Purpose Recently, several expression-profiling experiments have shown that adenocarcinoma can be classified into subgroups that also reflect patient survival. In this study, we examined the expression patterns of 44 genes selected by these studies to test whether their expression patterns were relevant to prognosis in our cohort as well, and to create a prognostic model applicable to clinical practice,
Patients and Methods Expression levels were determined in 85 adenocarcinoma patients by quantitative reverse transcriptase polymerase chain reaction. Cluster analysis was performed, and a prognostic model was created by the proportional hazards model using a stepwise method.
Results Hierarchical clustering divided the cases into three major groups, and group B, comprising 21 cases, had significantly poor survival (P = .0297). Next, we tried to identify a smaller number of genes of particular predictive value, and eight genes (PTK7, CIT, SCNN1A, PGES, ERO1L, ZWINT, and two ESTs) were selected. We then calculated a risk index that was defined as a linear combination of gene expression values weighted by their estimated regression coefficients. The risk index was a significant independent prognostic factor (P = .0021) by multivariate analysis. Furthermore, the robustness of this model was confirmed using an independent set of 21 patients (P = .0085).
Conclusion By analyzing a reasonably small number of genes, patients with adenocarcinoma could be stratified according to their prognosis. The prognostic model could be applicable to future decisions concerning treatment.
(C) 2004 by American Society of Clinical Oncology.