Relatively few studies have evaluated the role of germline genetics in predicting outcome in FL and most of these were small candidate gene studies. In addition, adapting statistical designs used in susceptibility GWAS for outcomes is challenging. In fact, at the genome-wide scale, discoveries from one cohort may fail to replicate in other cohorts due to both cohort heterogeneity, the « winner curse » phenomenon, and limited statistical power. To identify robust and reproducible SNPs associated with event-free survival (EFS) of FL patients, we developed a specific cross-validation statistical design in a multi-cohort study in order to better balance statistical power across cohorts and identify robust findings that replicate across cohorts. We conducted a GWAS study of FL patients treated with immunochemotherapy for EFS with adjustment for population stratification using, FLIPI score, and use of rituximab maintenance. Cohorts were from prospective clinical trials of the Lymphoma Study Association (LYSA, PRIMA [N=391], RELEVANCE [N=396], the Fondazione Italiana Linfomi (FIL, FOL05, [N=199]) and the Iowa/Mayo Clinic Lymphoma Specialized Program of Research Excellence (SPORE) (N=154), a prospective observational study. Genotyping for the GWAS was performed using Human Core exome (PRIMA, RELEVANCE, FOL05) and 660 Quad (SPORE) BeadChips (Illumina, San Diego, USA). Minimac4/1000G was used to impute additional SNPs. Statistical design is based on cross-validation statistical technique. Following a « leave-one-cohort-out » (LOCO) strategy, J train blocks are defined by combining iteratively the cohorts except cohort j which is left for validation purpose. A p-value threshold is defined to select the top SNPs from each train block, with subsequent validation based on p-value and hazard ratio estimate in the cohort that is left. Discordances between SNP hazard ratios from train and validation blocks indicates lack of reproducibility. At the end, SNPs are considered robust if they are discovered across several train blocks and validated with consistency. After standard quality control steps for genotyping, PCA for ancestry and population stratification (no exclusion) and imputation, 1,861,658 common SNPs for the five cohorts (N=1,140 patients) were analyzed. We will present in detail the cross-validation LOCO statistical design. Preliminary results for EFS indicated promising loci at chromosome 7 that replicated consistently across 4 out of the 5 train blocks in the unadjusted and adjusted Cox models. The LOCO design especially helps identifying SNPs with inconsistent results across cohorts. Some examples of divergent results across cohorts will be presented. The LOCO design can be used successfully to identify more robust results that are consistent across cohorts in GWAS, particularly in survival studies. Mots clés GWAS ; follicular lymphoma Déclaration de liens d'intérêts Les auteurs n'ont pas précisé leurs éventuels liens d'intérêts