PURPOSE: Interest is increasing in studying gene-environment (G x E) interaction in disease etiology. Study designs using related controls as a more appropriate control group for evaluating G x E interactions have been proposed but often assume unrealistic numbers of available relative controls. To evaluate a more realistic design, we studied the relative efficiency of a 1:0.5 case-sibling-control design compared with a classical 1:1 case-unrelated-control design and examined the effect of the analysis strategy. METHODS: Simulations were performed to assess the efficiency of a 1:0.5 case-sibling-control design relative to a classical 1:1 case-unrelated-control design under a variety of assumptions for estimating G x E interaction. Both matched and unmatched analysis strategies were examined. RESULTS: When using a matched analysis, the 1:1 case-unrelated-control design was almost always more powerful than the 1:0.5 case-sibling-control design. In contrast, when using an unmatched analysis, the 1:0.5 case-sibling-control design was almost always more powerful than the 1:1 case-unrelated-control design. The unconditional analysis of the case-sibling-control design to estimate G x E interaction, however, requires no correlation in E between siblings. CONCLUSIONS: In most settings, a matched analysis may be required and a 1:1 case-unrelated-control design will be more powerful than a 1:0.5 case-sibling-control design.
PURPOSE: Interest is increasing in studying gene-environment (G x E) interaction in disease etiology. Study designs using related controls as a more appropriate control group for evaluating G x E interactions have been proposed but often assume unrealistic numbers of available relative controls. To evaluate a more realistic design, we studied the relative efficiency of a 1:0.5 case-sibling-control design compared with a classical 1:1 case-unrelated-control design and examined the effect of the analysis strategy. METHODS: Simulations were performed to assess the efficiency of a 1:0.5 case-sibling-control design relative to a classical 1:1 case-unrelated-control design under a variety of assumptions for estimating G x E interaction. Both matched and unmatched analysis strategies were examined. RESULTS: When using a matched analysis, the 1:1 case-unrelated-control design was almost always more powerful than the 1:0.5 case-sibling-control design. In contrast, when using an unmatched analysis, the 1:0.5 case-sibling-control design was almost always more powerful than the 1:1 case-unrelated-control design. The unconditional analysis of the case-sibling-control design to estimate G x E interaction, however, requires no correlation in E between siblings. CONCLUSIONS: In most settings, a matched analysis may be required and a 1:1 case-unrelated-control design will be more powerful than a 1:0.5 case-sibling-control design.