Hormones and Menopausal Status as Predictors of Depression in Womenin Transition to Menopause

Background  Associations between depressed mood and hormonal changes during transition to menopause are controversial. To our knowledge, there has been no prospective study of these associations in women commencing when they are premenopausal.

Objective  To longitudinally study the associations among reproductive hormones, menopausal status, and other predictors of depressed mood in midlife women.

Design  Cohort study with 6 assessment periods during a 4-year interval. Blood samples were collected 12 times during the follicular phase (days 2-6).

Setting  Philadelphia County, Pennsylvania.

Participants  A randomly identified, population-based, stratified sample of African American (n = 218) and white (n = 218) women aged 35 to 47 years with regular menstrual cycles, no hormonal or psychotropic medication use, and no serious physical or mental health problems at enrollment.

Main Outcome Measures  Center for Epidemiologic Studies Depression Scale score and history of depression via the Primary Care Evaluation of Mental Disorders.

Results  There was an increased likelihood of depressive symptoms during transition to menopause and a decreased likelihood after menopause after adjustment for other predictors of depression, including history of depression, severe premenstrual syndrome, poor sleep, age, race, and employment status (P = .03). The likelihood of depressive symptoms decreased for individuals with a rapidly increasing follicle-stimulating hormone profile (P≤.001) and also decreased with age compared with premenopausal women (P = .02). Participant aggregate profiles with increasing estradiol levels were significantly associated with depressive symptoms in bivariate analysis (P = .053).

Conclusions  Depressive symptoms as assessed herein increased during transition to menopause and decreased in postmenopausal women. Hormone associations provided corroborating evidence that the changing hormonal milieu contributes to dysphoric mood during transition to menopause.

WHETHER DEPRESSIVE symptoms are associated with changes in reproductivehormone levels such as those that occur during transition to menopause isa controversial issue in women's health. The low estrogen levels of postmenopausalwomen were long believed to be associated with involutional melancholia, andthe low estrogen theory was supported by the results of estrogen administration,which reduced depressive symptoms in some studies,^1-3 althoughothers found no greater effect than with placebo use.⁴ Arecent study⁵ indicated that estrogen deficiencymay play a role in depression in premenopausal women, although other researchers⁶ suggested that the destabilizing effect of repeatedfluctuations during menstrual cycles rather than low levels of estrogen maybe the important factor in dysphoric mood in cycling women.

Epidemiologic data⁷ indicate that therisks of depression are greatest in the reproductive years, when hormone levelsfluctuate during the menstrual cycle. Several studies^8-10 reportedthat irritability, nervousness, and dysphoria increased in the transitionto menopause, observations that are consistent with the possibility that fluctuationsor rates of change of estradiol are associated with depressive symptoms invulnerable women. Suppression of these hormonal fluctuations in premenopausalwomen eliminated menstrually related dysphoric symptoms; after pharmacologicsuppression of ovarian function, the administration of physiologic doses ofestrogen induced dysphoric symptoms in women with premenstrual syndrome (PMS)but not in controls.¹¹ However, in perimenopausalwomen with major depression, estrogen treatment effectively reduced the depression.^12-14 These diverse studiesnot only indicate that the association between estrogen and depression iscomplex and varies with context¹⁵ but alsosuggest that estrogen has a role in depressive symptoms in the menopausaltransition.

This study examines depressive symptoms and reproductive hormones duringa 4-year interval in the Penn Ovarian Aging Study, an ongoing longitudinalstudy of ovarian aging in a population-based cohort. The objectives are (1)to identify the prevalence of depressive symptoms during this interval; (2)to determine associations among measures of depressed mood, menopausal status,reproductive hormones, and other predictors of depressive symptoms reportedin the literature; and (3) to conduct a secondary analysis of these associationsfor the subgroup of women with a diagnosis of major depressive disorder (MDD)during the study. We hypothesize that the strongest predictor of depressedmood in mid-life women is a history of depression and that fluctuations orchanges in hormone levels are associated with depressed mood during transitionto menopause.

The population-based cohort was identified by random digit dialing tohouseholds in Philadelphia County as previously described¹⁶ andstratified to enroll equal numbers of African American and white women. Eligibilitycriteria for enrollment into the cohort included 35 to 47 years of age, menstrualcycles in the reference range (22-35 days) for the previous 3 months, andat least 1 ovary and uterus. Exclusion criteria were current use of psychotropicor hormonal medications, including hormonal contraception and replacementtherapies; pregnant or breastfeeding; serious health problems known to compromiseovarian function (eg, diabetes mellitus, liver disease, and breast or endometrialcancer); alcohol or other drug abuse within the past year; and non-Englishspeaking. The study was approved by the University of Pennsylvania institutionalreview board, and written informed consent was obtained from participants.

A total of 436 women (75% of those eligible) were enrolled in the cohort(218 were African American and 218 were white). At the sixth assessment period,approximately 4 years later, data were obtained from 353 participants (81%of the cohort). Comparisons of baseline data between the study group and the83 individuals who discontinued participation showed no significant differencesin demographic background variables or in measures of depressive symptoms,diagnosis of MDD, menopausal status, or hormone levels.

Data were collected during 6 assessment periods at approximately 8-monthintervals. Each assessment period had 2 visits, scheduled between days 2 and6 of 2 consecutive menstrual cycles, to obtain blood samples for hormone assays.This narrow visit window was selected to provide an accurate assessment ofbasal follicle-stimulating hormone (FSH) levels in the early follicular phase,when changes associated with ovarian aging were expected to be most pronounced.¹⁷ Repeated measures of the other study hormones (estradiol,dehydroepiandrosterone sulfate, luteinizing hormone, and testosterone) arealso believed to be most reliable in the follicular phase.¹⁸

Trained research interviewers administered a structured questionnaire,measured height and weight to determine body mass index, and collected bloodsamples for hormone assays. The structured interview focused on overall healthand included demographic background information, menstrual cycle dates, reproductivehistory, general health status and behaviors (including medications, smoking,alcohol and caffeine consumption, and history of depressive disorders), andcommon menopausal symptoms. Current age was determined by subtracting thebirth date from each interview date, and the standard categories of populationstudies and census data were used (ie, 35-39, 40-44, 45-49, and 50-54 years).

The primary outcome measure was the Center for Epidemiologic StudiesDepression Scale (CES-D),¹⁹ a 20-item self-reportinventory for assessing current depressive symptoms. The standard CES-D cutoffscore of 16 or greater was used to define high depressive symptoms; a higherCES-D cutoff score of greater than 24 was also examined as a closer approximationof a clinical diagnosis of depression. The diagnosis of DSM-IV MDD was obtained during each assessment period by trained researchinterviewers administering the Primary Care Evaluation of Mental Disorders,²⁰ which is a standardized and validated diagnosticassessment procedure designed for primary care research and practice thatyields DSM-IV diagnoses for mood, anxiety, alcohol,and eating disorders.

Blood samples were collected between days 2 and 6 of the menstrual cycleor 1 month apart in nonmenstruating women. The samples were centrifuged andfrozen in aliquots at −70°C. Assays were conducted at the GeneralClinical Research Center, University of Pennsylvania Medical Center, in batchesthat included 4 visits per participant to reduce the within-subject variabilitydue to assay conditions. Estradiol, FSH, luteinizing hormone, dehydroepiandrosteronesulfate, and testosterone levels were measured by radioimmunoassay using commercialkits (Coat-A-Count; Diagnostic Products Corp, Los Angeles, Calif). Assayswere performed in duplicate for all hormones, and they were repeated if thevalues differed by more than 15%. The interassay and intra-assay coefficientsof variation were calculated from the assays conducted for the 6 assessmentperiods, with all coefficients being less than 5%.

Menopausal status was determined from the data on menstrual bleeding,which included menstrual dates recorded in the individual's daily symptomdiaries, the menstrual dates at each study interview, the dates of the 2 previousmenstrual periods, the number of menstrual periods between assessments, andcycle length. The definitions for the menopausal status groups were basedon the Staging System for Reproductive Aging in Women, a consensus group statementsponsored by the National Institutes of Health, the North American MenopauseSociety, and the American Society for Reproductive Medicine.²¹ Atevery assessment period, each participant was assigned to one of the followingcategories based on bleeding patterns:

1. Premenopausal: regular menstrual cycles in the 22- to 35-day range

2. Early transition: change in cycle length of 7 days or longer ineither direction from the participant's own baseline for at least 2 cycles

3. Late transition: 3 to 11 months of amenorrhea

4. Postmenopausal: 12 months or more of amenorrhea without hysterectomy.

We required 2 cycles of change for the early transition and atleast 3 cycles of amenorrhea for the late transition phase because we consideredfewer cycles to be too unreliable.

The hormone variables were constructed to provide a summary profileof the hormone values for each participant during the study. An aggregateprofile for each hormone for each participant was calculated by fitting aregression model using the participant's hormone values during the study andage as the time axis. To capture the shape of the profile over time, a polynomialof 2 (3 for estradiol) was used. A mathematical transformation of the modeldesign space (age matrix) using orthogonal polynomials was conducted to constructindependent tests on the shape (linear trend, quadratic trend, or cubic) ofthe hormone profile during the study.^22,23 Theseregression coefficients summarized (1) the trajectory (direction) of the hormonelevels during the study (ie, the linear, quadratic, and cubic terms), (2)the estimated mean square error term that summarized the participant-specificfluctuations in the hormone level (residual heterogeneity), and (3) the numberof hormone measures for each participant (maximum of 12), and they were usedin the analyses of this study.

The selection of predictor variables was based on their significancein previous studies^24-29 andthe goals of this project and included race (white or African American), currentage, history of depression (reported in the baseline structured interviewor a Primary Care Evaluation of Mental Disorders diagnosis of depressive disorderat baseline), menopausal status (defined in the "Menopausal Status" subsection),poor sleep as reported at each assessment period on the validated St Mary'sSleep Questionnaire³⁰ (categorical responsesof badly, fairly badly, or very badly for the item "How well did you sleeplast night?"), hot flashes (yes or no), current smoking, severe PMS (yes orno; the "yes" group included only those who also rated severe interferencewith functioning as a 4 on a 4-point scale), history of postpartum depression,currently employed, education (high school or less vs more than high school),marital status (married, living with partner vs all other), and living children(yes or no). Current use of antidepressant medication was an exclusion criterionat enrollment but was queried at each follow-up and examined for its associationwith the outcomes in multivariate analyses.

The frequencies and distributions of the CES-D scores and the diagnosisof MDD were examined at each assessment period. Multivariate logistic regressionfor repeated measures was used to estimate the effects of the covariates ondepressive symptoms as assessed by CES-D scores (≥16 vs <16). All availabledata for each participant were included in the repeated-measures models. Varianceestimates for the Wald statistics of the true logistic coefficients were adjustedfor the repeated observations from each participant using generalized estimatesequations.³¹ In multivariate modeling, themodel began with race, age, history of depression, and menopausal status.The hormone variables (described in the "Hormones" subsection) were initiallyexamined in models adjusted for race, age, and history of depression. Allother potential predictors with P≤.20 in the preliminarybivariate analyses were then added sequentially to the models. Hypothesizedinteractions were examined. The final selection of covariates was guided bywhether the variable remained statistically significant at P≤.05 and whether its inclusion modified other significant associationsin the model by 15% or more.³² Secondary analysesusing the same procedures were conducted for the outcomes of CES-D score greaterthan or equal to 25 and for a diagnosis of MDD (yes or no).

A secondary analysis was conducted for the "new cases" of CES-D scoresof 16 or greater during the 4-year study using a conditional (fixed-effects)logistic model.³³ A second subanalysis wasconducted for the new cases of an MDD diagnosis during the study. Only factorsthat changed over time in the primary analyses (ie, menopausal status, poorsleep, severe PMS, hot flashes, and FSH levels) were included to identifywhich factors affected within-subject changes in CES-D scores or MDD diagnosis.The log of the mean FSH level (≥20 vs <20 mIU/mL) for each participantat each assessment period was examined. The hormone profiles were not usedin this analysis because the profiles, which represented participant aggregatesummaries over time, were inappropriate for the conditional model.

All analyses were performed using a statistical software package (SASversion 8.0; SAS Institute Inc, Cary, NC). Statistical tests were 2-tailed,with P≤.05 considered significant.

Of the 353 participants at the sixth assessment period, 21 did not haveusable blood samples, leaving 332 participants for this study. Another 36observations from participants who were using hormone therapy or oral contraceptives,who were pregnant or breastfeeding, or who had a hysterectomy or an ovariectomywere not included in the analyses. The mean (SD) age of the cohort at theend point was 44.6 (3.5) years (range, 38-52 years). Half of the cohort wasAfrican American (n = 165) and half was white (n = 167); 59% had an educationbeyond high school, and 82% were employed.

All participants were premenopausal at enrollment. Menopausal statusat each assessment period is given in Table1. After 4 years, 73% of participants remained premenopausal, 21%were in early transition (≥7-day change in either direction in cycle lengthcompared with baseline for at least 2 cycles), 3% were in late transition(3-11 months with no menses), and 3% were postmenopausal (no menses for ≥12months). Participant ages ranged from 38 to 51 years in the premenopausalgroup and from 44 to 51 years in the postmenopausal group.

In Table 1, CES-D scoresof 16 or higher increased during transition to menopause and were lower aftermenopause (P = .047). A DSM-IV diagnosis of MDD ranged from 10% to 13% in premenopausal women andwas less frequent (≤4%) in women in the transition phases, although thenumbers in the transition phases were too small for analysis (Table 2). A history of depression at baseline was 21%. As expected,participants with a history of depression were statistically significantlymore likely to report high CES-D scores and to have an MDD diagnosis duringthe study than were those with no history of depression (Table 3).

The bivariate associations of the study variables with CES-D scoresand with MDD diagnosis during the 4-year study are given in Table 4. Menopausal status was significantly associated with CES-Dscores (P = .047), with an increased likelihood ofscores of 16 or higher in the transition phases and a lower likelihood aftermenopause compared with premenopausal women. All but 3 of the other 14 studyvariables were statistically significantly associated with CES-D scores. Theodds ratios (ORs) for the associations with MDD were generally similar tothose for the associations with CES-D scores. The associations of menopausalstatus and age did not reach statistical significance because of the muchsmaller numbers with MDD (10%-15% of the cohort) and because there were only3 participants with MDD in the late transition and postmenopausal phases (Table 4).

There was a significant inverse association between FSH levels and CES-Dscores adjusted for age, race, and history of depression (Table 5), indicating that the likelihood of depressive symptomsdecreased in participants with rapidly increasing FSH levels (quadratic profile, P = .002). The associations between the participant aggregateFSH profiles and menopausal status are depicted in Figure 1, which shows that FSH levels (summarized during the 4-yearstudy) were low in the premenopausal group and increased in the transitiongroups.

Women who experienced increased estradiol levels during the study weremore likely to report high CES-D scores (quadratic profile, P = .05), although the overall association of estradiol did not reachstatistical significance (Table 5).This pattern of higher estradiol levels observed in early transition to menopause³⁴ is believed to result from an increased ovarian follicularresponse to increased FSH levels.³⁵ An associationbetween decreasing estradiol levels and CES-D scores was also observed (cubicprofile, P = .07), but it was limited by the smallnumber of participants who reached menopause during the study. The hormoneprofiles of luteinizing hormone, dehydroepiandrosterone sulfate, and testosteronewere not significantly associated with CES-D scores. Only 3 women with MDDreached the late transition or postmenopausal phase in this study, an insufficientnumber for determining associations between the hormone profiles and the MDDdiagnosis.

Table 6 gives the finalmultivariate model of independent predictors of depressive symptoms controlledfor all other variables in the model. Women in early transition to menopausewere 55% more likely to report high CES-D scores, and women in late transitionto menopause were nearly 3 times more likely to report these depressive symptomscompared with premenopausal women (OR, 2.89; 95% confidence interval [CI],1.29-6.45; P = .01). Depressive symptoms decreasedwith age, with the 50- to54-year age group being 66% less likely to experiencedepressive symptoms compared with the 35- to39-year age group (OR, 0.34; 95%CI, 0.11-1.03; P = .06). The likelihood of depressivesymptoms decreased for participants with a rapidly increasing FSH profileduring the study (P≤001). Women with a historyof depression were more than twice as likely to report depressive symptoms(OR, 2.45; 95% CI, 1.61-3.72; P<.001). AfricanAmerican women remained nearly twice as likely to report depressive symptomsas white women (OR, 1.89; 95% CI, 1.35-2.63; P<.001).Other significant predictors of depressive symptoms included severe PMS, poorsleep, and lack of employment (Table 6).

The same set of variables was examined in a secondary analysis of thesubgroup with a diagnosis of MDD (assessed at each of the 6 study periods).The major predictor of MDD was a history of depression, with women who hada history of depression being nearly 5 times more likely to have an MDD diagnosisduring the 4-year study (OR, 4.75; 95% CI, 3.17-7.13; P<.001). Other significant predictors of MDD controlled for all othervariables in the model were severe PMS, poor sleep, and hot flashes. Adjustmentfor antidepressant medications, which were more likely to be used in the MDDgroup, did not alter the results. Menopausal status and age were not significantpredictors of MDD, but the number of women in the MDD group who reached thelate transition or postmenopausal phase was too small for meaningful analysis.Similarly, the lack of association with FSH profiles was likely to be becauseonly 8 women with a diagnosis of MDD reached an FSH level greater than 20mIU/mL during the study.

The results of a secondary analysis of CES-D scores using a cutoff scoregreater than 24 (to provide a closer approximation to clinical depression)were similar to those given for MDD in Table 6 (data not shown).

We conducted a secondary analysis of the subgroup of women who werenew cases of high CES-D scores during the study (n = 152) to determine whichfactors measured at each assessment period (menopausal status, poor sleep,PMS, hot flashes, and FSH levels) were associated with these new cases withno previous history of depression. The new cases were nearly twice as likelyto occur in the early transition phase compared with the premenopausal phase(OR, 1.94; 95% CI, 1.01-3.74; P = .048), and agewas inversely associated with high CES-D score (OR, 0.81; 95% CI, 0.69-0.96; P = .01) (Table 7).Although higher FSH levels were nearly twice as likely to be associated withhigh CES-D scores, the association was not statistically significant in thissmall subgroup. The same analysis was conducted for the new cases of MDD (n= 53). Although the OR for new cases of MDD increased in the early transitionphase, the number of participants was too small for reliable analysis (Table 7).

In this longitudinal, population-based cohort study, depressive symptomsas measured by the CES-D significantly increased during transition to menopauseand decreased after menopause after controlling for other important variables,including history of depression, age, PMS, poor sleep, hot flashes, race,and employment status. Consistent with the association of menopausal phase,depressive symptoms also decreased with age, with the oldest women (50-54years) being 66% less likely to report depressive symptoms. When we examinednew cases of depressive symptoms in women who had no history of depressionbefore enrolling in the cohort, we found that depressive symptoms were nearlytwice as likely to occur in the early transition phase. New cases of MDD diagnoseswere also more likely to occur in the early transition phase, although thesmall numbers in the MDD group were insufficient for statistical significance.These findings add support to another recent community-based study³⁶ of more than 2000 women that used the Edinburgh DepressionScale and identified increased depressive symptoms in the transition frompremenopause to perimenopause. The Study of Women's Health Across the Nation³⁷ also reported that rates of "psychologic distress"were highest in early perimenopausal women. In the Massachusetts Women's HealthStudy,³⁸ the percentage of women with depressivesymptoms (as assessed by the CES-D) also was highest in the perimenopausalgroup compared with premenopausal and naturally postmenopausal women.

The hypothesis that fluctuations in estradiol levels were associatedwith depressive symptoms was supported by the evidence that women with increasingestradiol profiles, which occur with ovarian aging in early transition tomenopause,³⁴ reported more depressive symptoms.Decreasing estradiol levels, which occur closer to menopause, were marginallyassociated with more depressive symptoms, although less than 5% of the cohortreached menopause during the study, providing little power for this analysis.It will be important to further investigate the associations between fluctuationsin estradiol levels and depressive symptoms as larger numbers of women inthis cohort reach menopause.

Less expected was the statistically significant inverse associationbetween depressive symptoms and FSH profiles, indicating that the likelihoodof depressive symptoms decreased in the presence of rapidly increasing FSHlevels during the 4-year study. It is unlikely that FSH contributes directlyto depressive symptoms, but, as a marker of ovarian aging, these results providecorroborating hormonal evidence that menopausal status is associated withthe dysphoric mood observed in menopausal transition.

In secondary analyses, a history of depression was the strongest predictorof MDD in this cohort. Women who had a history of depression were nearly 5times more likely to have an MDD diagnosis during the study. Severe PMS, poorsleep, and hot flashes also independently predicted MDD. The vasomotor findingssupport previous results, which indicated that perimenopausal women with vasomotorsymptoms were more than 4 times more likely to be depressed.³⁹ However,the present results also showed that hot flashes were not statistically significantlyassociated with the subclinical levels of depressive symptoms as assessedby the CES-D, a difference that warrants further study of the relationshipbetween vasomotor and depressive symptoms. Inasmuch as only 3 individualswith MDD reached the late transition or postmenopausal phase, the data wereinsufficient for determining associations of menopausal phase and hormonemeasures in the secondary analysis of MDD. Nonetheless, the findings indicatethe importance of further study of the associations between hormonal changesand major depressive episodes in larger numbers of women with a diagnosisof MDD.

African American women remained nearly twice as likely as white womento report depressive symptoms after adjustment for the other study variables,including employment, menopausal status, history of depression, PMS, poorsleep, hot flashes, and FSH level. Racial differences may have occurred becauseof unassessed variables that were beyond the focus of the project such aslife events or other psychosocial factors that differentially affect race.Another possibility is a differential response bias in the self-reported CES-Dscore, although the racial difference was also found with the MDD diagnosis,which was obtained by interview. Similar findings of an increased prevalenceof depressive symptoms in African Caribbeans compared with European whiterespondents in a community-based study⁴⁰ inBritain suggest that the racial difference is not simply an artifact of studymeasures but is an important factor for further study and treatment of depressivedisorders.

There are several limitations to this study. The study was not designedto address questions about postmenopausal status, and few women reached menopausein the 4-year study (<5% of the cohort). The hormone measures were in thefollicular phase and do not address other questions that require luteal phasemeasures, determination of the midcycle peak, or 24-hour blood sampling. Theeffect of measurement error, which results in bias toward the null hypothesisand which can potentially be increased if several variables such as hormonesand symptoms are measured with error, cannot be excluded. Only a subgroupof women in the randomly identified, population-based cohort had a diagnosisof MDD during the study (10%-15%, which is consistent with population estimates⁷), and further study of MDD to confirm the presentresults is needed. Participant reports of severe PMS (which was a strong predictorof depressive symptoms and MDD) were not confirmed by prospective daily symptomratings, although the prevalence of 14% for women with severe PMS at baselineis well within the range identified in epidemiologic studies.⁴¹ Finally,the participants represented a population-based sample of African Americanand white women, and the results cannot be generalized to women in nonurbanareas or to other racial groups.

A major strength of this study is the unique focus on hormones and symptomsin early transition to menopause, a period which has had little systematicexamination to date. The cohort was premenopausal at enrollment and was followedat 8-month intervals for 4 years to assess symptoms and hormones in conjunctionwith the ovarian decline that leads to menopause. The response rates in thislong-term prospective study remained high, and the multiple observations duringthe study were sufficient for the main aims and conclusions of this study.

In summary, depressive symptoms increased during transition to menopauseas identified by menstrual bleeding patterns. A significant inverse associationof participant aggregate profiles of FSH with depressive symptoms providedstrong corroborating evidence that the changing hormonal milieu contributesto dysphoric mood in this transition period.

Corresponding author and reprints: Ellen W. Freeman, PhD, Departmentof Obstetrics/Gynecology, University of Pennsylvania School of Medicine, MuddSuite, 3701 Market St, Suite 820, Philadelphia, PA 19104-5509 (e-mail: freemane@mail.med.upenn.edu).

Submitted for publication September 9, 2002; final revision receivedJune 9, 2003; accepted July 2, 2003.

This study was supported by grants RO1-AG-12745 (Dr Freeman) and 2MO1RR-00040-37(General Clinical Research Center, University of Pennsylvania Medical Center)from the National Institutes of Health, Bethesda, Md.