, but are challenging to implement. We determined the performance of a novel image-based method for CRC CMS subtyping (imCMS) using routine pathology images from the SCOT trial and investigated association of imCMS subtypes with clinicopathologic and tumor molecular and immune factors, treatment benefit, and survival. Methods We applied a novel imCMS classifier to stage II/III CRCs from the SCOT trial (ISRCTN59757862) which examined 3 versus 6 months of adjuvant oxaliplatin-based chemotherapy following curative-intent surgery. We determined correlates of imCMS subgroups and investigated imCMS prognostic and predictive value for CRC recurrence-free interval, overall survival (OS), and survival after CRC recurrence (SAR) by univariable and multivariable Cox regression. Results A total of 2,816 cases were informative for imCMS classification, of which 664 (24%) were imCMS1 (immune), 987 (35%) imCMS2 (canonical), 433 (15%) imCMS3 (metabolic), 653 (23%) imCMS4 (mesenchymal), and 79 (3%) of mixed subtype. Analysis confirmed expected associations of imCMS1 with deficient DNA mismatch repair (dMMR) and imCMS3 with KRAS mutation and revealed substantially lower intraepithelial CD8 + and intrastromal FoxP3 + T-cell infiltrate (CD8 IE -FoxP3 IS ) in imCMS4 tumors. imCMS classification was strongly prognostic independent of clinicopathologic factors, dMMR, and CD8 IE -FoxP3 IS in multivariable analysis but did not predict differential benefit from chemotherapy duration or regimen. The combination of imCMS with CD8 IE -FoxP3 IS groups defined CRC subgroups with substantially different recurrence risk. Analysis of OS and SAR revealed notable discordance in imCMS3 and imCMS4 prognostic value before and after relapse. Conclusion imCMS classification improves CRC prognostication beyond existing tumor factors and is potentially deliverable in clinical practice. imCMS prediction of therapeutic benefit requires further study.