Re: CYP2D6 Genotype and Tamoxifen Response in Postmenopausal Women With Endocrine-Responsive Breast Cancer: The Breast International Group 1-98 Trial and Re: CYP2D6 and UGT2B7 Genotype and Risk of Recurrence in Tamoxifen-Treated Breast Cancer Patients

the role of common and rare genetic variation in CYP2D6 in determining response to tamoxifen therapy in women with breast cancer has been controversial for nearly a decade. In two recent articles by regan et al. (1) and rae et al. (2), no association between CyP2D6 metabolizer status predicted on the basis of genotype and outcome in women with breast cancer treated with adjuvant hormone therapy was demonstrated. In an accompanying editorial (3), Kelly and Pritchard suggest that the controversy has finally been laid to rest. So what is different about the null findings of the new studies (1,2) that ends a controversy that previous studies have not [eg, (4,5)]? A key factor influencing the judgment of Kelly and Pritchard seems to be the fact that the studies were nested within randomized controlled trials. Indeed, they state that an important lesson to be learned from the controversy is that the use of randomized trials prospectively designed and retrospectively analyzed is critical to assess the role of biomarkers such as CyP2D6. the assumption that randomized trials are required to avoid bias in this context is a fundamental error in understanding of basic epidemiological principles. the primary purpose of randomization is to minimize the chance of correlation between the exposure of interest and potential confounding variables by randomly assigning the exposure of interest. In a genetic association study, the exposure of interest is the genotype, not the therapeutic arm. It is highly unlikely, but possible, that genotype will be correlated with other possible confounding variables. the two new studies are not randomized controlled trials of the exposure of interest. Indeed, it is not possible to randomize germline genotype or any other biomarker. one potential benefit of evaluating the role of germline genotype or other biomarker for outcomes by nesting the study within a randomized controlled trial of another exposure is that other important factors may have been carefully measured. We believe that purely observational studies may have equally good standardization of the population of interest and the outcome of interest. Kelly and Pritchard did not discuss some flaws in the study designs that may render the findings invalid. A major limitation of both new studies is the use of tumor DNA to determine germline genotype. one measure of genotyping quality is duplicate concordance. this was not reported by regan et al. (1). Consistency of genotype frequencies with those expected under hardy–Weinberg equilibrium (hWe) is another measure of genotyping quality that ought to be standard reporting practice. the proportions of the three CYP2D6*4 genotypes (rare homozygous [8.6%], heterozygous [20.5%], and common homozygous [70.9%]) in 3828 patients are provided in table 2 (1). A standard χ2 test shows highly statistically significant deviation of these frequencies from those expected under hWe (χ2 = 416, P = 10−92). Deviation of this magnitude would suggest a severe measurement error. there are two possible reasons for this problem. First, CYP2D6 has a pseudogene, and it is known that this pseudogene can interfere with an assay for the CYP2D6*4 allele. Second, loss of heterozygosity in a tumor would result fewer heterozygotes than expected, as is observed here. rae et al. (2) mentioned that the allele frequencies were in hWe. this statement is technically incorrect as hWe relates to genotype frequencies and not allele frequencies. Furthermore, the test statistic and P value are not given and genotype frequencies are not provided, so it is not possible for the reader to verify the calculations. the rae et al. study (2) comprised 588 patients treated with tamoxifen, of whom just 70 died from breast cancer during follow-up. No power calculation was given, but a study of this size would have approximately 50% power at a type I error rate of 0.05 to detect a hazard ratio of 1.5 in a high-risk group comprising 30% of the cohort. We published a large observational study that comprehensively evaluated CYP2D6 genotypes in a large, well-conducted, but not perfect, observational study 2 years ago (4). In terms of patient sample numbers (3155 patients), our study was larger than any previously published and is larger than both the new studies combined. We also found no evidence for predicted CyP2D6 metabolizer status and breast cancer– specific mortality. the article was not cited in either of the new articles or the editorial. Further support for our findings was published by Lash et al., although this study also suffered from the limitation of using tumor-derived DNA (5). If the valid results from observational studies had ended the controversy, some patients may have been spared the unnecessary expense of having a test of no clinical utility.