Forum:Very high ratio of cases to controls - how to increase statistical power?

Question

[mfshiller]

I have a dataset with about 700 cases and only about 100 controls. That means that my power to find relevant SNPs through normal gwas analyses is very, very small (less than 10%). How can I circumvent this? Is there any literature out there regarding this? Haven't been able to find much. Appreciate any suggestions.

[karl.stamm]

Is this a population that is otherwise well studied? If a model organism, there's plenty of databases of variants you can filter with. Sometimes a case/control is just not going to work and you need a mendelian trio sort of study.

[mfshiller]

This is in humans. It's a complex trait so not really mendelian. Given the class imbalance, I'm not sure using logistic regression is advisable in this case. Fisher's exact test-based approaches should be more robust to class imbalance, so maybe that is the way to go.

[karl.stamm]

Can you bring in the ESP6500 and assume they're generally unaffected individuals? Not matched controls, but if your trait is low frequency then a general pop dataset will have the controls genotype.

[mfshiller]

That's actually one of the solutions I thought about. Either:

1. increasing sample size
2. correcting for class imbalance with oversampling
3. jackknife or bootstrap