Menopausal Hormone Therapy and Breast Cancer Risk in the NIH-AARP Diet and Health Study Cohort

A significant elevation in breast cancer risk among users of menopausal estrogen plus progestin therapy (EPT) has been observed across multiple studies (1), including the recently conducted Women's Health Initiative (WHI) clinical trial (2, 3). When results from this trial were announced in 2002, clinical guidelines were reformulated to recommend that women use hormones at low doses for short periods only for symptom management, leading millions of women to discontinue usage (4). The strength of evidence between EPT use and increased breast cancer risk was reinforced when population data subsequently showed marked declines in breast cancer incidence (5, 6).

Limited data exist on how risks change after cessation of use, but resolution of this is now more important than ever. Further, questions exist about the role of unopposed estrogen therapy (ET), which has been linked to elevated breast cancer risks in numerous observational studies (7) but not in WHI (2). WHI data surprisingly showed somewhat of an inverse relationship (2), although recent analyses suggest that this might reflect more extended intervals between menopause and age at first use of hormones among trial participants (8). With respect to EPT, results from WHI focused on continuous regimens of 0.625 mg conjugated equine estrogen plus 2.5 mg/d medroxyprogesterone acetate, but data relating to other regimens are limited. In particular, it has not yet been possible to resolve whether usage of progestins sequentially affects risk differently than continuous usage (9-16).

Although many investigations support that hormones exert stronger relationships among thin women (13, 17-23), it is unclear whether the increasing prevalence of obesity in the United States could explain some of the divergence of results between more recent studies and earlier investigations, which nearly all showed increased breast cancer risks related to ET (1, 17). Whether there continues to be an increased risk associated with certain hormone usage patterns among heavier women is also unclear.

Increasing evidence suggests that hormone associations may be dependent on tumor characteristics, including stage or grade (24-26), histology (9-13, 25, 27-34), and hormone receptor status (12, 19, 24, 26, 27, 33-35), although results have not been entirely consistent. Whereas most studies show that hormone associations are generally stronger for tumors with favorable clinical characteristics, WHI showed the reverse (36). Some of the difficulty in resolving tumor subtype associations may reflect that most investigations have assessed relationships with clinical characteristics in isolation of each other despite there being a high degree of intercorrelation. For instance, recent studies have suggested stronger hormone associations for lobular than ductal tumors (25, 27, 29, 31, 32), but most have not considered how relationships by histology are affected by estrogen receptor (ER) positivity, also linked to enhanced hormone associations.

To clarify these unresolved questions, we evaluated relationships within the large prospective NIH-AARP (formerly known as the American Association of Retired Persons) Diet and Health Study.

The NIH-AARP Diet and Health Study Cohort was established in 1995 to 1996 when a questionnaire requesting information on demographic characteristics, dietary intake, and health-related behaviors was sent to 3.5 million AARP members (37). Recipients of the questionnaire included members aged 50 to 71 years who resided in one of six U.S. states (California, Florida, Louisiana, New Hampshire, North Carolina, and Pennsylvania) or two metropolitan areas (Atlanta, GA, and Detroit, MI). A total of 617,119 (17.6%) persons returned the questionnaire, with 567,169 (16.2%) satisfactorily completing it. In 1996 to 1997, a second risk factor questionnaire was sent to collect additional information on diet, family history of cancer, anthropometry, physical activity, and use of menopausal hormones, with a 59.4% response rate (n = 337,074). After excluding participants who died (n = 1,619), moved out of the study area before their second questionnaires were received and scanned (n = 547), had proxies complete their baseline (n = 6,959) or second (n = 3,424) questionnaires, or were male (n = 188,117), 136,408 potentially eligible women remained. The Special Studies Institutional Review Board of the National Cancer Institute approved this study.

The second questionnaire, administered in 1996 to 1997, collected detailed data on ever-use, dates of first and last use, total duration of use, regimen, usual dose, and pill used for the longest period for both estrogens and progestins. Current usage was defined according to usage patterns at the time of administration of this questionnaire. Subjects were asked about individual years of usage up to 10 years, with longer-term use collected only as ≥10 years. We used reported ages at first and last use to estimate longer-term usage for selective analyses.

We considered women who reported taking both estrogen and progestin pills to have used EPT if the reported dates of first use were within 90 days of each other or if reported durations of use were identical. Such women were further subdivided according to whether their usual pattern of use of progestins was sequential (for <15 days per cycle) or continuous (every day of the cycle). The 408 and 2,058 women who reported taking progestins for 15 to 19 or 20 to 25 days per cycle were categorized as having used the continuous regimen.

Cohort members were followed annually for address changes and vital status. Address changes were identified through linkage to the U.S. Postal Service National Change of Address database, U.S. Postal Service updates received with undeliverable mail, use of other address change update services, and participants' notifications. Vital status was updated through linkage to the Social Security Administration Death Master File and verified by the National Death Index.

Based on annually updated residence information and using first and last names, address, sex, date of birth, and social security number obtained from the baseline questionnaire, incident cases of in situ and invasive breast cancers were identified by probabilistic linkage to the cancer registries in the eight states from which study subjects were derived. All suspected matches underwent review to reject the potential matches that were unlikely to be true (an estimated 4%), and uncertain matches underwent final manual review. An earlier validation study that compared registry findings with self-reports and medical records estimated that linkage validly identified ∼90% of all incident cancers among study participants (38). The cancer registry ascertainment area was recently expanded to include three additional states (Texas, Arizona, and Nevada) to capture cancers occurring among participants who moved to those states during follow-up.

Dates of diagnosis and tumor characteristics were obtained from the cancer registries. Using histologic codes from the International Classification of Diseases for Oncology, Third Edition, we classified breast cancer into ductal (code 8500 or 8523), lobular (codes 8520 or 8524), and mixed ductal/lobular tumors (code 8522).

We excluded 9,022 women who reported a history of cancer other than nonmelanoma skin cancer on either questionnaire (including 942 breast cancers), 745 women who were missing all information on menopausal hormone use, and 3 women with no follow-up. Analyses therefore focused on 126,638 women.

Study entry and follow-up began at the age at which the second questionnaire was scanned, and continued until June 30, 2002 or the earliest of the following: participant diagnosed with breast cancer, moved out of the registry catchment area, or died from any cause. The censoring or exit date was chosen to account for declines by ∼50% in hormone use (39) subsequent to release in July 2002 of findings from the WHI that linked EPT with increased breast cancer risks (40). During follow-up in our study, a total of 3,657 women developed breast cancer (607 in situ cancers, 3,035 invasive cancers, and 15 with missing stage information).

Because the Florida, Pennsylvania, and Michigan state cancer registries did not collect hormone receptor information, ER and progesterone receptor data were available for only 1,451 (39.7%) and 1,403 (38.4%), respectively, of the breast cancer cases. Analyses regarding hormone receptor status excluded women from these three states. There were no substantial risk factor differences between women from states where hormone receptor information was available and states where it was unavailable.

We used Cox proportional hazards regression (using SAS 8.2 software; SAS Institute), with age as the timescale and ties handled by complete enumeration (41), to estimate the relative risk (RR) and 95% confidence intervals (CI) of developing breast cancer. Tests of the proportional hazards assumptions for exposures and other variables included in statistical models revealed no departures.

Although most women who use ET have had a hysterectomy, older women with intact uteri had opportunities to take ET before added progestins became routine. We therefore analyzed ET associations in the entire cohort as well as in the 51,237 women with hysterectomy at baseline. We limited the assessment of EPT to the 73,986 women with intact uteri at baseline. A total of 1,415 women were excluded from these analyses due to missing data.

We initially evaluated potential confounding by all identified risk factors but ultimately chose a parsimonious combination of variables that were associated with both exposure and outcome and changed the hormone therapy variable estimates compared with estimates from models adjusted for only age at entry. Our statistical models adjusted for age at entry, race/ethnicity, age at first birth, menopausal status (including oophorectomy status), age at menopause, number of breast biopsies, family history of breast cancer in a first-degree relative, and number of mammograms in the 3 years preceding the second questionnaire. Adjustment for additional risk factors, including years of education, parity, physical activity levels, alcohol consumption, and usage of nonsteroidal inflammatory drugs, had minimal effects on hormone risks. To assess possible biases associated with unknown types of or ages at menopause (including those with simple hysterectomies; refs. 42, 43), we conducted sensitivity analyses with different assumptions. Our results were unchanged by eliminating those women who were unable to report ages at menopause or by setting age at menopause among women with simple hysterectomies to the date of first usage of menopausal hormones.

The 126,638 women contributed 671,546.2 person-years. The mean durations of follow-up (and ranges) were 2.76 years (0.003-5.62) for those who developed breast cancer and 5.38 (0.003-5.67) for those who did not. The mean ages at entry and exit were 62.6 and 67.9 years, respectively. The standardized incidence ratio for breast cancer among AARP members compared with the National Cancer Institute Surveillance, Epidemiology and End Results rate (ages 50-79 years) was 1.10 (95% CI, 1.07-1.13).

Most women in the cohort were White, postmenopausal, and in their 60s when they completed the second questionnaire (Table 1). Whites were at a higher risk than non-Whites. Breast cancer risk was positively associated with age at first birth, age at natural menopause, body mass index (BMI), number of breast biopsies, a family history of breast cancer, number of recent mammograms, and number of alcoholic drinks per day. Inverse relations of risk were observed with age at menarche, parity, and age at bilateral oophorectomy.

A total of 38.8% of the person-years were contributed by women who never used hormones, whereas 27.9% were contributed by ET-only users and 21.8% by EPT users (either EPT alone or EPT use preceded by ET, with both types of users having similar risks). The remaining women (11.5%) used other combinations of use of hormones (PT followed by EPT, ET followed by PT, PT followed by ET, EPT followed by PT, EPT followed by ET, EPT unknown, other combinations, or unknown).

Compared with nonusers of hormones, women who had used hormones were more likely to be White, married, oral contraceptive users, and thin (BMI < 25 kg/m²) and to have had multiple breast biopsies. ET users were more likely to have had a surgical menopause and given birth at younger ages. EPT users were more likely to be younger, be college graduates, have reported more frequent recent mammograms, and be in excellent or very good health at baseline.

Among all women, ET-only use was associated with a RR of 1.15 (95% CI, 1.04-1.27; Table 2), with no further evidence of increase seen for longer durations of use. Current users were at a slightly higher risk (RR, 1.24) than former users (RR, 0.99). There was no evidence that risk rose with increasing durations of use among current users. Among formers users, no trend in risk was apparent according to time since last use, although recent quitters were at reduced risk (RR, 0.71; 95% CI, 0.51-0.98). Analyses restricted to women with a hysterectomy at baseline confirmed only weak associations with ET, although estimates were slightly lower than those observed among all women.

Among women with intact uteri, EPT use was associated with a significantly increased risk (RR, 1.82; 95% CI, 1.65-2.01; Table 3). The increased risks were driven by high risks associated with current usage, with women having discontinued hormone use ≥5 years previously showing no evidence of any increased risk. Among current users, risks increased with extended durations of use, rising to 2.44 (2.13-2.79) among women exposed for ≥10 years. Women who had used progestins sequentially (<15 days/mo) were at a slightly lower risk (RR, 1.63; 95% CI, 1.42-1.86) than those who had used them continuously (RR, 1.98; 95% CI, 1.78-2.21). These differences became even more marked when attention focused on current long-term users (≥10 years), with the risks rising to 2.01 (1.65-2.45) for sequential users and 2.76 (2.35-3.25) for continuous users. Risk did not vary in any systematic way with dose of or actual days progestins were used apart from there being somewhat higher risks for those who used progestins for ≥10 versus <10 days.

Analyses restricted to the most common modes of therapy, namely sequential regimens containing 5.0 or 10.0 mg/d medroxyprogesterone acetate or continuous regimens containing 2.5 or 5.0 mg/d medroxyprogesterone acetate, produced similar associations to those of all reported sequential or continuous regimens (data not shown).

The risks associated with ET (among all women) and EPT (among women with intact uteri) were evaluated further according to the presence or absence of breast cancer risk factors (Table 4). The risks associated with hormone use did not vary substantially by most risk factors. However, there were significant interactions between body mass and both years of ET and EPT (interaction P values for both of 0.01). In thin women, ET use of ≥10 years was associated with a RR of 1.60 (95% CI, 1.31-1.96), although there was no increased risk among the heaviest women. Extended EPT use led to nearly a 3-fold increased risk among the thin women, but even among the heaviest women, there was an ∼2-fold increased risk.

Relationships with hormone use were also explored according to tumor characteristics (Table 5). ET relationships did not vary substantially or in any systematic manner by tumor stage (in situ versus invasive) or ER status (negative versus positive). However, relationships with ever and current usage were slightly stronger for low-grade and mixed ductal/lobular tumors albeit without evidence of a trend in risk with increasing durations of use.

Although EPT associations also did not vary substantially by tumor stage, there was some variation in risks by the other clinical characteristics. Whereas there was no increased risk associated with long-term use for ER- tumors, over a 3-fold increased risk was seen for ER+ tumors. Further, higher hormone-associated risks were seen for low-grade and mixed histology tumors. Whereas long-term EPT use related only to a 1.7-fold increased risk of grade 3 tumors, the elevation in risk for grade 1 tumors approached a 5-fold excess. A similar excess 5-fold increased risk associated with long-term use was seen for mixed ductal/lobular tumors, which contrasted with RRs of 2.54 for ductal and 1.37 for lobular tumors.

The majority of the tumors with available hormone receptor data were ER+, with somewhat higher rates of ER positivity seen for lobular and mixed ductal/lobular tumors (Table 6). The high correlation between the various tumor markers made it difficult to discern the independence of effects. However, among women with intact uteri, there appeared to be no relationship of EPT use to risk of ductal tumors that were ER- (RR, 1.02) but over a doubling of risk for such tumors that were ER+ (RR, 2.61; 95% CI, 2.10-3.24). A similar increase in risk associated with EPT use was seen for ER+ lobular tumors, but even higher risks were observed for ER+ tumors with mixed histologies (RR, 6.37; 95% CI, 3.45-11.79). Similarly elevated risks associated with EPT use were seen for low-grade, ER+ tumors. Somewhat less predictive of risk were combinations of tumors defined jointly by histology and grade.

The cumulative incidence of developing breast cancer in this population was 54.5 per 10,000 person-years. Among all women, ET users had a 9% higher cumulative incidence compared with non-HT users (48.8 versus 44.7 per 10,000 person-years). Among women with intact uteri, the cumulative incidences compared with nonusers were 75%, 59%, and 89% higher, respectively, for all EPT users, sequential users and continuous users (76.4, 69.5, and 82.6 per 10,000 person-years, respectively) compared with non-HT users (43.7 per 10,000 person years).

Results from this large prospective study confirmed elevations in breast cancer risk associated with current usage of menopausal hormones. The highest risks were seen among current, long-term EPT users, although ET use also led to elevated risks among thin women. Risks quickly dissipated after discontinuation of hormone use, with no increases in risk seen ≥5 years after discontinuation.

These results are generally consistent with other recent prospective (18-20, 22, 27, 33, 44-46) and case-control (13, 16, 29, 30, 32, 47) studies and the growing consensus that EPT use is a breast cancer risk factor. However, somewhat reduced risks seen for ET in WHI starkly contrasted with results from many observational studies. This included data from the large Collaborative Group on Hormonal Factors and Breast Cancer (17), which showed clear increases in breast cancer risk among current long-term ET users. Further confusion regarding the relationship has arisen because some recent U.S. observational studies failed to show increased risks of ET use related to breast cancer risk (7, 9, 10, 12, 15, 16, 48). These studies contrast with most studies in Europe, which have shown fairly consistent risk increases associated with ET use (18, 21, 27, 44, 46). This has raised questions as to whether such differences could reflect variations in the types of estrogens prescribed in the United States versus Europe (conjugated equine estrogens versus the more potent estrogen 17β-estradiol), although results from the recent Million Women Study (18) found no substantial differences in risk between the two estrogen types.

In the United States, use of ET and its typical users have changed in the last 30 years. Although today ET use is generally limited to women with hysterectomy, this was not the case before the mid-1990s (4). Thus, much of the early literature on ET and breast cancer focused on women with intact uteri (17). These temporal changes occurred concurrently with a change from use for symptoms to use for potential disease prevention (49) and with trends of increasing obesity (50). Thus, compared with earlier studies, ET users in our study included more hysterectomized and heavy women, both factors that would tend to attenuate ET risks.

Our results and those from most other contemporary investigations (13, 17-23) indicate that a complete assessment of hormone effects must assess modifying effects of body size. Weak associations between ET and breast cancer were in stark contrast to nearly a doubling in risk for EPT. However, among thin women, we found that long-term ET use was associated with a 60% significantly elevated risk, similar to findings of most European investigations (18, 21, 44, 46). EPT relationships were also stronger among thin women, but of note was that EPT continued to exert increases in breast cancer risk even among heavier women. Thus, even among the heaviest women in our study, we still observed a 2-fold increased risk associated with long-term use of EPT.

The growing trends toward more American women becoming obese (50) thus raises questions as to whether hormone effects may have been underestimated in recent U.S. studies, including the WHI trial where >80% of study subjects were classified as having heavier than ideal body masses. In the WHI trial, some attempts were made to address effect modification by body size; EPT relations were somewhat stronger among thin women (51), but this was not true for ET (52). These analyses, however, were limited by small numbers of thin women and short follow-up periods. Additionally, it has recently been shown that much of the lower breast cancer risk seen in the WHI unopposed estrogen trial was among women who had not previously used hormones (8), resulting in adjustment for the time from menopause to first use of ET raising risk estimates to those of the trial's observational subcohort data. This further complicates the assessment of the extent to which hormone effects may have been underestimated by an inability to adequately take weight into account. Thus, resolution of the effect of obesity on breast cancer hormone relationships (both ET and EPT) will require additional large observational studies.

An additional difficulty in resolving associations of ET on breast cancer risk in observational studies is that this mode of therapy is now recommended only for women who have undergone a hysterectomy, many of whom also keep their ovaries (53). It is impossible to determine a “true” age at menopause for women who undergo a simple hysterectomy. The resulting potential bias to estimates of hormone associations has received extensive previous attention (42, 43). However, there is no single ideal approach for addressing this problem, although women with simple hysterectomy comprise a substantial proportion of all hormone users (especially ET users). We conducted different sensitivity analyses to assess the extent of this bias and were not convinced that our risk estimates were confounded by imprecise information. Because of the older age of our respondents, we also had a unique opportunity to compare ET results between those who received it in conjunction with a hysterectomy and those who never had any surgical intervention, who comprised a sizeable number of all of our ET users. The lack of difference in results between these two groups was reassuring. Nonetheless, it is possible that our results for ET were somewhat biased because of the inability to precisely identify age at menopause among women with a simple hysterectomy, although this would, if anything, have tended to underestimate the risks presented.

Previous studies have provided discrepant results regarding how the administration of progestins in EPT affects breast cancer risk, with some showing higher risks associated with continuous therapy (10, 14, 16, 21, 34, 45, 46, 54-56), another showing a higher risk associated with sequential therapy (15), and still others showing no major differences in the two regimens (9, 12, 13, 18, 29, 47). We found somewhat stronger associations for continuous than sequential EPT use, particularly for current and long-term use. Our finding emphasizes the importance of future investigations being able to evaluate high-risk use patterns to clarify differences according to hormone regimens.

As in previous investigations (12, 19, 24, 26, 27, 33-35), we found that combined therapy was strongly related to ER+ tumors but unrelated to ER- tumors. Our findings are consistent with recent U.S. preferential declines in the incidence of ER+ tumors (5, 6) following the 2002 WHI report and subsequent cessation of this regimen. We also found a somewhat stronger relationship with low-grade tumors, in line with several previous investigations (24-26).

Numerous studies have reported stronger hormone associations for invasive lobular than ductal tumors (9-13, 25, 27, 29-34). Few epidemiologic studies, however, have been able to assess associations according to newer diagnostic criteria, which incorporate classification of mixed ductal-lobular carcinomas as well as the more conventional categories of ductal and lobular tumors. Two studies that have assessed hormone risks for mixed tumors found EPT risks intermediate between those for pure ductal and lobular tumors (30, 31), but we, like one other study (27), found the highest hormone risks for mixed tumors, with little difference in hormone associations for ductal versus lobular tumors.

In comparing our results with those of others, the generally older age of our study subjects must be considered, resulting in the majority of tumors being ER+. This may explain some of the discrepancy in our results versus those of others. As noted elsewhere (25), the high degree of intercorrelation between clinical features complicates the interpretation of epidemiologic associations. For instance, our mixed histology tumors tended to be ER+, the group in whom hormone risks were highest. EPT use was unrelated to ER- ductal tumors but led to a 2.6-fold increased risk of ER+ ductal tumors. Our findings stress the complexity of breast cancer as a heterogeneous disease and support the need for additional development of statistical approaches for assessing effects of correlated tumor characteristics.

Some study limitations may have affected our findings. Updated hormone therapy data were not available after the second questionnaire. However, the short follow-up period should have minimized potential exposure misclassification, particularly because hormone use increased in the United States during this time, leading many participants to continue usage through 2002. Nonetheless, we could not evaluate whether cessation of or changes in use after baseline differed by exposure or breast cancer status. Further, reported duration of use at baseline would have systematically underestimated the true total duration of hormone therapy use in the population during the study period.

Additionally, although the baseline questionnaire was sent to a large and representative group of women ages >50 years, it generated a relatively low response rate. A greater proportion (∼60%) of its respondents completed the second questionnaire, which was the basis for deriving most of the hormone exposures considered. Nonetheless, compared with nonrespondents, respondents to the second questionnaire were more likely to be older, Caucasian, educated, in good health, thin, and current or longer-term hormone users, which may have limited the generalizability of our results. Finally, we lacked hormone receptor data for a substantial proportion of the cancers. Although this hindered our ability to assess relationships according to hormone receptor status, we do not believe that this biased our results, given that the absence of data was systematic rather than unsystematic (relating to whether cancer registries recorded such data). It was further reassuring that other risk factors did not differ substantially between state registries that recorded hormone receptor data and those that did not record such data.

In summary, our study provides unique data relevant to recent patterns of hormone usage in the United States. Of particular concern were our EPT results, which were consistent with an increase of ∼33 cases of breast cancer for every 10,000 users (compared with 4 additional cases for every 10,000 ET users). However, certain subgroups of users, notably thin women, may experience even higher risks, particularly for certain tumor subgroups. Thus, a true appreciation of hormone effects on breast cancer risk is dependent on considering modifying effects of body size and variation in risks according to combined breast cancer clinical features.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.