Purpose: A 128-gene signature has been proposed to predict outcome in patients with stages II and III colorectal cancers. In the present study, we aimed to reproduce and validate the 128-gene signature in external and independent material.
Methods: Gene expression data from the original material were retrieved from the Gene Expression Omnibus (GEO) (n = 111) in addition to a Danish data set (n = 37). All patients had stages II and III colon cancers. A Prediction Analysis of Microarray classifier, based on the 128-gene signature and the original training set of stage I (n = 65) and stage IV (n = 76) colon cancers, was reproduced. The stages II and III colon cancers were subsequently classified as either stage I-like (good prognosis) or stage IV-like (poor prognosis) and assessed by the 36 months cumulative incidence of relapse.
Results: In the GEO data set, results were reproducible in stage III, as patients predicted to be stage I-like had a significant lower risk of relapse than patients predicted as stage IV-like (P = 0.04, Gray test). Results were not reproducible in stage II patients (P > 0.05, Gray test). In the Danish data set, two of four stage III patients with relapse were correctly predicted as stage IV-like, and the remaining patients were predicted as stage I-like and unclassifiable, respectively. Stage II patients could not be stratified.
Conclusions: The 128-gene signature showed reproducibility in stage III colon cancer, but could not predict recurrence in stage II. Individual patient predictions in an independent Danish material were unsatisfactory. Additional validation in larger cohorts is warranted.