To perform this analysis, we aggregated data from multiple public cohorts and collaborated with different academic groups to have access to private data from prospective series and one clinical trial (Alliance CALGB9581). This project was approved by the Vall d'Hebron Institute of Oncology Ethics Committee. Patients signed informed consent for exploratory biomarker research on samples prospectively collected in accordance with the guidelines of Institutional Review Boards from each organization/clinical trial. Table 1 summarizes the final study population that included 2636 patients diagnosed with stage II/III CRC, untreated (N = 1656) or treated (N = 980) with adjuvant chemotherapy, with clinicopathological and molecular annotation for variables of interest. Transcriptomic data were normalized following standard bioinformatics procedures for CMSclassifier and MCPcounter application independently in each cohort (see supplementary Table S1, available at Annals of Oncology online for details on gene expression platform and tissue source). We obtained CMS1, CMS2, CMS3 and CMS4 Random Forest posterior probabilities as a continuous value (each ranging from 0–1) and final CMS labels using CMSclassifer R-package [13 (link)]. Likewise, the abundance of immune- and nonimmune-stromal cell populations was estimated from gene expression data using MCPcounter R-package [18 (link)]. Given the fact that MCPcounter scores are affected by gene expression platform and tissue source, microenvironment cell infiltration scores were scaled (from 0 to 1) first within three subgroups [Affymetrix in fresh frozen samples; Agilent in fresh frozen samples; and Almac-Affymetrix in formalin-fixed paraffin-embedded (FFPE) samples] and then rescaled after data aggregation to facilitate cross-study comparisons. Multiple imputation of random missing values was carried out via the mice R package in the aggregated cohort (supplementary Table S2, available at Annals of Oncology online).