Reproductive factors and postmenopausal hormone use in relation to endometrial cancer risk in the Nurses' Health Study cohort 1976–2004

Authors

  • Stalo Karageorgi,

    Corresponding author
    1. Department of Environmental Health, Harvard School of Public Health, Boston, MA
    2. Program in Molecular and Genetic Epidemiology, Harvard School of Public Health, Boston, MA
    • Harvard School of Public Health, 677 Huntington Avenue, Boston, MA 02115, USA
    Search for more papers by this author
    • Fax: +617-432-1722

  • Susan E. Hankinson,

    1. Channing Laboratory, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA
    2. Department of Epidemiology, Harvard School of Public Health, Boston, MA
    Search for more papers by this author
  • Peter Kraft,

    1. Program in Molecular and Genetic Epidemiology, Harvard School of Public Health, Boston, MA
    2. Department of Epidemiology, Harvard School of Public Health, Boston, MA
    Search for more papers by this author
  • Immaculata De Vivo

    1. Program in Molecular and Genetic Epidemiology, Harvard School of Public Health, Boston, MA
    2. Channing Laboratory, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA
    3. Department of Epidemiology, Harvard School of Public Health, Boston, MA
    Search for more papers by this author

Abstract

Endometrial cancer is a disease primarily driven by cumulative exposure to estrogen unopposed by progesterone. Reproductive factors associated with changes in endogenous hormone levels and use of exogenous hormones such as postmenopausal hormones influence the risk of disease. The authors used the Nurses' Health Study, comprised of 121,700 nurses, to assess the above associations. Over 28 years of follow-up, 778 adenocarcinoma cases were diagnosed and 1,850,078 person-years were accumulated. Cox proportional hazards models were used to estimate relative risks (RR) and 95% confidence intervals (CI). A late age at menarche decreased the risk independent of body mass index (BMI) (P-trend = 0.02). A late age at menopause increased cancer risk (P-trend = 0.0003). An advanced age at last birth reduced the risk (P-trend < 0.0001), however, an inverse association with age at first birth and parity diminished after adjustment for age at last birth. Compared with never users, an increased risk was observed among long-term (≥5 years) users of both estrogen (E) (RR = 7.67, 95% CI: 5.57, 10.57) and combined estrogen plus progesterone (E+P) (RR = 1.52, 95% CI: 1.03, 2.23). Normal-weight (BMI < 25) women had the highest risk following E or E+P use (P-interaction-E = 0.0008, P-interaction-E+P = 0.02). The findings from this study underscore the importance of hormonal mechanisms in endometrial carcinogenesis.

Endometrial cancer is the most common gynecologic malignancy and the 4th most common type of cancer in women. Around 40,000 cases are diagnosed annually in the United States at an average age of 62 years.1 The key components in endometrial cancer, a hormone-associated cancer, include estrogen and progesterone. The accepted etiological hypothesis, known as the unopposed estrogen hypothesis, explains that exposure of the endometrium to estrogen without concurrent exposure to progesterone stimulates endometrial cell proliferation, increasing the likelihood of genetic errors and malignant transformation.2, 3 Progesterone opposes the estrogen effect by down-regulating estrogen receptors and promoting endometrial cell differentiation.3 An imbalance in these 2 hormones may be caused by excessive estrogen or a deficiency in progesterone.

Reproductive factors are likely risk factors for endometrial cancer because they are linked to endogenous hormone levels. Menarche, menopause and pregnancy are accompanied by major changes in endogenous hormone levels. Since 1988, at least 5 prospective cohort studies4–10 and at least 9 case-control studies (nested case-control,11 population-based12–15 and hospital-based16–20) reported on reproductive factors and endometrial cancer risk. Most of the studies agree that early age at menarche and a late age at menopause increase risk.7, 8, 10, 13, 14, 20 However, among studies that describe an inverse association with age at menarche not all controlled for body mass index (BMI),7, 8, 10 which is a strong risk factor for endometrial cancer and a determinant of age at menarche. Nulliparity has been consistently associated with an increased risk, and a decrease in risk with increasing number of births has been observed in almost all studies.4, 5, 7, 8, 11–13, 15, 16, 18 The role of timing of birth is less clear. Results are largely inconsistent on age at first birth as a possible risk factor, and recent studies report on a decreased risk associated with late age at last birth.6, 7, 11, 16, 17, 19 To our knowledge, no studies have presented results examining the number of births controlling for age at last birth.

Studies on exogenous hormone use, specifically postmenopausal hormone use (PMH), provide more direct evidence of support for the unopposed estrogen hypothesis and highlight the importance of duration of exposure. Estrogen hormone use was initially prescribed to help ameliorate menopausal symptoms and then increasingly prescribed for long-term use to prevent conditions such as osteoporosis and heart disease. Following a rise in sales in the 1970s and a rise in endometrial cancer incidence in the 1980s,21 subsequent studies confirmed a 2- to 10-fold increased risk depending on the duration of use.22 Progestins were added to estrogen use (combined hormone use) in a sequential (only for a number of days/month in addition to daily use of estrogen) or continuous manner (daily use of both estrogen and progesterone) to counteract the effect of estrogen. The addition of progesterone to estrogen lowered the risk compared to estrogen only therapy,23–35 however, the risk associated with the different types of combined hormone regimen remains to be elucidated. Studies that examined sequential use for <10 days/month point to an increased risk.24, 29, 31 Results on sequential hormone use for ≥10 days/month and daily use are inconclusive varying among a positive, null or inverse association.24–26, 29–31, 34, 35

We used the Nurses' Health Study (NHS), a prospective cohort study with information collected every 2 years on variables of interest. We examined the association among age at menarche, age at natural menopause, parity, age at first birth, age at last birth and PMH use by type and duration and endometrial cancer risk over a follow-up period of 28 years.

Material and Methods

Study population

The NHS is a prospective cohort study established in 1976, when 121,700 married registered nurses residing in 11 U.S. states and between the ages of 30 and 55 years completed a baseline mailed questionnaire inquiring about various disease exposures and personal health status. Revised questionnaires were mailed biennially to update exposure and disease information. The follow-up rate through 2002, as measured in percent of total possible person-years, was 93.7%. Deaths in this cohort were identified by next-of-kin reports, the U.S. Postal Service, or through searches of the National Death Index.

In this analysis, nurses who reported a hysterectomy, cancer other than nonmelanoma skin cancer, and surgical menopause were excluded at baseline and at each subsequent follow-up cycle. This resulted in 95,638 eligible subjects at baseline followed from the age in months at the date of return of the 1976 questionnaire until the end of the study, June 1, 2004. Each subject contributed person-time until the age in months at death, report of endometrial cancer, loss to follow-up, or the end of the study, whichever came first. A total of 1,850,078 person-years were accumulated over 28 years of follow-up. Postmenopausal women contributed 990,397 person-years.

Endometrial cancer cases

Information on endometrial cancer was collected beginning in 1978 and at each subsequent questionnaire cycle. Women who reported being diagnosed with endometrial cancer were asked for permission to review their medical records. Only cases of invasive Type I endometrioid adenocarcinoma (ICD-O-3 histology codes 8380-83) confirmed by medical records were included in the analysis (88% of total cases). Over 28 years of follow-up, 778 cases of invasive adenocarcinoma were identified; of which 658 were postmenopausal.

Exposure assessment

Information on age at menarche was collected at baseline and women were asked to report the age at which their menstrual periods began. Information on age at menopause was collected at baseline and every 2 years thereafter. For each questionnaire women were asked whether their menstrual periods have ceased permanently, at what age and for what reason (natural or surgical). Only nurses with natural menopause were considered in this study, and the first consistent report of age at menopause was accepted for the nurses. Age at menopause in the NHS is reported with a high degree of reproducibility and accuracy.36

Women were asked to report if they had a pregnancy lasting 6 months or more and the year of birth in 1976 and in cycles following up to 1984. In 1996 they were asked to report their lifetime pregnancy (≥6 months) history. Information from above questionnaires was used to derive variables for parity, age at first birth and age at last birth that were updated every 2 years.

PMH use and duration of use in months was first asked in 1976. Beginning in 1978 and at each 2-year follow-up, women were asked whether they currently used PMH, number of months used during the 24 months prior to the questionnaire, and the type of PMH. In 1978 the majority of users (79%) of a known type reported using estrogen; therefore, type of PMH for users in 1976 was classified as estrogen. Type of hormone was classified into oral conjugated estrogen (E), combination estrogen plus progesterone (E+P) and other hormones. Results for other hormones are not presented because this represents a heterogeneous group of hormones (nonconjugated estrogens, patch hormones and vaginal hormones). Duration of ever hormone use was calculated as the cumulative duration of all types of PMH use reported over follow-up. Duration of a specific type was calculated as the cumulative duration over all reported time of use of that specific type. Between 1988 and 1990 women were asked whether they used progesterone continuously, approximately 1–2 weeks, or had not used. From 1992 forward, the number of days/month of progesterone use was asked as follows: not used, <1 day, 1–8 days, 9–18 days, 19–26 days and 27+ days.

Covariate assessment

Information on potential confounders such as BMI, diabetes and smoking was collected at baseline and updated every 2 years. Updated BMI (kg/m2) was calculated using height reported in 1976 and weight reported at each cycle. Height and weight are accurately reported in the NHS.37 Because BMI is an important confounder, women missing weight in 1 cycle had their weight carried forward once from the prior cycle. Women missing weight for 2 consecutive cycles were excluded until they reported their weight again. Nurses were asked whether they were past or current cigarette smokers and the number of cigarettes smoked per day. Pack-years were calculated by multiplying smoking duration in years by packs of cigarettes smoked per day (20 cigarettes per pack).

First-degree family history of endometrial cancer was collected only in 1996. Information on oral contraceptive (OC) use and duration of use in months was collected up until 1982, at which time fewer than 500 women reported using OCs.

Statistical analysis

Multivariate Cox proportional hazards models stratified by age in months at the start of follow up and calendar year, were used to estimate relative risks (RRs) and 95% confidence intervals (CIs). The time scale used was follow-up time in months which is equivalent to age in months. Exposures were mutually adjusted and covariates included in the model were selected based on previous findings. Important predictors of the outcome that were not confounders were included to increase power of the analysis. The final model included the following variables: age at menarche, age at menopause, parity, age at first birth, age at last birth, PMH duration, OC duration, BMI, smoking pack-years, report of diabetes and family history of endometrial cancer. To examine PMH use in detail, PMH duration in the above model was replaced with ever use, type and type by duration.

All exposures and covariates were categorized and an indicator variable was created for each category except BMI, which was used continuously. The categories chosen are listed in the tables. For covariates missing data, a missing category was created. All exposures and covariates were updated biennially at the date in months of return of each questionnaire, except age at menarche and family history of endometrial cancer, which were used as baseline variables. For exposures, the most recent or current report was used to estimate risk. Analysis was restricted to postmenopausal women to examine age at menopause and PMH use, and parous women to examine parity, age at first birth and last birth. In this case, women started contributing person-time after they became menopausal or parous. Exclusive use of E+P hormones was examined by excluding women who reported using E at baseline and at each follow-up cycle.

Tests for trend were performed by entering the exposure of interest as a continuous linear term in the model. Only interactions based on biological plausibility or previously reported7, 26, 31, 34, 35, 38 were considered. We investigated effect modification of exposures by BMI, smoking and PMH use/type. Tests of interaction were assessed using the Wald test for interaction.

Results

The mean age among all participants at the start of follow-up in 1976 was 41.8 years and the average time of follow-up was 19.3 years. The mean age at diagnosis for all cases was 62.2 years. The distributions of covariates among all women at baseline are outlined in Table 1. Approximately 15% of women were postmenopausal at baseline. Fifty percent of women had never used OCs and 40% were nonsmokers. Seventy percent of the women were normal-weight at baseline and 3% reported a family history of endometrial cancer in 1996.

Table 1. Distribution of baseline characteristics among all women in the Nurses Health Study in 19761,2
inline image

Age at menarche was inversely associated with endometrial cancer risk and the association was stronger among postmenopausal women (P-trend = 0.02) (Table 2). Women who started menstruating at age ≥15 years were at a 34% decreased risk compared with women who started menstruating at age ≤11 years (95% CI: 0.46, 0.94). The difference between age-adjusted and multivariate results for age at menarche was because of confounding by BMI. Age at menopause was positively associated with endometrial cancer risk (P-trend = 0.0003) (Table 2). Women who reached menopause at an age ≥ 55 years were at a 53% increased risk compared with women who reached menopause between the ages of 45 and 49 years (95% CI: 1.13, 2.06).

Table 2. Relative risks for age at menarche and age at menopause in relation to endometrial cancer in the Nurses' Health Study, United States, 1976–2004
inline image

Nulliparous women were at a 24% increased risk compared with uniparous women (95% CI: 0.89, 1.74) (Table 3). Among parous women, in the multivariate model not adjusted for age at first and age at last birth, we found a decrease in risk with increasing number of births (P-trend = 0.0001) (Table 3). After additional adjustment for age at last birth, the trend was attenuated (P-trend = 0.04) and the RRs were no longer statistically significant. The RRs strengthened slightly after additional adjustment for age at first birth (Table 3).

Table 3. Relative risks for reproductive factors in relation to endometrial cancer among parous women in the Nurses' Health Study, United States, 1976–2004
inline image

Age at last birth was inversely associated with endometrial cancer risk (P-trend <0.0001) (Table 3). Results were robust and did not change after adjustment for parity or age at first birth. Women who had their last child at an age ≥40 years were at a 49% decreased risk compared with women who had their last child between the ages of 25 and 29 years (95% CI: 0.34, 0.77). The inverse association between age at last birth and endometrial cancer risk was similar across strata of parity (data not shown). Restricting the analysis to women with ≥2 births did not significantly change the results for age at last birth or age at first birth (data not shown). A significant inverse association emerged between age at first birth and endometrial cancer risk after adjustment for parity, but diminished after additional adjustment for age at last birth (Table 3). In this population, parity and age at last birth, as well as age at last birth and age at first birth were moderately positively correlated (r-Pearson, parous only = 0.51 and 0.47, respectively). Age at first birth and parity were weakly negatively correlated (r-Pearson, parous only = −0.26). Results for parity, age at first birth, and age at last birth did not change after the analysis was restricted to postmenopausal women (data not shown).

We examined use of PMHs in relation to endometrial cancer risk in the NHS (Table 4). The difference observed between age-adjusted and multivariate results throughout all PMH analyses was caused by confounding by BMI. Compared with never PMH users, ever use of E was associated with a 2.78 increase in risk (95% CI: 2.17, 3.57). Among ever E users, short-term (<5 years) users were at a 52% increased risk (95% CI: 1.08, 2.13) whereas long-term (≥5 years) users had a 7.67 increase in risk (95% CI: 5.57, 10.57), 5 times the risk of short-term users (P-trend, E-duration = <0.0001, data not shown). Among long-term (≥5 years) E users, current users were at a 10.78 increased risk (95% CI: 7.53, 15.44) and past users had a 4.17 increase in risk (95% CI: 2.44, 7.12). Past long-term users had a substantial decrease in risk after 2 years of nonuse (RR = 2.23, 95% CI: 1.02, 4.84) (data not shown).

Table 4. Relative risks for postmenopausal hormone use in relation to endometrial cancer among postmenopausal women in the Nurses' Health Study 1976–2004
inline image

Among all women, ever E+P use was associated with a 59% increase in risk (95% CI: 1.26, 2.01) compared with never users (Table 4). After excluding E users from the analysis, ever E+P use was associated with a 33% increased risk (95% CI: 1.01, 1.75). Excluding both E users and women missing PMH did not alter the findings for E+P (data not shown). Among the ever E+P users, long-term E+P users (≥5 years) were at a higher risk of 52% increase in risk (95% CI: 1.03, 2.23) (P-trend, E+P-duration = 0.005, data not shown). We also examined the change in risk with the addition of progesterone (days/month) to estrogen and observed a significant increased risk for 1–8 days/month and a nonsignificant increased risk for 9–18 days/month and 27+ days/month (Table 4) (P-trend = 0.03, data not shown).

The risk associated with E use varied by BMI (P-interaction = 0.0008) (Table 5). Compared with never users within BMI strata, normal weight (BMI < 25) users were at a considerably higher risk (RR = 4.03, 95% CI: 2.42, 6.72) than overweight (BMI 25–29) (RR = 1.96, 95% CI: 1.18, 3.25) than obese (BMI ≥ 30) users (RR = 1.06, 95% CI: 0.56, 1.98). Results were similar for E+P use (P-interaction = 0.02) (Table 5). No other significant interactions were observed.

Table 5. Relative risks for type of postmenopausal hormone use in relation to endometrial cancer by BMI among postmenopausal women in the Nurses' Health Study 1976–2004
inline image

Discussion

This is the first report on reproductive factors and PMH use in relation to endometrial cancer risk in the NHS. Our findings on age at menarche, age at natural menopause and PMH use are consistent with previous reports. We further provide insight into the independent effects of age at first birth, age at last birth and parity. The strengths of this study include the use of large prospectively collected data with frequently updated information on exposures, covariates and hysterectomy status of participants followed for a long period of 28 years.

We observed that early age at menarche and late age at menopause increase endometrial cancer risk.7, 8, 10, 13, 14, 20 Not all previous studies that reported an association with age at menarche adjusted for BMI in their analysis.7, 8, 10 In our data, following adjustment for BMI the association was attenuated however it remained significant, suggesting that the relationship is not wholly explained by BMI. One of the most pronounced hormonal changes observed with the onset of menstrual periods is a rise in estradiol levels and then the fall in estradiol levels with the cessation of menstrual periods.39, 40 The increased risk associated with early menarche and/or late menopause has therefore been attributed to a longer lifetime exposure to endogenous estrogen.13, 41 Alternatively, progesterone deficiency associated with anovulatory cycles that are more common during early and late reproductive life may also contribute to endometrial cancer risk.14, 42 It is also possible that early vs. late onset of menstruation is linked to different hormonal profiles in adult life, such as high levels of estradiol43 or low levels of sex-hormone binding globulin.44, 45

We observed that late age at last birth (≥40 years) reduces the risk of endometrial cancer whereas an inverse association for age at first birth diminished after adjustment for age at last birth. Our findings are in concordance with previous reports that observed an inverse association with age at last birth,6, 7, 11, 16, 17, 19 and no association with age at first birth4, 8, 13, 14, 17–19 or an association that diminished after controlling for age at last birth.7, 9, 11 Interestingly, the trend we observed in decreasing risk with increasing parity also disappeared after adjustment for age at last birth, indicating that the risk reduction observed for high parity is better explained by a late age at last birth. With the exception of a few studies,10, 14, 17, 46 previous reports showed a decreasing risk with increasing number of births,4, 5, 7, 8, 11–13, 15, 16, 18 however both studies did not control for age at last birth.

The largest reduction in risk was observed for women with a last pregnancy in the 4th decade. This time coincides with perimenopause, a 10 year period preceding menopause, during which estrogen and progesterone levels decrease.39 Studies show that following a pregnancy, parous women have lower estrogen levels than nulliparous women,44, 47–49 therefore an advanced age pregnancy may protect against cancer by further decreasing estrogen levels in women approaching menopause. Alternatively, exposure to high levels of progesterone throughout pregnancy may be especially effective at a time when progesterone deficiency is common at later ages. Additionally, prolonged exposure to progesterone may facilitate the removal of premalignant lesions, as observed after treatment of women with hyperplasias with progestins in the PEPI trial.20 Premalignant cells may also be mechanically removed by exfoliation during parturition,5 however, such a mechanism would be restricted to women having a vaginal birth as opposed to a C-section. Alternatively, mechanical elimination of premalignant lesions may happen during uterine involution.50 There is also a suggestion of an immunity-mediated effect of pregnancy; sera from parous women commonly recognized endometrial tumor antigens which failed to be recognized by sera from nulliparous women.51 If the immune system is involved, then a pregnancy at a late age may act as a booster vaccine against endometrial cancer. Alternatively, it is possible that women capable of having children at a late age may possess a healthy uterus or experience fewer anovulatory cycles.11, 19

A strong duration-risk relationship was observed for menopausal estrogen therapy. Consistent with a meta-analysis of 37 studies from 1970–1994 by Grady et al. in 1995,22 and subsequent studies,25, 29, 31, 34 we observed an 8-fold increase in risk for long-term (≥5 years) E use. Current users were at a higher risk that past users indicating that the risk associated with E use decreases following cessation of use, however, an elevated risk may remain even after ≥5 years22 or ≥10 years since last use.24, 33 Our limited data suggests that the risk of E only decreased substantially after 2 years since cessation of use among long-term users (≥5 years).

Since Grady et al.,22 at least 6 case-control studies23–31 and 4 cohort studies32–35 have been published on E+P. We had the advantage of having information on previous use of E among E+P users and excluded these users from the analysis eliminating residual confounding by previous E use. Consistent with the unopposed estrogen hypothesis and the studies above we observed a substantially lower risk for long-term (≥5 years) E+P than E use. The number of days of progesterone administration within a month is an important factor for characterizing E+P risk. Despite the low power, we observed an increased risk with the addition of progesterone for <10 days/month as previously reported.24, 29, 31 Studies in which progesterone was added for ≥10 days/month observed an increased25, 31 or null risk.24, 26, 29, 34, 35 Daily use was associated with an increased,26 null24 or decreased risk.25, 30, 31, 35 In this study, use for 9–18 days/month and daily use were associated with a nonsignificant increased risk. The inconsistencies between studies may partly be explained by the small number of users in some studies or the use of different ranges of days/month progesterone. The only 2 available clinical trials, the PEPI trial in 1996 and the WHI trial in 2002, reported a null association with daily use (<5 years) and endometrial hyperplasia or endometrial cancer.20, 52

Normal-weight women were at the highest risk of cancer rather than overweight than obese women following use of E or E+P, compared with never users.26, 31, 34, 35 Obese postmenopausal women have higher levels of circulating estrogen53 therefore this finding suggests that an estrogen threshold level may exist beyond which any additional estrogen does not have an influence.3 Also possible is adaptation; obese women are acclimated to high levels of estrogen. In this study, obese women (18%) were less likely to use E or E+P than overweight (31%) than normal-weight (55%) women. The lower risk observed in the stratum of obese women is not attributed to residual confounding by BMI given that we adjusted for BMI within strata. Consistent with the Million Women Study,35 among obese women, E+P use was associated with a decreased risk compared with never use.

In summary, the findings from our study strongly support the role of reproductive factors and PMH use on endometrial cancer risk. The associations are restricted to Caucasian women with adenocarcinoma of the uterus. Age at menarche was inversely associated with cancer risk independent of BMI; and age at natural menopause was positively associated. A late age at last birth rather than age at first birth or parity was a more important reproductive factor for reducing the risk. Normal-weight (BMI < 25) women were particularly susceptible to the adverse effects associated with both E and E+P hormones. Overall, the findings from this study underscore the importance of hormonal mechanisms in endometrial carcinogenesis.

Acknowledgements

Ms. Stalo Karageorgi was supported by the HSPH-Cyprus Initiative for the Environment and Public Health funded by the Republic of Cyprus. The authors thank Dr. Monica McGrath and Prof. David Hunter for their contributions to this study.

Ancillary