Purpose: In order to aid the development of patient-tailored therapeutics, we attempted to identify a relapse-related signature that allows selection of a group of adenocarcinoma patients with a high probability of relapse. Patients and Methods: Whole-genome expression profiles were analyzed in 117 lung adenocarcinoma samples using microarrays consisting of 41,000 probes. A weighted voting classifier for identifying patients with a relapse-related signature was constructed with an approach that allowed no information leakage during each training step, using 10-fold cross-validation and 100 random partitioning procedures. Results: We identified a relapse-related molecular signature represented by 82 probes (RRS-82) through genome-wide expression profiling analysis of a training set of 60 patients. The robustness of RRS-82 in the selection of patients with a high probability of relapse was then validated with a completely blinded test set of 27 adenocarcinoma patients, showing a clear association of high risk RRS-82 with very poor patient prognosis regardless of disease stage. The discriminatory power of RRS-82 was further validated using an additional independent cohort of 30 stage I patients who underwent surgery at a distinct period of time as well as with the Duke data set on a different platform. Furthermore, completely separate training and validation procedures using another data set recently reported by the Director's Challenge Consortium also successfully confirmed the predictive power of the genes comprising RRS-82. Conclusion: RRS-82 may be useful for identifying adenocarcinoma patients at very high risk for relapse, even those with cancer in the early stage.
All Science Journal Classification (ASJC) codes
- Cancer Research