This is probably down to the impact the SNP has on the transcription/ translation process. You could use something like VEP to predict the effect of the variant, i.e. if the effect is deleterious, causes a shift in the reading frame, etc. If you don't have sequencing data and this was done on a chip, you could find the reference for that gene, alter the single base that you associate with risk, and run it through the VEP