In this population-based study, we observed no substantial association between CYP2D6 inhibition and breast cancer recurrence. The adjusted odds ratio associating no functional CYP2D6*4 allele with breast cancer recurrence in the ER+/TAM+ group (OR = 1.4, 95% CI = 0.84 to 2.3) approximately equaled the corresponding adjusted odds ratio in the ER−/TAM− group (OR = 1.3, 95% CI = 0.60 to 2.9). Taken together, these results should caution against overinterpretation of the former association because no association via the tamoxifen pathway is plausible to explain the latter association. The near-null association persisted regardless of whether we assessed CYP2D6 inhibition only by *4 genotype, by intake of medications that inhibit CYP2D6 function, by combination of *4 genotype with medication history, by limiting the ER+/TAM+ group to women who received no chemotherapy, by limiting the dataset to women with confirmed ER expression, or by analysis of the potential bias arising from misclassification of CYP2D6 function because we genotyped only the *4 allele.
The study design took advantage of the long-standing high-quality clinical database maintained by the DBCG (42
) and the tissue archives maintained by the pathology departments of Danish hospitals (59
). Linking these resources yielded a study effectively immune to selection bias and, to our knowledge, the largest study to date of the association between CYP2D6 inhibition and breast cancer recurrence. The 184 recurrent cases with at least one *4
allele in the ER+/TAM+ group represent approximately one-third of the accumulated cases with genetically identified reduced function reported in studies thus far.
Data quality was high, with perfect concordance between genotypes in DNA extracted from tumor tissue and lymph node tissue in the validation subset and with alleles at the CYP2D6*4
locus in Hardy–Weinberg equilibrium among all control subjects and consistent with the expected allele frequencies. Concordance of re-assayed ER expression with ER status at diagnosis was consistent with earlier reports (60
) and showed the expected associations with recurrence risk. The association between failure to express the ER at re-assay and recurrence risk in the ER+/TAM+ group provides the study design with face validity and demonstrates the study’s ability to detect associations between rare events and recurrence risk.
All recurrences in the validation subset were confirmed by medical record review, which is consistent with an earlier and much larger validation study that reported a positive predictive value of 99.4% for breast cancer recurrence recorded by the DBCG (62
). We observed perfect agreement for all of the matched factors except for one patient’s menopausal status. Only duration of tamoxifen therapy differed frequently between the registry data and medical records, with many patients who were originally assigned to tamoxifen therapy of short duration receiving tamoxifen for longer durations. This discrepancy likely resulted from modifications to the tamoxifen protocol occurring after the patient’s diagnosis and initial assignment to tamoxifen durations of 1 or 2 years. The discrepancy actually strengthens our study’s results because it suggests that the null association did not arise from some patients receiving short-duration tamoxifen therapy.
The major limitation of this study was absence of genotyping data for CYP2D6
alleles other than *4
. Complete genotyping of the DNA samples extracted from paraffin-embedded tissues required more resources than were available to us. To address this limitation, we implemented a quantitative bias analysis, which assumed that case patients with recurrence were more likely to carry alleles with reduced function than were control subjects. All of the parameters of the bias model were informed by published external data sources. Assuming an accurate bias model, this bias analysis showed that the near-null results would have changed very little with complete genotyping data. A second limitation was the absence of information on tamoxifen adherence by case patients and control subjects. About half of tamoxifen-treated patients fail to complete the intended duration of their tamoxifen therapy (63
). Among the 20 ER+/TAM+ patients included in our medical record review, six did not complete their intended duration of tamoxifen therapy, two because of recurrent breast cancer. Tamoxifen adherence is related to recurrence risk (64
) and may be predicted by CYP2D6
), in which case, adherence to the intended duration would be a causal intermediate between CYP2D6
genotype and recurrence. Results adjusted for adherence would be more biased than without adjustment, usually toward the null (66
Earlier studies of the association between CYP2D6 inhibition and risk of breast cancer recurrence or mortality in tamoxifen-treated women have reported widely heterogeneous results (55
), and there is no adequate explanation for this heterogeneity (18
). Although the heterogeneity presents an important barrier in interpreting a quantitative synthesis (18
), a recent meta-analysis concluded that the effect of CYP2D6 inhibition on recurrence risk may be relatively small (55
). This conclusion is consistent with our results, with reasonable bounds that can be placed on the expected association (18
), and with the well-understood pharmacology of tamoxifen therapy (18
), which requires that tamoxifen and its metabolites overwhelm estrogen in competition for binding to the ER. It is common for genetic research to initially show strong associations, sometimes in more than one study (69
), only to find that the strength of association diminishes or disappears as evidence accumulates (70
). Despite early enthusiasm for the potential to identify poor candidates for tamoxifen therapy by CYP2D6
), the true association between CYP2D6 inhibition and breast cancer recurrence risk in tamoxifen-treated patients is likely to be null or small.