Multiple gene expression classifiers from different array platforms predict poor prognosis of colorectal cancer.

This study aimed to develop gene classifiers to predict colorectal cancer recurrence. We investigated whether gene classifiers derived from two tumor series using different array platforms could be independently validated by application to the alternate series of patients. Colorectal tumors from New Zealand (n = 149) and Germany (n = 55) patients had a minimum follow-up of 5 years. RNA was profiled using oligonucleotide printed microarrays (New Zealand samples) and Affymetrix arrays (German samples). Classifiers based on clinical data, gene expression data, and a combination of the two were produced and used to predict recurrence. The use of gene expression information was found to improve the predictive ability in both data sets. The New Zealand and German gene classifiers were cross-validated on the German and New Zealand data sets, respectively, to validate their predictive power. Survival analyses were done to evaluate the ability of the classifiers to predict patient survival. The prediction rates for the New Zealand and German gene-based classifiers were 77% and 84%, respectively. Despite significant differences in study design and technologies used, both classifiers retained prognostic power when applied to the alternate series of patients. Survival analyses showed that both classifiers gave a better stratification of patients than the traditional clinical staging. One classifier contained genes associated with cancer progression, whereas the other had a large immune response gene cluster concordant with the role of a host immune response in modulating colorectal cancer outcome. The successful reciprocal validation of gene-based classifiers on different patient cohorts and technology platforms supports the power of microarray technology for individualized outcome prediction of colorectal cancer patients. Furthermore, many of the genes identified have known biological functions congruent with the predicted outcomes.