Depression and Its Influence on Reproductive Endocrine and Menstrual Cycle Markers Associated With Perimenopause:  The Harvard Study of Moods and Cycles FREE

IN A RECENT study of premenopausal women, Young et al¹ reported lower follicular-phase plasma estrogen levels and higher luteinizing hormone (LH) levels in women with depression than in control subjects. Since decreased estrogen influences a range of neuroendocrine systems related to mood, behavior, and cognition,² it is conceivable that women with a history of depression may have a long-standing compromised hypothalamic-pituitary-ovarian axis. An early age at menarche is associated with developing depression later in life³ and also an earlier onset of menopause.⁴ Depression may therefore have either an intermediary or an independent effect on age at natural menopause.

Recent longitudinal population-based studies suggest that a normal transition to menopause does not appear to increase a woman's risk of depression.^5- 8 However, it has been shown that the perimenopausal transition is a time of increasing vulnerability for depressive episodes, particularly among women with a history of mood disorder.^9- 10 While investigators have described the risk of depression among perimenopausal women, few studies have assessed the impact of a history of major depression on an early transition to menopause. One population-based case-control study reported that those who underwent a natural menopause before age 47 years were 2 to 3 times more likely to have reported a history of depression that required medication treatment for 1 year or longer compared with those who made a normal transition to menopause.¹¹ On the basis of these findings, we speculated (1) whether depression or its treatment could be a preceding risk factor for a precipitous decline in ovarian function and (2) whether depression itself could be a marker for a precipitous decline in ovarian function that precedes the cessation of menstruation by several years. By observing a cohort of premenopausal women with and without major depression, we sought to determine whether a lifetime history of major depression was associated with an earlier decline in ovarian function based on changes in menstrual cycle characteristics and early follicular-phase gonadotropin measures.

The Harvard Study of Moods and Cycles was derived from a population-based cross-sectional sample of approximately 6000 women between the ages of 36 and 44 years selected from 7 Boston, Mass, metropolitan area communities. After a 73% response rate was achieved to a self-administered questionnaire that assessed menopausal status and past and current depressive symptoms, 4161 women composed the target population used to identify a sample with and without a lifetime history of major depression. Further details pertaining to the characteristics of this framed target population can be found elsewhere.¹² We enrolled in the Harvard Study of Moods and Cycles 332 women who met DSM-IV criteria for major depression and 644 women with no current or previous history of major depression on the basis of in-person administration of the Structured Clinical Interview for DSM-IV.¹³ Depression severity was measured by means of Hamilton Rating Scale for Depression interviews,¹⁴ but few women (n = 16) were experiencing an acute depressive episode (scores >17) at study enrollment. In an earlier report, our group presented (1) the sampling and selection criteria used to identify and classify women with and without a lifetime history of major depression,(2) the detailed data collection methods, (3) comparisons between women who were and were not enrolled within depressed and nondepressed target populations, and (4) justification that the quality of the data obtained from our sample of women ensured a high degree of internal validity.¹⁵

Our main outcome of interest was inception of perimenopause as a marker of early ovarian decline. The World Health Organization defines perimenopausal transition as the time immediately preceding menopause, beginning with endocrine, biological, and clinical changes and ending a year after the final menstrual period.¹⁶ During this phase, the rate of depletion of ovarian follicles increases sharply and the aging ovaries lose their ability to undergo ovulation and produce estrogen and progesterone. Often, the first sign of perimenopause is menstrual irregularity, resulting from erratic ovarian hormone secretion and decreased frequency of ovulation. In the early phase of perimenopause, women may experience a shortening of their menstrual cycles by 2 to 7 days, an increased quantity of menstrual flow, and midcycle spotting.¹⁷ Later in perimenopause, skipped periods or longer cycles are common, as fewer follicles remain and anovulation occurs. Up to 90% of women may experience the onset of changes in menstrual patterns, typically occurring between the ages of 40 and 44 years.¹⁷ Median age at inception of perimenopause has been estimated at 47 to 48 years.^18- 19

Informed by the Massachusetts Women's Health Study,^20- 22 the Seattle Midlife Women's Health Study,¹⁷ and related epidemiologic research,^18,23 we defined time of entry into perimenopause as the time, in months, from the baseline interview to the follow-up interview in which a woman reported the occurrence of any of the following events within the previous 6 months: an absolute change of 7 days or more in menstrual cycle length as compared with baseline; a change in menstrual flow amount (≥2 flow categories, eg, from light or moderately light to moderately heavy or heavy) or duration (absolute change of ≥2 days); or periods of amenorrhea lasting at least 3 months. Since women entered the study at different ages (36-45 years at baseline), our main analyses were limited to "time to perimenopause."

In-person interviews were conducted with all study participants regarding their demographic and lifestyle characteristics, menstrual and reproductive history, past and current medical conditions, and use of hormonal and nonhormonal medications at study enrollment. Telephone interviews were conducted every 6 months during a 36-month follow-up to track menstrual and medical history changes.

All participants provided early follicular-phase (day 2, 3, or 4 of the menstrual cycle) blood specimens at study enrollment and then every 6 months during a 36-month follow-up. The blood specimens were used for the measurement of serum gonadotropins via radioimmunoassays (Coat-A-Count; Diagnostic Products Corporation, Los Angeles, Calif). The World Health Organization First International Reference Preparation (68/40) of human pituitary LH was used as the standard for LH, and the World Health Organization Second International Reference Preparation (78/549) of human pituitary follicle-stimulating hormone (FSH) was used as the standard for FSH. Estradiol was measured in duplicate and averaged by means of the radioimmunoassay (Coat-A-Count). Intra-assay coefficients of variation were less than 4% for the gonadotropins and less than 7% for estradiol. Interassay coefficients of variation were less than 10% for the gonadotropins and less than 15% for estradiol.

Although 332 depressed and 644 nondepressed women were initially recruited, certain women were excluded from these analyses because of complexities surrounding the measurement of either hormonal or menstrual cycle indicators of the perimenopausal transition (Table 1). For analyses that prospectively assessed either menstrual cycle or hormonal changes as markers of perimenopause, women who were lost to follow-up within 6 months of study enrollment or reported taking exogenous hormones at study enrollment were excluded. We also excluded all women who reported at baseline enrollment menstrual cycle irregularities, longer or shorter cycle lengths of 7 days or greater, or a history of amenorrhea (more than 3 missed menstrual cycles). These women were excluded to ensure that the remaining study population comprised truly premenopausal women. For the analyses that assessed hormonal changes over time, we excluded individual blood specimens at certain follow-up periods that were drawn while a participant was taking exogenous hormones, pregnant, or breastfeeding. In addition, we excluded an additional 25 depressed and 46 nondepressed women who were unable to provide at least 2 valid (no exogenous hormonal influence) menstrually timed early follicular-phase blood specimens or whose FSH levels were greater than 20 mIU/mL at study enrollment (suggestive of postmenopausal levels).

Table 1 also illustrates the proportion of women who dropped from the study after each follow-up period. During the 3-year follow-up, we retained more than 80% of the entire cohort, and the dropout rate from the study was not appreciably different between depressed and nondepressed cohort members. Thus, any differences observed between depressed and nondepressed women with respect to changes in menstrual cycle characteristics or hormonal values were not likely to be due to differential loss of subjects in the depressed or nondepressed cohort over time.

All analyses were conducted with SAS statistical software.²⁴ We compared the distribution of background history characteristics between women with and without a lifetime history of depression to identify potential confounding factors related to both depression status and our hormonal or menstrual markers of perimenopause. Crude odds ratios with 95% confidence intervals were used as the measure of association.

We used survival analysis to assess the influence of depression on time to perimenopause. The risk period was defined as the time interval in months from the baseline interview to the follow-up interval in which a woman reported the inception of perimenopause or until the first of any censoring event. Censored data were from women who became pregnant (n = 41), had a hysterectomy or bilateral oophorectomy (n = 8), initiated hormone replacement therapy (n= 11), or dropped out of the study before becoming perimenopausal (n = 112). Women who remained premenopausal were censored at the end of the study observational period (n = 304). We examined the distribution of the outcome variable among all women, stratified by selected baseline characteristics and also by depression status at study enrollment. Kaplan-Meier product-limit survival analyses were used to estimate the median time to inception of perimenopause for the whole sample and to make univariate comparisons between depressed and nondepressed women.

Cox proportional hazards regression models²⁵ were used to estimate the hazard ratio and 95% confidence intervals for the relationship between depression and inception of perimenopause during 36 months of follow-up, while adjusting for age, age at menarche, parity, and education. Although they are considered potential intermediates along the causal pathway, we further adjusted for body mass index and cigarette smoking to determine the direct effect of depression on outcome. With the Proportional Hazards Regression procedure in SAS software (version 8.0), we used the "exact" option to handle a high proportion of ties as a result of imprecise measurement of event times.²⁴

Measures of FSH, LH, and estradiol became normally distributed after log transformation. The means of the log-transformed hormonal values were calculated for each of the blood donation periods (ie, baseline, 6-month follow-up, 12-month follow-up, etc) for depressed and nondepressed subjects. A model was constructed to test the change in hormonal values over time and between women with and without major depression by means of longitudinal analysis for repeated measures, with depression status and other potential confounders as the independent variable and each of the repeated hormonal values as the dependent variable.

To address our concern that use of hormone replacement therapy may be related to the onset of perimenopausal symptoms, we conducted separate "competing risks" analyses²⁶ in which 9 women who initiated hormone replacement therapy before the onset of perimenopausal symptoms were reclassified as having had the outcome event. These analyses showed no appreciable differences when compared with the original outcome classification, which censored these women.

Women with a history of depression were slightly older than women with no depression history (Table 2), but were similar in race, educational attainment, past use of oral contraceptives, and a history of having had 1 ovary removed. Compared with women with no history of mood disorder, women with a history of depression reported an earlier age at menarche, fewer live births, a greater body mass index at study enrollment, and a history of marital disruption and were somewhat more likely to be current cigarette smokers.

On the basis of our definition of perimenopause discussed in the "Methods" section, Figure 1 illustrates the time in months to perimenopause among women with and without a history of major depression (A) and further stratified within women with a lifetime history of depression by Hamilton depression scores less than or equal to 8 or more than 8 (B). During the 36-month follow-up, a greater proportion of women with a history of depression reached the perimenopausal end points sooner than women with no depression history. The greatest differences were observed after 18 months of follow-up and were largely confined to the women with a history of depression with more severe depressive symptoms (Hamilton scores >8) at the time of study enrollment. Fewer than 10 women with no history of depression had Hamilton scores greater than 8, precluding any further stratification.

Using Cox proportional hazards modeling, we observed that women with a history of depression had a 20% increased rate of entering perimenopause sooner compared with women with no depression history, after adjustment for age, parity, age at menarche, education, cigarette smoking, and body mass index (Table 3). This association was stronger among the women with depression history who were taking antidepressants at the time of study enrollment. Also, as illustrated in Figure 1, women with a history of depression and Hamilton depression scores exceeding 8 had twice the rate of entering perimenopause sooner than that of women with no depression history. To distinguish the use of antidepressants from depression symptoms, we created a variable that combined the Hamilton score and current use of antidepressants. As shown in Table 3, among the women with a history of depression and low Hamilton scores, the use of antidepressants had little effect on the rate of transition to perimenopause. However, among women with a history of depression and Hamilton scores greater than 8, antidepressant use substantially increased the rate of transition to perimenopause by nearly 3-fold.

To determine whether the association between history of depression and perimenopause was dependent on the proximity of a depressive episode to study enrollment, we included recency of last depressive episode in the survival model. This analysis suggested that the greatest rate of entry to perimenopause was among the women with a history of depression whose most recent episode occurred more than 5 years before study enrollment. We also observed little difference in the rate of transition to perimenopause by age at first onset of depression or the number of previous depressive episodes (data not shown).

Because women had to be premenopausal at the time of study enrollment, we had a priori concerns that the older women (aged 41-45 years) may differ systematically from the younger women (aged 36-40 years). We therefore stratified by age to assess the association between depression and perimenopausal transition. We found that the influence of a history of depression on the perimenopausal transition was more observable among the younger cohort members. The more subtle effect among the older women could be a consequence of differential selection bias. Thus, a greater proportion of women with a history of depression compared with those without might have been deemed ineligible for the study because of the presence of perimenopausal symptoms at the time of eligibility screening for study enrollment.

We assessed differences at baseline and the change over time in early follicular-phase FSH, LH, and estradiol levels between women with and without a lifetime history of major depression. We limited our assessment to the blood drawn at study enrollment, and then only during the first 3 follow-up periods in which at least 80% of eligible women with and without a lifetime history of depression were able to provide specimens. The proportion of study subjects providing menstrually timed blood specimens declined over time and dropped precipitously at 24 months after enrollment, and more rapidly among women with a history of depression compared with nondepressed cohort members (data not shown). Although all cohort members were premenopausal at study enrollment, as study subjects began to enter their late reproductive years and the start of the perimenopausal transition, they began having irregular periods and amenorrhea, and some began hormonal therapy to regulate their menstrual cycles or to alleviate perimenopausal symptoms. In addition, it is known that as women approach menopause, the cyclical variability in gonadotropin and estradiol levels measured in the follicular phase increases substantially.²⁷ Thus, although the increase in FSH level over time is one of the best predictors of the decline in ovarian function, when measured cross-sectionally every 6 months, it is possible that a woman with a menopausal FSH level at some point during that time interval might be missed. Therefore, hormonal value comparisons between groups will be subjected to greater amounts of misclassification with increasing proximity to perimenopause.

Using the first 4 blood donations (baseline and 3 subsequent follow-up specimens), we assessed mean log-transformed FSH, LH, and estradiol values in women with and without a lifetime history of depression (Table 4). Each hormone was examined separately, and analyses were completed for all women and then repeated for the younger (aged 36-40 years at study entry) and older (aged 41-45 years at study entry) cohorts.

The FSH level increased during each period for all women and remained higher in women with, compared with women without, a lifetime history of depression. The magnitude of this difference was more pronounced among older cohort members. The LH level also increased during each period, with higher levels observed among women with, compared with women without, a history of depression. Estradiol levels increased modestly over time and were generally lower among women with, compared with women without, a history of depression.

One of the strongest risk factors for an early onset of menopause is cigarette smoking, and our group²⁸ has shown that cigarette smoking is also associated with depression. However, most studies have not assessed the temporal relationship between this association. Our group has shown in an earlier report that depressed women are at greater risk of becoming smokers, and smokers are also more likely to develop mood disorder.²⁸ Although controlling for cigarette smoking will likely attenuate the true association between depression and early perimenopausal transition, even after this adjustment we still observed an association suggesting that a substantial amount of the effect of depression on perimenopause is operating independent of cigarette smoking.

The validity of our reported associations is influenced by the accuracy of both our measured exposure of interest (lifetime history of depression) and our assessed outcomes (perimenopausal transition). It is unlikely that misclassification of our exposure of interest was a problem in this study. Well-established in-person structured clinical interviews were performed with all participants and were then systematically peer reviewed by our clinical experts (M.W.O. and L.S.C.). The outcome of menstrual cycle markers for perimenopause was based on 2 population-based surveys^17,21 in which the authors determined that no particular menstrual cycle change was a sole predictor of perimenopause. We therefore defined the transition to perimenopause as a change over time in any one of several key predictors, provided that menstrual cycle characteristics were consistent and stable before study enrollment. Although we observed a modest association between depression and menstrual cycle markers of perimenopause in our entire study population, what might explain the stronger association among our younger cohort members? Since women were required to be premenopausal at the time of study enrollment, it is likely that a greater number of women 41 to 45 years of age compared with women 36 to 40 years of age would not have been eligible for enrollment in our study because of either hormone use or menstrual cycle indicators of having already entered perimenopause at the time of study recruitment. If a sizable number of these women excluded were more likely to have experienced depressive disorder, then our sample of older depressed women may not represent the reproductive endocrine profile of premenopausal depressed women in this age group. This is not likely to be the case among the women aged 36 to 40 years, since far fewer of them will likely have been excluded from study participation as a consequence of either hormonal therapy or menstrual cycle markers related to perimenopause.

How accurate are hormonal markers of perimenopausal transition? It is known that as women approach perimenopause, there can be substantial variability in both gonadotropin and estradiol production.²⁷ We observed within our data some individuals who at one follow-up assessment had early follicular-phase premenopausal FSH levels and postmenopausal estradiol levels, and then at a subsequent follow-up assessment showed a reversal of this hormonal pattern. Thus, postmenopausal levels of FSH and estradiol can be seen despite continued menstrual bleeding. Therefore, when cross-sectionally assessed FSH and estradiol levels are evaluated prospectively during the late reproductive years, substantial variability will be observed. In contrast, LH levels usually remain in the reference range and appear to be less variable.²⁹ This may explain, in part, why our association of depression with LH levels was much greater than that observed with either FSH or estradiol levels. If the cycle-to-cycle variability in FSH and estradiol is nondifferentially distributed between women with and without a history of major depression, our estimates of the association between a history of depression and either higher FSH or estradiol values may in fact be an underestimate of the true association. If there is greater misclassification among our women with a history of depression, then the true direction of an association may be less clear. However, Cramer et al,²⁷ in an analysis restricted to women with no depression history, observed substantial hormonal variability, suggesting that this misclassification applies to all women.

Of further concern as women approach perimenopause is the difficulty in obtaining menstrually timed blood specimens. Many women begin hormone replacement to either regulate their menstrual cycles or alleviate somatic symptoms associated with perimenopause, which prohibits an accurate assessment of hormonal profiles. Furthermore, many women who remain free of hormone therapy begin having either irregular cycles or amenorrhea, which increases the difficulty of obtaining the timed specimens. In our analyses, we chose to include only data during the first 4 periods in which more than 80% of both cohort members were able to provide menstrually timed specimens. In addition, we eliminated any specific blood draws in which either pregnancy, breastfeeding, or exogenous hormones might have influenced the results. Thus, although we cannot rule out that our associations of higher FSH and LH levels and lower estradiol levels among women with, compared with women without, a lifetime history of depression could be due to measurement bias, we also cannot rule out that the associations we observed may actually be an attenuation of the true association.

Is there a biological rationale for a more rapid decline in ovarian function among women with, compared with women without, a lifetime history of major depression? One could hypothesize that depression may alter more permanently the response of the hypothalamic-pituitary-gonadal axis regulation to stressful events, as already seen in other neurobiological models of psychiatric disorders such as posttraumatic stress disorder.^30- 31 Our finding of a stronger association among women with greater Hamilton scores at study enrollment may also suggest that the context in which depressive symptoms occur may play a role in the association between depression and an alteration of the hypothalamic-pituitary-gonadal axis.

An association between a lifetime history of depression and an earlier transition to perimenopause may result in a prolonged exposure to a hypoestrogenic state, which has been associated with bone density loss, sexual dysfunction, a decline in cognitive function, and a potential increased risk of cardiovascular disease. Likewise, there are substantial morbidity and economic burden associated with major depression. The compounded burden of illness associated with depression and end-organ sequelae related to estrogen deficiency may be considerable. It is unclear, however, whether an earlier transition to perimenopause would, in fact, result in earlier menopause. If not, women with a history of depression may also have to face a longer period of menopausal transition before reaching menopause at an age similar to that of women with no history of major depression. Under this scenario, women with a history of depression would be exposed to a more prolonged perimenopausal period—a time that has been associated with greater vulnerability for new or recurrent depression.

Corresponding author and reprints: Bernard L. Harlow, PhD, Obstetrics and Gynecology Epidemiology Center, Brigham and Women's Hospital, 221 Longwood Ave, Boston, MA 02115 (e-mail: bharlow@rics.bwh.harvard.edu).

Submitted for publication October 25, 2001; final revision received April 30, 2002; accepted May 9, 2002.

This study was supported by grant R01-MH-50013 from the National Institute of Mental Health, Bethesda, Md.

We thank Evelyn Li and Vanessa Pratomo for laboratory assessment of hormonal measures; Jill MacRae, Nancy Mansbach, Rebecca Pettingill, Lynn Rogue, and Elizabeth Heckshire for psychiatric and medical history interviewing; Allison Fraer for database management and development; Donna Spiegelman, ScD, and Rebecca Liberman, MPH, for assistance in statistical programming and analyses; and Daniel Cramer, MD, ScD, and Robert Barbieri, MD, for advice regarding the interpretation of the hormonal data.