Objective: We compared four haplotype-based approaches for the analysis of gene-environment interactions when haplotype-phase is ambiguous. The methods employ different versions of the expectation maximization algorithm and differ in the choice of the reference group and in the way the risk of disease is modeled (retrospective versus prospective). Furthermore, the methods are based on distinct assumptions (such as Hardy Weinberg equilibrium). The haplotype-based methods were also compared to single-marker logistic regression (LR).
Methods: We simulated case-control scenarios where the risk variant was directly genotyped (direct scenario) or in linkage disequilibrium with the genotyped markers (indirect scenario).
Results: The retrospective methods tended to be more powerful for detecting interactions than the prospective methods. In the indirect scenarios, the power of all methods was decreased. However, the power of the retrospectives methods was high in some scenarios and the interactions may only be detectable when using these approaches. Furthermore, we observed that the precision of one prospective method was clearly lower than the precision of the retrospective methods.
Conclusion: For the analysis of gene-environment (GxE) interactions in case-control data, the investigated retrospective methods can be an attractive alternative to haplotype-based methods which do not account for the retrospective sampling design.
Copyright 2009 S. Karger AG, Basel.