Background: Many BC survivors report fatigue. The relevant genomic correlates of fatigue after BC are not well understood. We applied a previously validated machine learning methodology (Oh 2017) to GWAS data to identify biological correlates of fatigue induced after tx. Methods: We analyzed 3825 BC pts with GWAS data (Illumina InfiniumExome24 v 1.1) from the CANTO study (NCT01993498). The outcome of this study was post-tx fatigue 1 year after the end of primary chemotherapy/radiotherapy/surgery using the EORTC C30 fatigue subscale (overall fatigue) and the EORTC FA 12 fatigue domains (physical/emotional/cognitive). For each domain, we limited the study group to those with zero baseline fatigue and defined severe fatigue change as score increase above the third quartile. We tested univariate correlations between severe fatigue in each domain and 496539 SNPs as well as relevant clinical variables. The machine learning prediction model based on preconditioning random forest regression (PRFR) (Oh et al., 2017), was then built using the SNPs with ancestry adjusted univariate p-value < 0.001 and clinical variables with Bonferroni adjusted p-value < 0.05. The model was validated in a holdout subset of the cohort. Gene set enrichment analysis (GSEA) was performed using MetaCore to identify key biological correlates relevant to tx-induced fatigue. Results: Distinct results were found by fatigue domain (table). GSEA showed that the cognitive fatigue model SNPs included biomarkers for cognitive disorders (p = 1.6 x 10^-12) and glutamatergic synaptic transmission (p = 1.6 x 10^-8). Conclusions: A SNP based model had differential performance by fatigue domain, with a potential genetic role on risk and biology for tx induced cognitive fatigue. Further research to explore biomarkers of tx induced fatigue are needed.

Funding Source: French National Agency for Research (ANR), Susan Komen for Cure (CCR17483507) to IVZ