7657 Background: Our aim was to determine the genes predictive for relapse-free and overall survival in stage I-II NSCLC. Methods: We analyzed the gene expression profiles in lung cancer specimens collected from 70 NSCLC patients (pts) who underwent curative pulmonary resection between 1999 and 2004 in two Polish centers (Gdansk, Bialystok). There were 54 men and 16 women aged 37–77 yrs (median 62.5 yrs), 45 with squamous cell ca, 22 with adenoca and 3 with large cell ca. Eight pts were staged pT1, 59 pT2 and 3 pT3; there were 49 and 21 pN0 and pN1 pts, respectively. 30 pts had a relapse and 32 pts died (median follow-up 36 months). Samples of tumor tissue were collected intraoperatively and snap-frozen, total RNA was isolated and gene expression profiling was carried out by Affymetrix HG-U133 2.0 Plus oligonucleotide microarray. Samples were pre-processed with RMA algorithm, gene selection was carried out by Support Vector Machines and Bayesian Compound Covariate Classifier, using own procedures and BRB-Array software developed by Simon and Peng Lam. Survival time prediction was carried out by method developed by Bair and Tibshirani (PLoS Biology 2004). Results: Based on the microarray gene expression profiling, the relapse could be predicted with 75.0% specificity and 53.3% sensitivity (positive predictive value [PPV] 61.5%, negative predictive value [NPV] 68.2%). The classifier, obtained by cross-validation of 70-sample dataset, consisted of 170 transcripts. Further gene selection was based on the prediction of the relapse-free survival: 1,919 genes, selected by fitting Cox proportional hazard model (p<0.05) and further used to predict survival by 4 principal components, distinguished between pts with high and low risk of relapse (p<0.05, log-rank test). The prediction of death was possible with 66.7% specificity and 57.1% sensitivity (PPV 53.3%, NPV 70%), but the distinction between high- and low-risk pts was significantly weaker than based on lymph node involvement (N0 vs. N1). Thus, for the final selection of genes this clinical variable was incorporated into the model as a covariate. Conclusions: Prediction of the risk of relapse in stage I-II NSCLC based on the gene expression profile is feasible, with NPV of 68.2%. No significant financial relationships to disclose.