If the initial results found an increased risk per copy of, say, the A allele (additive model), then a significant increased risk for carriers of T allele (dominant model, in other direction), does not constitute replication. It is in principle possible that the direction of association can change due to differences in linkage disequilibrium across study populations. However, this “flip flop” phenomenon can occur only in very specific situations that are unlikely when the study populations have similar continental ancestry. [40] The burden of proof is on investigators to show evidence for how difference in LD in their study populations could produce a “flip flop” if they wish to claim replication, even though different alleles are associated with risk. Merely citing the possibility of “flipflopping” does not suffice.
