Limitations of the case-only design for identifying gene-environment interactions

The case-only design, which requires only diseased subjects, allows for estimation of multiplicative interactions between factors known to be independent in the study population. The design is being used as an alternative to the case-control design to study gene-environment interactions. Estimates of gene-environment interactions have been shown to be very efficient relative to estimates obtained with a case-control study under the assumption of independence between the genetic and environmental factors. In this paper, the authors explore the robustness of this procedure to uncertainty about the independence assumption. By using simulations, they demonstrate that inferences about the multiplicative interaction with the case-only design can be highly distorted when there is departure from the independence assumption. They illustrate their results with a recent study of gene-environment interactions and risk of lung cancer incidence in a cohort of miners from the Yunnan Tin Corporation in southern China. Investigators should be aware that the increased efficiency of the case-only design is a consequence of a strong assumption and that this design can perform poorly if the assumption is violated.