Sample Size

Power targets when ancestry groups contribute unequal samples

by Yun O.

Suppose a genome-wide association meta-analysis aims to detect a common-variant association while retaining several ancestry groups with sharply unequal sample sizes. Allele frequency, imputation quality and effect heterogeneity are uncertain across groups. For planning, should the target effect be a common clinically relevant effect, ancestry-specific effects drawn from a prespecified range, or the smallest effect detectable within each group? Which assumption gives the most defensible power calculation when diversity and total discovery yield may favor different allocations?

1
Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.

Crosspost to another Topic

Write your own title and commentary. The original is linked, not copied. To crosspost a crosspost, open its original first.

Comments

1
Sami N.

Power the ancestry-specific effects needed for the intended cross-group decision, using prespecified ranges for allele frequency, imputation quality, and heterogeneity, because a common effect can hide subgroup shortfalls.

Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.

1
Yun O.

For those prespecified ranges, would you require target power at the least favorable combination or average power over a distribution of plausible values?

Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.

0
Sami N.

If adequate detection within every ancestry group is required, I would use minimum power over a defensible joint scenario set as the primary constraint. That set should reflect plausible combinations, since combining every marginal extreme can create an implausible worst case. Assurance incorporates parameter uncertainty through prior distributions, as illustrated in [Williamson and colleagues’ cluster-trial study](https://pubmed.ncbi.nlm.nih.gov/39935065/). My recommendation for this ancestry setting is to use assurance as a complementary summary when those distributions are defensible: meeting an average target still permits low power in some scenarios, so it cannot establish the proposed minimum within each group.

Safety · report, block, mute

Blocking hides the author in your feeds and prevents direct replies between you. Muting hides a Topic. Public posts remain public.