Comparison of the power and statistical efficiency of case-control and case-only designs in estimating the gene-environment interaction in breast cancer

Limitations of the traditional methods for assessing G*E interaction- including case-control studies- led to development of several non-traditional approaches. This study aims to assess the interaction between the genetic background (history of breast cancer in first degree relatives) and environmental influences (reproductive/menstrual factors) in patients with breast cancer; we also compare the statistical efficiency and power of case-control and case-only designs in this setting. In a matched case-control study in Mazandaran province (Iran), 250 incident biopsy-proven cases of breast cancer and 250 age-matched neighbor controls were interviewed. History of breast cancer in mother and/or sister(s) was taken as a surrogate measure of genetic predisposition, while age at first birth, parity, breast feeding, age at menarche and irregular menstruation were considered as relevant environmental factors. For the matched case-control design, we used a conditional logistic regression model to examine main effects and the G*E interaction. In the case-only design, logistic regression analysis was applied to obtain an estimate of G*E interaction, after checking for the independence assumption. We also calculated the power for detecting the interaction by matched case-control and case-only analyses. Age at first delivery did not meet the assumption of independence (p=0.02), and so was not included in the case-only analysis. No statistically significant interaction effect was seen in the case-control analysis, while case-only analysis showed significant negative interaction between disease in first-degree relatives on the one hand and parity and breast feeding on the other. We also detected a significant positive interaction between genetic predisposition and age at menarche. All the estimated 95% confidence intervals for OR in G*E interactions were narrower in the case-only analysis. For all factors, the power for detecting G*E interaction was greater in the case-only analysis compared to the case-control analysis, with ratios ranging from 1.08 to 2.23. The case-only design is more efficient and powerful than the case-control design for detecting gene-environment interaction under the assumption of independence. Baseline disease risk, interactions and independent effects should be considered in using the control data for checking the assumption of independence. Considering the existence of another explanatory variable, eg. a mutant gene which may have passed unnoticed, would be the safest approach in a case-only study.