If the odds ratio of a SNP is < 1, does that mean the minor allele is protective?

Question

[maya123z]

I am reading a paper (link here) where they did a GWAS looking for Alzheimer's-associated genes. In their Table 2, a lot of the associated SNPs have an odds ratio that's less than 1. Does that mean that the minor allele is protective against Alzheimer's? Or, is this just because these are raw odds ratios and do not account for other covariates that were included in the main GWAS?

Answer 1

[Kevin Blighe]

Yes, 'protective' in the sense that having the variant statistically means that you have a reduced 'risk' (or lower 'odds') of having an Alzheimer's diagnosis, but in the confines of those subjects who were in the study. As the OR approaches 0, the level of reduced 'risk' is strengthened. The way that the study's subjects are diagnosed is obviously important. Other criteria are also important in order for the results to be applied to the wider population.

Thus, there are very important things to consider here before one makes any wild statements about such variants:

1. your study should be sufficiently powered such that interpretation of the statistical results can be done with confidence
2. if the upper confidence limit interval for the variant goes above 1, then it is of less interest and this will reflected in th associated p-value
3. if it doesn't have a statistically significant p-value, then it's not interesting; moreover, the p-value should reach genome-wide siginificance
4. Replication - can the results be replicated in different cohorts / stages?

From what I can see, the Table 2 variants that have OR<1.0 (and the study, generally) fulfill these criteria.

Kevin

[maya123z]

Thank you for the detailed answer! I'm curious, how trustworthy can we consider the odds ratios in this paper, considering they don't seem to be correcting for any covariates?

[Kevin Blighe]

Well, one must realise that we only have to correct for covariates when there is evidence of bias (known or unknown) and / or if there is a known confounding factor from the study design. One of the primary concerns is always ethnic differences, but the authors have controlled for that by just including non-Finnish Europeans.

I am not an Alzeimher's researcher, so, I don't know the common confounding factors that are considered. In cancer, we would adjust for things like smoking, BMI, etc. However, I look at the reputation of the journal and the authors and feel content that the results can be interpreted with confidence. This actually looks like one of the better GWAS studies, I must admit. All that said, GWAS studies are renowned for being 'non-reproducible', but there are notable exceptions.

It's all just statistics for now, i.e., until they prove a mechanism through which the variants exert their protective effects.

[NB]

I know this is irrelevant now but being a part of this project, I feel the need to mention that we did correct for covariates such as age, sex and principal components to adjust for possible population stratification. It is also mentioned in the methods section of the paper