Let's say you are analysing 3 contrasts together with mitch, each gene set will have enrichment scores for each of the 3 contrasts. When you prioritise by SD, you are looking for gene sets that have highly discordant regulation in the 3 contrasts. This can be interesting if you are doing single cell RNA-seq analysis and interested in how the different cell types are behaving in disease. https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-020-06856-9