Introduction

Tamoxifen (TAM), a selective estrogen receptor modulator, has been widely used to reduce the risk of breast cancer recurrence and prolong disease-free survival among women with non-metastatic hormone receptor-positive cancers. Adjuvant TAM use significantly lowers the risk of breast cancer recurrence and cancer-specific mortality [1, 2], and TAM use exceeding 5 years has been shown to afford further improvement in breast cancer outcomes [3]. Current American Society of Clinical Oncology guidelines recommend up to 10 years of adjuvant endocrine therapy for women with hormone receptor-positive cancer, composed of TAM, aromatase inhibitors, or a combination depending on menopausal status [4]. However, despite the established efficacy of TAM, among women who receive 5 years of adjuvant TAM therapy, approximately a third will experience a breast cancer recurrence and almost a quarter will die from their cancer within 15 years of diagnosis [1, 2]. Significant gaps persist in understanding the factors that modulate TAM response and the subsequent risk of adverse breast cancer outcomes.

TAM is a prodrug that exhibits relatively weak binding affinity for the estrogen receptor and must undergo biotransformation to achieve its pharmacologic effect [5]. Two metabolites exhibit much greater binding affinity for the estrogen receptor than TAM [6, 7] and one of these, 4-OH-N-desmethyltamoxifen (endoxifen), is found at especially high plasma concentrations and is thought to be TAM’s primary active metabolite [8].

The formation of endoxifen is catalyzed by the product of the CYP2D6 gene, CYP2D6, an enzyme central to the metabolism of many drugs [9]. Co-administration of TAM with medications metabolized through the same pathway has been shown to result in lower plasma concentrations of endoxifen [5, 8]. Concomitant use of TAM and CYP2D6-inhibiting medications has therefore been hypothesized to reduce TAM efficacy. CYP2D6-inhibiting medications span several major drug classes, and some of the strongest inhibitors are antidepressants, which are commonly used by women after breast cancer diagnosis for the treatment of depression, anxiety, and TAM- or chemotherapy-related hormonal side effects [10, 11].

Inherited genetic variation in CYP2D6 has also been shown to correlate with metabolic efficiency and to impact the concentration of TAM and its active metabolites in blood [5, 8, 12,13,14,15], normal breast tissue, and tumor tissue [16]. Up to 10% of the Caucasian population carries variant alleles implicated in near complete loss of enzymatic function (“poor metabolizers”) [17]. Though other enzymes are involved in TAM metabolism [18], between 27 and 58% of the variability in plasma Z-endoxifen concentration has been attributed to CYP2D6, depending on population and phenotype classification [13,14,15]; however, these studies consistently found that at least 90% of those classified as poor metabolizers had endoxifen concentrations at or below the level [19] recommended for inhibition of the estrogen receptor. The combination of CYP2D6-inhibiting medication use in women with impaired CYP2D6 metabolizer status has been shown to result in lower endoxifen levels than seen with either inhibiting factor alone [5, 8, 20] and may therefore be associated with an even greater risk for adverse breast cancer outcomes. Indeed, lower plasma endoxifen levels themselves have been shown to be negatively associated with survival in several studies [19, 21, 22], while others [23, 24] have found no association.

Results from studies of the association between CYP2D6 inhibition and adverse breast cancer outcomes have been heterogeneous, with several finding an increased risk [10, 25,26,27] (potentially only in relevant subgroups [28, 29]) and others finding a null association [11, 30,31,32,33,34,35,36,37,38,39]. However, the majority of studies have focused exclusively on either CYP2D6-inhibiting medication use or CYP2D6 genotype and have not considered their joint actions. In addition, most studies have employed limited genotyping strategies covering a small number of CYP2D6 alleles and few have assessed the full scope of CYP2D6-inhibiting medications, focusing instead on particular classes of drugs. Lastly, few have attempted to evaluate duration of concomitant use or the intensity of hypothesized CYP2D6 inhibition.

This study examined the associations of concomitant use of TAM and prescription CYP2D6-inhibiting medications (“inhibitors”) and CYP2D6 metabolic phenotype with adverse breast cancer outcomes in a well-characterized, population-based cohort of women with invasive breast cancer. Inhibitor use was assessed for all known prescription CYP2D6 inhibitors and further classified with respect to inhibitor strength and duration of concomitant use. In addition, this study applied a comprehensive CYP2D6 genotyping approach and assessed potential interactions between concomitant inhibitor use and metabolic phenotype on the risk of adverse breast cancer outcomes.

Methods

Study population

The study was conducted within the Quilt Study, a prognostic cohort of 2,337 women diagnosed with invasive breast cancer at ages 45–79 in the Seattle tri-county area between 1993 and 1999 that has been described previously [40]. Cases in the Quilt Study were originally ascertained through the Cancer Surveillance System (CSS), the Surveillance, Epidemiology and End Results (SEER) cancer registry serving the Seattle-Puget Sound region, as part of three prior population-based case–control studies: the Women’s Contraceptive and Reproductive Experiences Study (CARE, Seattle site only) [41]; the Puget Sound Area Breast Cancer Evaluation Study (PACE) [42]; and the Electric Power and the Risk of Breast Cancer (EMF) Study [43] (Fig. 1). The three case study populations were mutually exclusive of one another, and while population-based, the CARE and EMF studies sampled women with respect to race and county, and CARE additionally by age. As part of these three studies, information on demographics, tumor characteristics, and known risk factors for breast cancer incidence was obtained from the CSS and structured in-person interviews. As part of the subsequent Quilt Study, additional information on cancer treatment, recurrences, co-morbidities, prescription medications, and other exposures after diagnosis was collected through follow-up interviews and medical record reviews. Mean and median time from diagnosis to last chart review was 9.2 and 9.7 years, respectively.

Fig. 1
figure 1

Identification of study cohorts, including inclusion and exclusion criteria for each analysis. AA African American, CARE Women’s Contraceptive and Reproductive Experiences, EMF Electric Power and the Risk of Breast Cancer, ER estrogen receptor, Hx history, PACE Puget Sound Area Breast Cancer Evaluation, TAM tamoxifen

Full size image

The study population for this analysis was restricted to women in the Quilt Study diagnosed with local or regional stage estrogen receptor-positive (ER+) cancers, with no prior history of cancer at breast cancer diagnosis, with medical records available for review, and who used adjuvant TAM for at least 6 months after primary diagnosis and prior to any recurrence or second primary breast cancer (Fig. 1). Those women excluded due to lack of consent or unavailability of medical records were slightly younger, more likely to be non-White and peri-menopausal, to have local disease, and to be alive at end of follow-up, and slightly less likely to receive chemotherapy and radiation than women who met all study criteria (data not shown). The only statistically significant difference in demographic, tumor, or clinical/treatment characteristics between eligible women included and excluded in the study was in the proportion that received radiation therapy (61.6 vs. 68.9%, χ2 p = 0.012). Secondary analyses were restricted to the subset of women from the CARE and PACE studies with genotyping data on CYP2D6.

Data collection

Prescription medication use in targeted categories (including antidepressant/antianxiety medications, antihypertensives, and nonsteroidal anti-inflammatory drugs) and TAM use was abstracted from medical records. For each month following diagnosis and through the date of the last medical record review, women were classified as either having started, stopped, or after starting, continued using a medication in the absence of any evidence of discontinuation. CYP2D6 inhibitors were identified from the classifiers maintained by the U.S. Food and Drug Administration and Indiana University’s Clinical Pharmacology Research Institute, and further categorized with regard to inhibitor strength [44, 45]. Both entities classify inhibitor strength based on the change in in vivo plasma substrate concentration over time with and without co-administration of the inhibitor. Of the medications observed in our data, these two classifiers agreed on four out of five drugs classified as strong or moderate inhibitors, and on all drugs classified as strong inhibitors. For purposes of this study, medications were classified according to the highest inhibition level assigned by either classifier (Table 1).

Table 1 CYP2D6 inhibitors observed in study, by inhibitor strength and exposure classification
Full size table

DNA extracted from blood from women in the CARE and PACE studies was genotyped for eight common allelic variants in the CYP2D6 gene, as well as the number of copies of the gene, in the Fred Hutch Public Health Sciences’ Molecular Epidemiology Laboratory using validated TaqMan™ Drug Metabolism Genotyping Assay sets (Applied Biosystems, Foster City, CA), with the exception of the *9 allele. This allele was determined by fragment analysis on a 3730xl genetic analyzer, using a fluorescently labeled PCR product covering rs5030656 (primer set: 5′-GACCTGACTGAGCCCTTCCT forward, 5′ FAM-ATTCCTCCTGGGACGCTCAA reverse). Each single nucleotide polymorphism (SNP) corresponds to a known phenotype of tamoxifen metabolism, specifically extensive (EM), intermediate (IM), or poor (PM) metabolic efficiency (Table 2) [46, 47]. Because the SNP defining *2 occurs in multiple other alleles, we considered the *2 allele present only in the absence of those other allele-defining SNPs. Phenotypic diplotypes were determined as being extensive (EM/EM), poor (PM/PM), or intermediate (all others) and further collapsed into EM and IM/PM for analysis.

Table 2 SNPs used in determination of allelic variants and metabolic phenotype
Full size table

Breast cancer-specific mortality was identified from CSS, death certificates, or proxy report. Second breast cancer events (SBCEs) were defined as the first local, regional, or distant breast cancer recurrence or second primary breast cancer occurring at least 6 months after the initial cancer diagnosis. Since TAM broadly suppresses the proliferation of estrogen-sensitive tumor, CYP2D6 inhibition of TAM was hypothesized to have similar associations with risk of recurrence and risk of second primary breast cancer, although they were also explored separately. Data on recurrence were obtained from medical record review and interview, and data on second primary cancer were collected primarily from CSS. In addition, for women who died as a result of their breast cancer but for whom there was no record of distant recurrence, distant recurrence dates were imputed by subtracting from death dates the median period between first distant recurrence and death date in cohort members with available dates for both events. If the imputed date preceded the last recorded disease-free date, the later date was used instead.

Analysis

The relative risks of first SBCE and breast cancer-specific mortality associated with concomitant use of TAM and CYP2D6 inhibitors were estimated using Cox proportional hazards models. Women were followed from diagnosis for these events until their death, diagnosis with a non-breast cancer, date of last available medical record, or the end of follow-up on March 31, 2015, at which point they were censored. Because all women had to survive until their original case–control study interview, women entered the analysis (i.e., analyses were left-truncated) on their interview date (mean and median durations between diagnosis and interview were 280 and 242 days, respectively) or after 6 months of TAM use, whichever occurred later. SBCE analyses were also repeated without imputed distant recurrences.

Women who used TAM and a CYP2D6 inhibitor concurrently were considered exposed after 6 months of continuous or cumulative concomitant use and for the remainder of their follow-up time; until this point, they were classified in a time-varying manner as either having no concomitant use or as concomitant users of fewer than 6 months. Exposure assessment was conducted during the period between diagnosis and the first adverse breast cancer outcome; in the mortality analysis, exposure assessment was restricted to the period prior to the first SBCE in order to avoid ambiguous temporality. TAM users who had not used inhibitors concurrently with TAM served as the reference group for all analyses, although it was possible that these medications were used non-concurrently among these women. All analyses further restricted the exposed group to those whose concomitant use was of (i) strong or moderate CYP2D6 inhibitors or (ii) strong inhibitors alone.

Concomitant use of adjuvant TAM and an inhibitor for 6 or more months (main exposure) and duration of adjuvant TAM use were modeled as time-varying covariates. Analyses also adjusted for age at diagnosis (< 55, 55–69, 70+ years), BMI (prior to diagnosis; < 25, 25–29.99, 30+ kg/m2), tumor stage (local or regional), tumor grade (good, moderate, or poor differentiation, or undifferentiated), receipt of radiation, and receipt of chemotherapy. Adjusted Cox models were stratified by diagnosis year (1993–1995, 1996–1997, 1998–1999). All categorical covariates were modeled as dummy variables, and duration of prior adjuvant TAM use was modeled as a time-varying, continuous variable in years.

A subgroup analysis was conducted among women who underwent genotyping for the major CYP2D6 variant alleles from the CARE and PACE studies. Extensive metabolizers served as the reference group to which intermediate and poor metabolizers were compared. In the analysis of the interaction of concomitant inhibitor use and metabolic phenotype, phenotype (EM or IM/PM) and concomitant inhibitor use (none, < 6 months, 6+ months) were included in the model as dummy variable main effect terms along with terms for the interactions of concomitant inhibitor use category and phenotype. Risk estimates were derived using linear combinations of coefficients from the full model including all interaction terms.

Results

Nine-hundred sixty women from the Quilt Study cohort met study inclusion criteria (269 from CARE, 464 from PACE, and 227 from EMF, Fig. 1). Women who were concomitant inhibitor users for 6 or more months were slightly older than women without concomitant inhibitor use, were more likely to have been diagnosed in the years 1996–1997, had higher pre-diagnosis BMI, and were somewhat less likely to be postmenopausal than never concomitant users (Table 3). They were also more likely to be diagnosed at a regional stage, to receive chemotherapy, and had slightly longer durations of TAM use overall [median (IQR): 59 (36–61) vs. 57 (31–61) months].

Table 3 Baseline demographic, tumor, and treatment characteristics, by duration of concomitant medication use (never, fewer than 6 months, 6 or more months)
Full size table

There were 252 women who experienced a breast cancer recurrence or second primary breast cancer. Of these SBCEs, 19 were local recurrences, 3 were regional recurrences, 134 were distant recurrences (22 imputed), and 75 were second primary cancers; the remaining 21 SBCEs encompassed simultaneous events (e.g., a local and distant recurrence diagnosed at the same time). Two-hundred and twenty-two women were censored in the SBCE analysis at the time of their diagnosis with a non-breast cancer (n = 158) or death from another cause (n = 64). A total of 168 breast cancer deaths occurred during the follow-up period, with a mean cause-specific survival time of 14.2 years and a median of 16.6 years. In the mortality analysis, 214 women were censored at the time of their death due to another cause.

Approximately 70% of the observed follow-up time in the SBCE analysis occurred among never users of any concomitant CYP2D6 inhibitor, with an additional 8% among women who used a CYP2D6 inhibitor concomitant with TAM for fewer than 6 months, and the remaining 22% among women who used an inhibitor concomitant with TAM for 6 months or longer. The distribution of follow-up time was similar in the mortality analysis. Among women who used any CYP2D6 inhibitor concomitant with TAM for at least 6 months, approximately a third first used inhibitors at diagnosis, a third started them within 2 years following diagnosis, and a third initiated use more than 2 years after diagnosis. Approximately 45% of women who used inhibitors had no unopposed TAM use before beginning inhibitor use, and 27% did not begin inhibitor use until completing at least 2 years of unopposed TAM use.

Women who used CYP2D6 inhibitors concurrently with TAM for fewer than 6 months were considered to be at low risk for adverse events attributable to concomitant use, and the hazard ratios (HRs) associated with short-term concomitant use were not significantly different from 1.0 across all inhibitor strength categories (Fig. 2). Overall, 6 or more months of any CYP2D6 inhibitor use concomitant with TAM was not associated with any appreciable change in risk of SBCE or breast cancer mortality relative to non-concomitant users (Table 4; Fig. 2). Hazard ratio estimates were similar for SBCE and recurrences alone (data not shown), in line with the hypothesis that TAM inhibition would affect recurrence and second primary risk similarly; only results for the combined SBCE endpoint are presented. Among women whose first event was a second primary breast cancer, there were no significant differences in ER positivity by concomitant inhibitor use (χ2 p = 0.894), but information on ER status was available only for a small number of women (n = 42). Associations were not observed in any category of inhibitor strength for either SBCE or breast cancer-specific mortality (Table 4), and results did not differ when imputed distant recurrences were excluded. Concomitant inhibitor use appeared inversely related to the risk of adverse breast cancer outcomes, but the bounds of the 95% confidence intervals did not exclude the null.

Fig. 2
figure 2

Crude Kaplan–Meier survival curves showing time to earliest recurrence or second primary breast cancer by time-varying TAM-CYP2D6 inhibitor concomitant use. Because exposure status was assessed in the first 6 months following diagnosis, no events in this analysis could occur during this period. Individuals were classified as nonusers prior to first record of concomitant use, at which point they become exposed for < 6 months; after 6 or more months of cumulative concomitant use, they were considered exposed in the main analysis. Mos. months

Full size image
Table 4 Adjusted hazard ratios and 95% confidence intervals for the association of 6+ months concomitant inhibitor use with SBCE and breast cancer-specific mortality, by inhibitor strength and CYP2D6 metabolic phenotype
Full size table

Six hundred and sixty-five women had CYP2D6 genotyping results available, of which 111 (17%) were classified as EMs, 512 (77%) as IMs, and 42 (6%) as PMs. All metabolizer phenotype groups had similar distributions of concomitant use categories; overall, 64% were classified as never users and 26% were classified as concomitant users for 6 months or longer. No significant association between metabolizer phenotype and SBCE risk or breast cancer-specific mortality was observed among never users of concomitant inhibitors in adjusted analyses (Table 4). Use of CYP2D6 inhibitors was not associated with risk of SBCE or cancer mortality regardless of a women’s metabolic phenotype.

Discussion

Overall, this study found no increased risk of adverse breast cancer outcomes associated with concomitant use of CYP2D6 inhibiting medications and adjuvant TAM for 6 months or longer. No evidence of an increased risk of SBCE or cancer mortality was observed at any level of inhibitor strength. Furthermore, no association with metabolizer phenotype was observed among never users of concomitant CYP2D6 inhibitors, nor was there any evidence for an interaction of metabolizer phenotype and concomitant inhibitor use.

The main results of this study are consistent with other observational studies that have observed no association between CYP2D6 inhibiting medications, with or without additional information on CYP2D6 genotype, and risk of adverse breast cancer outcomes [11, 30,31,32,33, 37, 39]. Despite evidence that both use of CYP2D6 inhibitors and metabolizer phenotype are associated with reduced plasma endoxifen levels, only one observational study has observed a significant association between inhibitor usage and adverse clinical breast cancer outcomes [10].

Effects of impaired metabolic efficiency or pharmacologic inhibition of CYP2D6 on breast cancer outcomes have been hypothesized to be minimal, as TAM and its metabolites may still overwhelm estrogens in competition for ER binding sites in estrogen-sensitive tumor cells at standard doses [48]. If anything, deleterious effects of CYP2D6 inhibition are hypothesized to be most pronounced among pre-menopausal women [49], who have higher levels of endogenous estrogens than postmenopausal women and for whom TAM remains the preferred endocrine therapy. In an exploratory analysis, stratification by menopausal status revealed no difference in the association between concomitant inhibitor use and adverse breast cancer outcomes (data not shown).

This study has several notable strengths. As noted earlier, this study employed a population-based design and followed participants for events for up to 22 years. Population-based ascertainment of cases increases the generalizability of our findings, but also must be considered in light of nonparticipation or data unavailability. Overall, approximately 15% of women eligible declined to participate in the original case–control studies, and mortality among these women is likely to be greater, as has been shown in previous case–control studies of breast cancer [50]. In this analysis of the Quilt cohort, approximately 27% of women were excluded due to unavailability of medical charts. If the association between CYP2D6 inhibition and event-free survival differs in those not included in this analysis, this could introduce bias into our study.

This was also one of a few studies to assess CYP2D6 inhibition due to both medication use and genetic variation. The use of a broader genotyping strategy in this study represents a significant advantage over previous work and allowed for more precise classification of CYP2D6 metabolic phenotypes. Our study utilized germline DNA from blood rather than tumor, the specimen source for many past studies, alleviating possible concern regarding genotypic misclassification due to loss of heterozygosity in tumor tissue [51, 52]. In addition, this study included all prescription medications known to inhibit CYP2D6 activity in exposure assessment, rather than restricting to a single class of medications, and concomitant inhibitor use was classified with regard to intensity and duration of concomitant use, rather than relying on ever use of TAM and inhibitors. As prescribing practices for TAM changed over the study period, this approach allowed for variation in TAM usage among the study population when assessing concomitant use. However, it is unknown when during follow-up, after what duration, and for how long concomitant inhibitor use may affect risk of adverse breast cancer outcomes. Although longer durations of TAM have been found to provide greater clinical benefit [3], even 2 years of TAM use confers a long-term survival benefit relative to no use [53]. This study found that the majority of concomitant use began within 2 years of diagnosis, but we lacked sufficient data to explore the robustness of our findings to other risk periods and exposure definitions. Incorrect specification of the relevant exposure and/or risk period could explain our null findings.

This analysis is limited to prescription CYP2D6 inhibitors. While all inhibitors currently classified as having strong or moderate inhibiting activity are prescription medications likely to appear in medical records, several weak inhibitors are not prescription medications and therefore may not be captured (e.g., diphenhydramine). The use of medical records to classify drug exposures may also be less accurate than insurance claims or pharmacy fill data.

The present study is also limited by a small number of events among exposed women, leading to imprecise effect estimates and precluding assessment of associations with longer periods of CYP2D6 inhibition. The timing of inhibition and the duration of previous or subsequent unopposed TAM use may also affect long-term risk of adverse outcomes, but this study was underpowered to assess these factors.

Lastly, the only observed uses of moderate or strong CYP2D6 inhibitors in this cohort were of medications indicated for the treatment of depression or anxiety, conditions that may be independently associated with adverse cancer outcomes [54]. Despite concerns that this may have led to confounding by indication, no increased risk was observed.

Overall, this study found no evidence of increased risk of adverse breast outcomes among women who used CYP2D6-inhibiting medications concomitant with TAM, and no interaction of medication use with CYP2D6 metabolic phenotype. Findings were similar across all levels of inhibitor strengths and among pre-/peri- and postmenopausal women. If anything, there was suggestion of a reduced risk of adverse breast cancer outcomes in relation to concomitant inhibitor use, although the confidence limits for the risk estimates did not exclude the null. This could be suggestive of some degree of “healthy user/adherer” bias among long-term users of CYP2D6-inhibiting medications, which in this study were indicated primarily for the treatment of psychiatric and cardiac conditions. The suggested reduced risk associated with concomitant inhibitor use could be partially attributable to the influence of this uncontrolled confounding, as well as by other factors like social support and care access that were also not controlled for in this analysis.

Our findings are consistent with previous large observational studies and suggest that there is little clinical evidence at this time to support avoidance of CYP2D6-inhibiting medications among women using TAM endocrine therapy. It is possible that accounting only for CYP2D6 inherited variation and CYP2D6-inhibiting medication use does not fully capture factors impacting TAM efficacy, since additional enzymes are involved in TAM metabolism. A more comprehensive approach that can account for the entire TAM metabolic pathway may be needed to fully capture the effect of medication use and endogenous metabolic phenotypes on risk of adverse breast cancer outcomes.