Menstrual Cycle Characteristics in Adolescence and Early Adulthood Are Associated With Risk of Early Natural Menopause Free

Abstract

Context

Early natural menopause (i.e., before age 45 years) is associated with increased risk of adverse outcomes. Associations of earlier menopause with younger age at menarche and short and/or regular cycle length are suggested, but study findings are inconsistent and few address early menopause risk.

Objective

To evaluate the relationship between menstrual cycle characteristics in early life with incident early natural menopause.

Design

The prospective Nurses’ Health Study 2 (1989 to 2011).

Setting and Participants

Women ages 25 to 42 years and premenopausal in 1989 (N = 108,811).

Main Outcome Measure(s)

Risk of early natural menopause not due to surgery, radiation, or chemotherapy (n = 2794) was evaluated with Cox proportional hazards models. Anti-Müllerian hormone (AMH) levels were considered in a nested case-control sample (n = 820).

Results

In adjusted models, risk was associated with earlier age at menarche (P for trend = 0.05), shorter (P for trend < 0.0001), and more-regular cycles (P for < 0.0001). The hazard ratio (HR) for women with age at menarche ≤9 (vs. 12) years was 1.28 (95% CI, 0.99 to 1.67). Women reporting usual menstrual cycle lengths <25 days at ages 18 to 22 years had substantially higher risk of early menopause (HR, 1.70; 95% CI, 1.47 to 1.96) than women with 26- to 31-day cycles, whereas women with ≥40 day cycles had lower risk (HR, 0.44; 95% CI, 0.34 to 0.58). Women with irregular cycle length had lower risk compared with women with regular cycles (HR, 0.51; 95% CI, 0.43 to 0.60). Associations with AMH concentrations among the nested case-control subset were consistent with these findings.

Conclusion

Results from this large prospective study of early menopause suggest an influence of accelerated oocyte depletion on risk and may help clarify the etiology of early menopause.

In Western populations, the average age at natural menopause is approximately 51 years; early menopause, cessation of ovarian function before age 45 years, affects ∼10% of women (1). Women who experience early menopause have increased risk of a number of adverse health outcomes, including cognitive decline, osteoporosis, and cardiovascular disease (2–5), the latter of which remains the leading cause of premature mortality in U.S. women as of 2015 (6). Reproductive potential declines during the 10 years leading up to natural menopause; for women who experience menopause before age 45 years, this may have substantial consequences for family planning, particularly as women increasingly delay childbearing into the later reproductive years (1–7).

Age at menopause is related to several potentially modifiable lifestyle and dietary factors that may affect ovarian reserve and/or rate of decline (8–24). A role of genes related to DNA repair and maintenance has been suggested in studies considering genetic influences (25). Menopausal timing also may be affected by polymorphisms in genes that may influence menstrual cycle function by disturbing regulation of the hypothalamic–pituitary–ovary axis, including those for gonadotropins, anti-Müllerian hormone (AMH), and receptors (26, 27). Epidemiologic studies of menopausal timing and/or early menopause risk have considered menstrual cycle characteristics that may be related to timing and rate of ovarian decline (8, 9, 11–15, 18–22, 24, 28–30). Short cycle length has been linked with earlier age at natural menopause (9, 13, 28); in contrast, findings regarding associations of menopausal timing or early menopause risk with menstrual cycle regularity (8, 11, 12, 21, 24) and age at menarche (8, 11, 12, 14, 15, 18–22, 24, 28–30) have been inconsistent. Studies considering early natural menopause have been limited by retrospective or cross-sectional study designs, small sample sizes, and/or small numbers of women with natural menopausal occurring prior to age 45 years.

To address these gaps, using data from the Nurses’ Health Study 2 (NHS2), we describe results of an evaluation of the association between menstrual cycle characteristics in adolescence and early adulthood and risk of early menopause. Because of the large cohort size and timing and length of follow-up, many incident cases of early menopause were observed, resulting in high statistical power and a unique opportunity to evaluate relationships with early menopause risk. In addition, we used biospecimens from a subset of the participants to evaluate associations of menstrual cycle characteristics with AMH, a biomarker of ovarian reserve, to supplement analyses of risk of early natural menopause.

Materials and Methods

The NHS2 is a longitudinal study of 116,429 female U.S. registered nurses who responded to a mailed baseline questionnaire in 1989. Participants were 25 to 42 years old at baseline in 1989 and provided information on lifestyle, medical history, and health-related behaviors; follow-up of the cohort is ongoing. Since baseline, biennial questionnaires have been used to collect updated information regarding lifestyle, behaviors, and new medical conditions and diagnoses of disease; dietary information is collected via questionnaires sent every 4 years. The cumulative follow-up rate over time has been at least 89%. For the current analysis, eligibility was limited to participants who reported being premenopausal and with no prior cancer diagnosis (except for nonmelanoma skin cancer), without hysterectomy or oophorectomy, and with available information on menstrual cycle characteristics in youth and adolescence in 1989 (n = 108,811). The study protocol was approved by the institutional review boards at Brigham and Women’s Hospital and the Harvard T.H. Chan School of Public Health, both in Boston, Massachusetts.

In addition to primary analyses among the full cohort, additional analyses were performed using a nested case-control study design (n = 820) selected from among the NHS2 Blood Cohort. Between 1996 and 1999, NHS2 members without cancer diagnosis were invited to provide self-collected blood samples; the Blood Cohort comprises 29,611 women, ∼79% of whom were premenopausal at the time of blood sampling. Timed samples were collected from women who were premenopausal; not using hormone therapy, oral contraceptives (OCs), or other hormones; and had not been pregnant within the past 6 months. These participants were asked to provide one sample in the follicular phase (days 3 to 5) and one in the luteal phase (7 to 9 days before anticipated menses). Untimed samples were collected from women using hormones or unwilling to collect timed samples. Upon receipt, samples were centrifuged, separated, and stored at ≤−130°C. Members of the Blood Cohort were similar to women in the main NHS2 cohort with regard to mean body mass index (BMI), parity, smoking status, OC use, and other factors (31). For the nested case-control study of incident early natural menopause, details of which are published elsewhere (32), case subjects (n = 328) were women who contributed a blood sample before natural menopause before age 45 years. Control subjects (n = 492) were premenopausal at blood sample provision and included 328 women who experienced menopause at age ≥47 years matched by age at the time of blood collection (±4 months), as well as by fasting status, time of day, season of blood collection, and sample type (luteal phase or random timing); and 164 women who experienced natural menopause at ages 45 and 46 years.

Outcome assessment

A component of the biennial questionnaires included questions regarding whether participants’ menstrual periods had ceased permanently. Among those indicating that their periods had ceased, participants were asked the age when their menstrual periods ceased and whether cessation was related to surgery, radiation or chemotherapy, or occurred naturally. Information regarding use of replacement sex hormones was collected as well. We identified cases of early menopause as those women who reported natural menopause before age 45 years during follow-up. Furthermore, because information regarding cessation of menstrual periods was provided on each questionnaire, we were able to consider transient periods without menses followed by return of menstruation and avoid error in self-report of timing of menopause and misclassification of early menopause status that could occur in cross-sectional reports or prospective reports without updating.

Menstrual cycle characteristics in adolescence and youth

Questions on the baseline NHS2 questionnaire in 1989 were used to determine menstrual cycle characteristics in early life. Participants were asked to report the age (in years) at which their menstrual periods began, with options from ≤9 through ≥17 years. The number of years from onset of menstrual period until cycles became regular was recorded as ≤1 year, 1 to 2 years, 3 to 4 years, ≥5 years, or never. Usual cycle length at ages 18 to 22 years was reported as ≤21 days, 21 to 25 days, 26 to 31 days, 32 to 39 days, 40 to 50 days, >50 days, or too irregular to estimate. Additional questions assessed cycle regularity during high school and during ages 18 to 22 years, with the following response options: very regular (±3 days), regular, usually irregular, always irregular, or, no periods. In addition, use of OCs for at least 2 months or for a full year was assessed from each year from ages 13 through 42 years.

Covariates

Questionnaires

Baseline questionnaires were used for collection of current age, height, weight, ethnicity, maternal and paternal education levels, and smoking status, among other factors. Throughout follow-up, information was collected to update weight, smoking, parity, OC use, breastfeeding, and hormone therapy use, among other factors. Current BMI (calculated as weight in kilograms divided by height in meters squared) was determined for each questionnaire cycle using baseline height and updated weight. The reported average number of cigarettes smoked per day was used to determine current smoking status and amount. Physical activity was assessed in 1991, 1997, 2001, 2005, and 2009 using questions regarding average time spent per week in specific activities (e.g., walking, running, biking), and these data were used to calculate metabolic equivalent task–hours per week (33). Semiquantitative food frequency questionnaires were used for dietary assessments starting in 1991 and in 4-year cycles thereafter. Participants were asked to estimate how frequently they consumed 131 foods, beverages, and supplements on average over the preceding year (34–36). These questionnaires have been previously assessed for validity (36). Nutrient intake was adjusted for total energy using the residual method (37).

AMH measurement

AMH was measured at Children’s Hospital, Boston, Massachusetts, by an ultrasensitive ELISA from ANSH Laboratories (picoAMH; Webster, TX). The assay uses the quantitative sandwich enzyme immunoassay technique. The day-to-day variabilities of the assay at concentrations of 0.023, 0.087, and 0.373 ng/mL are 5.8%, 3.2%, and 4.3%, respectively. The coefficient of variation from samples from a blinded plasma pool assayed alongside our analytic samples was 8.6%.

Statistical analysis

Baseline characteristics of the study sample were evaluated by categories of self-reported usual menstrual cycle length during ages 18 to 22 years. Age-adjusted comparisons of these characteristics between cycle-length groups were performed using generalized linear models. Cox proportional hazards models were used to estimate hazard ratios (HRs) for early menopause related to menstrual cycle characteristics among women in the NHS2 cohort who were premenopausal at baseline (n = 108,811). Specific characteristics assessed in separate models included: age (in years) at menarche, years from menarche to regularity, menstrual cycle regularity in high school and at ages 18 to 22 years, and usual menstrual cycle length at ages 18 to 22 years. For all comparisons, women reporting use of OCs for that time (i.e., high school or ages 18 to 22 years) were included separately from cycle-length categories. Accrual of follow-up (in months) began on the date of return of the 1989 questionnaire and continued until report of age 45 years, menopause, or the end of follow-up in June 2011, whichever occurred first. Menopause due to surgery (e.g., hysterectomy or oophorectomy), radiation, or chemotherapy was considered a censoring event, as were dropout and death.

Analyses were stratified on age and questionnaire cycle. Age-adjusted models were run and potential confounding addressed using multivariable models adjusting for a priori potential confounders observed to be related to cycle length. Because of concerns regarding the nature of the causal relation of later-life OC use with early-life menstrual cycle characteristics, time-varying OC use was evaluated in initial models, but because of its potential role as a causal intermediate, it was not included as a covariate in final models.

For all models, 95% CIs were estimated and P values provided for likelihood ratio tests of contribution to global model fit. P for trend was estimated by fitting models with exposure variables as ordinal categorical variables and reflects the linear association of category with risk. Sensitivity analyses were run to assess potential sources of bias and residual confounding, and included models that excluded women who reported autoimmune conditions (i.e., systemic lupus erythematosus, rheumatoid arthritis, and multiple sclerosis), excluded women reporting diagnosed polycystic ovarian syndrome (PCOS), were restricted to women with BMI of 18.5 to 25.0 kg/m², and were restricted to nonsmokers.

In addition, association of AMH levels with menstrual cycle characteristics were evaluated among members of the nested case-control study (n = 820). To address the skewness of AMH levels, general linear models were used to estimate and compare means of log-transformed AMH levels by cycle characteristics, which were back transformed to yield estimates of geometric means. Geometric means were estimated for models including case and control subjects combined and also restricted to women with age at natural menopause >45 years only (n = 492). For Cox models, P values were determined to assess global model fit from likelihood-ratio tests, and P for trend was assessed to evaluate linear relations of category with log AMH. All statistical analyses were conducted with SAS software, version 9.4 (SAS Institute Inc., Cary, NC).

Results

Baseline characteristics of women included in the current analyses were assessed and compared by category of cycle length at ages 18 to 22 years (Table 1). Small to moderate differences among groups were observed for age, BMI, alcohol consumption, parity, dietary factors, and race. Larger differences among the groups were observed for pack-years of smoking, physical activity and total breastfeeding duration. Compared with those with cycle lengths of 26 to 31 days, women with short cycles had more pack-years of smoking (8.5 vs. 6.8) and proportion of current smokers at baseline (13.1% vs. 10.7%), higher metabolic equivalent task -hours per week of physical activity (35.1 vs. 28.5), and shorter duration of breastfeeding (12.2 vs. 13.9 months).

Results of Cox proportional hazards models of menstrual cycle characteristics and risk of early menopause are shown in Table 2. Minimal differences were observed between results of age-only adjusted models and those of multivariable models, and so fully adjusted model results are described here. Risk of early menopause was associated with all considered characteristics [i.e., age at menarche (P = 0.002); years to regularity (P < 0.0001); cycle regularity in high school (P < 0.0001) and ages 18 to 22 years (P < 0.0001); and cycle length at ages 18 to 22 years (P < 0.0001)]. Cycle length was collapsed to four length options because of better parameter-penalized global model fit and comparable results to the seven-option version of the variable. Compared with those who were 12 years old at menarche, the HR for those with menarche at age 9 years or younger was 1.28 (95% CI, 0.99 to 1.67), and though CIs for age-specific HRs from multivariable models all crossed 1, estimates were generally consistent with an inverse association of age at menarche and risk of early menopause (P for trend = 0.05). Longer cycle length at ages 18 to 22 years was related to lower risk (P for trend < 0.0001). Compared with those reporting cycle lengths of 26 to 31 days, those with cycles <25 days had an HR of 1.70 (95% CI, 1.47 to 1.96), whereas risk was lower for those with cycles 32 to 39 days long (HR, 0.49; 95% CI, 0.40 to 0.59) and those with cycles of ≥40 days (HR, 0.44; 95% CI, 0.34 to 0.58).

Years from menarche to regularity, cycle pattern in high school, and cycle pattern at ages 18 to 22 years were all strongly related to risk of early menopause (P for trend < 0.0001 for all). Women who reported that their cycles never became regular had a HR of 0.61 (95% CI, 0.52 to 0.72) compared with those whose cycles became regular within a year; estimates for those with 3 to 4 years (HR, 0.80; 95% CI, 0.69 to 0.94) or ≥5 years to regularity (HR, 0.73; 95% CI, 0.64 to 0.83) were of intermediate magnitude. Compared with those reporting very regular cycles, risk was nearly 50% lower among those reporting always having irregular cycles or no menses in high school (HR, 0.55; 95% CI, 0.48 to 0.64) or at ages 18 to 22 years (HR, 0.51, 95% CI, 0.43 to 0.60). Risk for women reporting usually but not always irregular cycles had lower magnitude estimates. Estimates for OC users in each category included in initial models were largely comparable to the referent group and are shown in Supplemental Table 1.

Results of analyses jointly considering cycle length and regularity at ages 18 to 22 years are shown in Table 3. For these models, cycle length was further collapsed for some strata of regularity due to small numbers and low statistical power resulting from the high degree of correlation between length and regularity. In these models, similar findings regarding cycle length were observed among women reporting regular cycles, with higher risk observed for women with regular short cycles (HR, 1.64; 95% CI, 1.40 to 1.93) and lower risk for those with regular long cycles (HR, 0.61; 95% CI, 0.48 to 0.78) compared with those with regular cycles of 26 to 31 days. Lower risk was restricted to those with cycles longer than 32 days among women reporting usually irregular (HR, 0.50; 95% CI, 0.37 to 0.66) or always irregular cycles (HR, 0.50, 95% CI, 0.37 to 0.68). Results of sensitivity analyses (i.e., exclusion of women with diagnosed autoimmune conditions, exclusion of women with diagnosed PCOS, analysis restricted to nonsmokers, and analysis restricted to those with normal BMI) were unchanged from the primary analyses (data not shown).

Table 4 lists results of general linear models evaluating associations of AMH levels with menstrual cycle characteristics among the nested case-control sample, for which cases were matched to controls by age (±4 months), as previously described. Geometric mean AMH level varied significantly by age at menarche, and generally, but not uniformly, was higher with later menarcheal age. Low AMH values were observed for age at menarche of ≥17 years, though this age was reported by only three case and three control subjects. A similar pattern of higher AMH levels with later age at menarche was also observed among 491 women with age at natural menopause >45 years (P for trend = 0.04), but geometric means were generally higher at each age. Higher AMH levels were observed with longer times from menarche to regularity, longer cycle lengths at ages 18 to 22 years, and more-irregular patterns in high school and ages 18 to 22 years (P for trend < 0.001 for all). For example, among 28 women reporting having cycle lengths ≥40 days at ages 18 to 22 years, the geometric mean AMH level was 8.8 (95% CI, 5.1 to 15.2) ng/mL at the time of sample collection; by comparison, women reporting cycle lengths <25 days (n = 42) had a geometric mean AMH level of 3.8 (95% CI, 2.4 to 5.9) ng/mL. Women reporting always irregular cycles at ages 18 to 22 years (n = 29) had a geometric mean AMH level of 10.6 (95% CI, 6.3 to 18.1) ng/mL at the time of sample collection, whereas women with regular cycles at ages 18 to 22 years (n = 147) had a geometric mean AMH level of 4.8 (95% CI, 3.8 to 6.1) ng/mL, and women reporting very regular cycles at ages 18 to 22 years (n = 182) had a geometric mean AMH level of 3.6 (95% CI, 2.9 to 4.5) ng/mL. As with models of age at menarche, similar patterns of results were observed in analysis restricted to control subjects as to those for all 818 case and control subjects, but with higher overall levels of AMH.

Discussion

In this large study of menstrual cycle characteristics in adolescence and incident early natural menopause performed among 108,811 women in the NHS2 who were premenopausal at baseline, we observed a very strong relationship of shorter cycle length and more-regular cycles at ages 18 to 22 years with higher risk of early menopause. We also observed an association of increased risk with earlier ages at menarche, though this was of lower magnitude than associations with cycle length and regularity. Early-life menstrual cycle length and regularity may provide an early indication of later life outcomes. In addition, among a nested case-control study sample (n = 820), we observed a strong relationship between these menstrual cycle characteristics and levels of AMH in blood samples provided premenopausally, with lower geometric mean AMH level related to shorter and more-regular cycles and earlier menarcheal age. These associations were qualitatively conserved when analysis was restricted to control subjects (i.e., women who did not experience early natural menopause).

The findings from these analyses may help clarify results of prior studies considering menstrual cycle characteristics with regard to menopause. Younger age at natural menopause is related to shorter cycle length at ages 20 to 35 years (9, 30), 25 to 35 years (18), or during “midlife” (28), even when using broad categories of cycle length. Of prior studies considering cycle regularity, general lifetime cycle irregularity (8, 11, 21) has been associated with older age at natural menopause compared with regular cycles; whereas no association was observed between age at natural menopause and usual cycle irregularity (12) or risk of early menopause and irregularity at ages 18 to 22 years (24). Findings regarding age at menarche have been more discordant, with many having observed earlier menopause with earlier onset of menses (11, 12, 15, 21, 23, 28–30), but even some very large studies observed no association (8, 14, 18–20, 24). In terms of relative contributions to ovarian decline, short vs. long cycle length may have a larger potential impact on the total number of ovulatory cycles by a given age than does age at menarche; small persistent differences in cycle length can correspond to substantial differences in the number of menstrual cycles occurring over a given time. In our study, we observed an approximately twofold higher risk for women reporting cycles of 26 to 31 days at ages 18 to 22 years compared with those with cycles of 32 to 39 days or ≥40 days. Associations of similar magnitude were observed for cycle regularity. We observed a statistically significant linear inverse relationship between menarcheal age and early natural menopause odds, but smaller-magnitude associations were observed, and individual age at menarche-specific estimates were not statistically significant in adjusted models.

AMH is a glycoprotein produced by small antral and preantral follicles (38) and is commonly used as a measure of ovarian reserve. Though recent studies have shown that it is unrelated to fecundity in a general population (39, 40), it has consistently been related to menopausal timing (41–44). In our own data, we have recently extended these findings, showing that AMH level strongly predicts risk of early menopause, as well (32). In our analyses evaluating AMH levels among those in the nested case-control subset, characteristics related to higher risk of early menopause were also observed to be related to lower AMH level. Notably, we observed these associations among all case and control subjects, as well as among control subjects only whose age at menopause ranged from 45 to 58 years. These observations may suggest general relationships of timing of cycle onset, cycle length, and cycle regularity with age at menopause, rather than specifically with risk of early menopause.

The NHS2 is a prospective study with 20 years of follow-up. Many prior analyses have been cross-sectional studies and/or used retrospectively determined age at menopause. In contrast, we used detailed longitudinal reports to classify age at menopause and cases of early menopause. In addition, the large sample size and extensive covariate assessment in the NHS2 help address possible confounding and/or effect modification. We performed a number of sensitivity analyses to address possible biases related to autoimmune conditions, as noted in Methods; PCOS, and potential residual confounding due to misclassification of smoking and BMI. In all cases, results were consistent with those of the full cohort.

Nevertheless, several limitations should be considered. We used self-report for age at menopause and all menstrual cycle characteristics. A study of 6591 women in the comparable NHS population suggested self-reported menopausal status to be reproducible and valid; among women who were premenopausal in 1976 and reported having natural menopause on the 1978 questionnaire, 82% reported their age at menopause to within 1 year on the following two questionnaires (45). Similarly, studies of retrospective self-report of age at menarche suggest it to be valid and to have high concordance to prospectively assessed menarcheal timing (46–49). Still, reliance of self-reported recall of menstrual cycle characteristics from as many as ≥20 years in the past increases the probability of measurement errors. Using these baseline measures to compare women by prospectively determined early menopause status is most likely to result in nondifferential misclassification; however, to the extent that reporting patterns of retrospective menstrual cycle characteristics in early adulthood and menopausal timing could be related, bias otherwise cannot be ruled out.

In addition, the possible role of PCOS adds some complication to these findings, despite sensitivity analysis results. Related to associated genetic variants and its etiology, PCOS has clear and well-established associations with menstrual cycle characteristics (50, 51), and thus at least some of the observed risk may be related to this condition. Assessment of menstrual cycle characteristics from adolescence may help reduce the impact of PCOS on findings, though research suggests an early age of onset for some women (52). The observation of linear trends in risk and the large magnitude of estimates in conjunction with sensitivity analyses help strengthen our findings; however, careful attention to PCOS will be an important consideration for future studies. Complexities are also introduced by the role of OCs. OC use may be initiated because of menstrual cycle irregularities, as well as other complications; therefore, for these analyses, we grouped OC users with those missing information. The role of OCs in development of early menopause has received considerable attention and findings are ambiguous (11, 12, 18, 20, 21, 24, 28, 53). In our initial models, estimates for OC users were generally comparable to those in the reference category (i.e., those with regular cycles and with lengths of 26 to 31 days); however, these models addressed only use vs. nonuse, and despite also considering time-varying use as a covariate, because of the potential issues described and a desire to address OC use more comprehensively, these results were not included here. We plan to address this important topic more comprehensively in studies using these data.

In summary, in this study following 108,811 members of the NHS2 for incident early menopause, we observed risk to be strongly related to menstrual cycle length, regularity, and age at menarche. Taken together, these results are consistent with the rate of ovulation as a unifying mechanism for the associations observed. We adjusted for a wide range of diet, lifestyle, and behavioral factors that may affect menstrual characteristics; however, determinants of cycle length and regularity and the etiology of these associations merit additional consideration.

Abbreviations:

Abbreviations:
 
• AMH

  anti-Müllerian hormone

 
• BMI

  body mass index

 
• HR

  hazard ratio

 
• NHS2

  Nurses’ Health Study 2

 
• OC

  oral contraceptive

 
• PCOS

  polycystic ovarian syndrome

Acknowledgments

Financial Support: This work was supported by the National Institutes of Health Grants R01HD078517 (to E.R.B.-J.), R01CA67262 (to S.E.H.), and UM1CA176726 (to S.E.H.).

Disclosure Summary: The authors have nothing to disclose.

REFERENCES