Increased Estradiol and Improved Sleep, But Not Hot Flashes, Predict Enhanced Mood during the Menopausal Transition

Abstract

Background:

The antidepressant effect of estrogen in women undergoing the menopause transition is hypothesized to be mediated by central nervous system effects of increasing estradiol on mood or through a pathway involving suppression of hot flashes and associated sleep disturbance. Estrogen therapy (ET) and the hypnotic agent zolpidem were selected as interventions in a three-arm, double-blind, placebo-controlled trial to distinguish the effects of estradiol, sleep, and hot flashes on depression.

Methods:

Women with depressive disorders, hot flashes, and sleep disturbance were randomly assigned to transdermal 17β-estradiol 0.05 mg/d, zolpidem 10 mg/d, or placebo for 8 wk. Changes in serum estradiol, perceived sleep quality, objectively measured sleep, and hot flashes were examined as predictors of depression improvement [Montgomery-Åsberg Depression Rating Scale (MADRS)] using multivariate linear regression.

Results:

Seventy-two peri/postmenopausal women with depression disorders were randomized to 17β-estradiol (n = 27), zolpidem (n = 31), or placebo (n = 14). There was no significant difference between groups in depression improvement (overall MADRS decrease 11.8 ± 8.6). Increasing estradiol (P = 0.009) and improved sleep quality (P < 0.001) predicted improved mood in adjusted models but reduced hot flashes (P = 0.99) did not. Post hoc subgroup analyses revealed that the therapeutic effect of increasing estradiol levels on mood was seen in perimenopausal (P = 0.009), but not postmenopausal, women.

Conclusions:

For women with menopause-associated depression, improvement in depression is predicted by improved sleep, and among perimenopausal women, by increasing estradiol levels. These results suggest that changes in estradiol and sleep quality, rather than hot flashes, mediate depression during the menopause transition. Therapies targeting insomnia may be valuable in treating menopause-associated depression.

The risk of clinical depression is increased in women during the menopause transition (1, 2), but the basis for this increased risk is not well understood. During the perimenopause and early postmenopause, depression frequently co-occurs with insomnia and hot flashes, two other common menopause-associated symptoms (2–4, 6). Hot flashes are the primary symptom of the menopause transition (7), and both depression (1, 2) and sleep disruption (4, 8) are more common in peri-/postmenopausal women than similar-age premenopausal women. Sleep disturbance is both a core symptom of depression (9), and a risk factor for the development of new-onset or recurrent depression (10). Many studies have shown that the risk of depression in midlife women is increased by the presence of hot flashes and sleep disturbance (2, 6, 11), but other studies have concluded that hot flashes and sleep disturbance are not risk factors for menopause-associated depression (12). The strong associations between depression, hot flashes, and insomnia in some studies suggest that such symptoms may be causally related either because of a shared etiology or because changes in estradiol lead to a cascade of symptoms, with hot flashes causing sleep disturbance, which results in depression. However, how these menopause-related symptoms interrelate is not known.

Some have hypothesized that menopause-associated depression occurs because hot flashes disrupt sleep (13), and, by extension, that estrogen therapy (ET) treats depression because it suppresses hot flashes, which results in improved sleep, thereby improving mood because of a cascade effect (13, 14). This domino hypothesis suggests that the pathway from hot flashes to depression is mediated by sleep disruption, that repeated awakenings from sleep due to hot flashes leave women feeling fatigued and moody. Supporting this hypothesis is evidence that sleep deprivation impairs mood (15), that ET is most effective for treating sleep problems in women with hot flashes (16), and that ET is a highly effective treatment for depression in perimenopausal women, most of whom have hot flashes (17, 18), but that ET does not improve depression in older postmenopausal women without hot flashes (19).

Although the domino hypothesis is a widely touted clinical model for understanding the basis of menopause-associated depression and the beneficial effect of ET on mood, it has not undergone substantial empiric investigation. Recent observational studies exploring this model suggest that, although hot flashes and sleep are each strongly predictive of poor mood, the effect of sleep is stronger than that of hot flashes, and sleep does not mediate the effect of hot flashes on mood (20).

Alternative explanations for the increased risk of menopause-associated depression posit that depression is caused by the effect of fluctuations in estradiol on neurotransmitter systems in brain regions that regulate mood (1, 12). If menopause-associated depression results from variability of estradiol, then ET may treat depression because it increases estradiol levels, which regulates brain regions that influence mood.

Although ET is known to treat menopause-associated depression (17, 18), the mechanism of its therapeutic benefit remains unknown. Understanding the pathways through which estrogen treats depression may provide insight into the pathophysiology of menopause-associated depression and inform the development of treatment strategies for depression in this susceptible population. The aims of this study were to identify the pathways through which ET has a therapeutic effect on menopause-associated depression, specifically whether increasing estradiol levels, suppressing hot flashes, and/or improving sleep disturbance is required to treat depression.

We conducted a randomized, placebo-controlled clinical intervention study in peri/postmenopausal women who had a combination of a depressive disorder, hot flashes, and insomnia related to hot flashes to examine these pathways. Participants were randomly assigned to one of three interventions: 1) estrogen therapy, expected to increase estradiol levels and treat all three symptoms; 2) the hypnotic agent zolpidem, used as an active comparator, expected to treat sleep disturbance but not increase estradiol levels or treat hot flashes or depression; and 3) placebo. These treatment interventions were selected to generate a range of changes in estradiol levels and menopause-associated symptoms. We hypothesized that depression would improve in women treated with ET more than in those receiving a specific treatment for sleep disruption, suggesting that increasing estradiol levels treats depression in this population through its effect on the central nervous system, rather than by improvement of sleep disturbance associated with hot flashes.

Subjects and Methods

We designed an intervention study for peri- and postmenopausal women with unipolar depressive disorders, hot flashes, and sleep disturbance that would selectively increase estradiol levels and treat individual symptoms to create variability in serum estradiol levels, sleep disturbance severity, and hot flash frequency. To achieve this goal, 72 peri- and postmenopausal women with unipolar depressive disorders, hot flashes, and sleep disturbance were randomly assigned to 8 wk of treatment with transdermal 17β-estradiol 0.05 mg/d, zolpidem 10 mg/d, or placebo in a 2- to 2-to-1 ratio. Transdermal estradiol results in the most stable estradiol levels and the most physiological ratio of estradiol to the less bioactive estrone (21). To maintain the double-blind intervention, all participants wore an identical patch (17β-estradiol or placebo) continuously and took one pill (zolpidem or placebo) each night at bedtime. The study protocol was approved by the Partners Healthcare Systems Institutional Review Board, and written informed consent was obtained.

Subjects

Subjects were women 40–60 yr old who were perimenopausal (variable cycle length >7 d different from normal or more than two skipped cycles and an interval of amenorrhea of >2 to <12 months) or postmenopausal (amenorrhea >12 months or bilateral oophorectomy) according to standardized criteria (22). Hysterectomized women with ovarian preservation were included if their serum FSH was greater than 20 IU/liter.

Subjects were eligible if they met criteria for a depressive disorder and an insomnia syndrome involving awakenings induced by nocturnal hot flashes (23) and reported 14 or more hot flashes per week for 2+ wk. A unipolar depression diagnosis (major depression, dysthymia, or minor depression) was made using the Structured Clinical Interview for axis I Diagnostic and Statistical Manual of Mental Disorders-IV Disorders (24). Subjects were also required to have a Montgomery-Åsberg Depression Rating Scale (MADRS) score of 15–31 (25), consistent with mild to moderate depressive symptoms (26). Women judged to have severe depression (MADRS score >31, suicidal ideation, or psychotic symptoms) were excluded, as were those with bipolar disorder, panic disorder, obsessive-compulsive disorder, anorexia nervosa, or substance-use disorders.

Also required for eligibility was significant sleep disturbance meeting threshold criteria for an insomnia syndrome for at least 1 month (27) that included three or more awakenings per night occurring in association with hot flashes 3 or more nights per week and a deleterious impact of sleep disruption on daytime well-being or function. Using clinical interviews and the Sleep Disorders Questionnaire (28), those diagnosed with another primary sleep disorder [obstructive sleep apnea (OSA), periodic limb movement disorder, or narcolepsy] or with sleep symptoms judged to be unrelated to hot flashes were excluded.

Also excluded were women using hormonal medications, hypnotic agents, antidepressants, and other prescribed or over-the-counter agents with potential effects on hot flashes, sleep, or mood, abnormal mammograms, and any contraindication to or adverse response to ET or zolpidem. Additionally, women were excluded if they had any menstrual dysfunction or amenorrhea related to other etiologies that could obscure determination of menopause status.

Procedures

Participants were assessed at study entry, after a 1-wk run-in period and then after 1, 3, 5, and 8 wk of treatment. Sleep disturbance and hot flashes were document prospectively before treatment initiation during the run-in.

Depression symptoms were assessed using the clinician-rated MADRS (range 0–60) (25) and the self-rated Beck Depression Index (BDI; range 0–63) (29). Perceived sleep quality was documented using the Pittsburgh Sleep Quality Index (PSQI; range 0–21) (30). Hot flashes were measured subjectively using a daily diary throughout the study. In addition, during 2 nights before and at the end of treatment, objective hot flashes were recorded using a skin-conductance monitor (see below) concurrent with an actigraphic watch to measure objective sleep patterns. The Actiwatch-Score (Mini Mitter Co., Inc., Bend, OR) was used to calculate time spent in bed, time spent awake after sleep onset, sleep efficiency (percent of time spent asleep between bedtime and wake-up time), sleep-onset latency (minutes), total sleep time (number of minutes spent asleep), and number of awakenings. Data were collected in 30-sec epochs, with each awakening defined as a total activity count greater than a sensitivity threshold of 80 during each epoch.

One third of the study participants were randomly selected to undergo a screening polysomnogram (PSG) at entry to estimate the prevalence of co-occurring sleep apnea and periodic limb movement syndrome (PLMS). The PSG was conducted and scored according to standard procedures to define specific sleep stages, measures of sleep disruption, OSA (respiratory disturbance index >15), and PLMS (periodic limb movement arousal index >20). Because only a subgroup of women underwent the PSG, those with OSA or PLMS on the PSG remained eligible to participate.

Concurrent with 2-night actigraphy assessment before and at the end of treatment, hot flashes were recorded objectively using a sternal skin-conductance monitor (Biolog ambulatory recorder; UFI, Morro Bay, CA). Documentation of objectively measured hot flashes was not required for eligibility. This skin-conductance monitor is used widely and has good reliability and validity (31). Consistent with standard procedures, objectively measured hot flashes were defined as an increase in sternal skin conductance of at least 2 μmho during a 30-sec period using a 20-sec lockout period (32).

Hormone analysis

FSH was measured using a two-site monoclonal nonisotopic system according to the manufacturer's directions (AxSYM; Abbott Laboratories, Abbott Park, IL), as described previously (33), and expressed in international units per liter as equivalents of the Second International Pituitary Standard 78/549. Serum estradiol was measured using an automated, random access, microparticle enzyme immunoassay (AxSYM; Abbott Diagnostics, Inc., Abbott Park, IL) with an analytical sensitivity 20 pg/ml, intraassay coefficient of variation of 2.1–4.5%, and interassay coefficient of variation of 6.5–9.6%.

Statistical analysis

Multivariate linear regression models with mean change in depression symptoms on the MADRS from baseline to 8 wk as the dependent measure were built to determine the independent effects of estradiol, sleep, and hot flashes on mood. Models were adjusted for treatment assignment and baseline MADRS scores. Predictors of interest were changes from baseline to study end in serum estradiol, perceived sleep quality (PSQI), and hot flashes, as measured subjectively by the diary. Alternate assessments of hot flashes (objective nocturnal symptoms, subjective daytime only symptoms, and subjective nighttime only symptoms) and sleep (actigraphically measured time spent awake after sleep onset, sleep efficiency, sleep onset latency, and total sleep time) were also examined. The association of treatment assignment with age, race, and body mass index was examined to determine whether these demographic characteristics should be included as covariates, but none met criteria (P < 0.10) and were therefore not included in the models. Post hoc subgroup analyses by menopause status were conducted to determine whether predictors of treatment response differed between peri- and postmenopausal women. Statistical significance was assumed at the two-sided α = 0.05 level.

Results

Of 82 women eligible to participate in the study, 72 (87.8%) were randomized to treatment in a 2- to 2-to-1 ratio (Fig. 1). Of those, 27 (38%) received 17β-estradiol, 31 (43%) received zolpidem, and 14 (19%) received placebo. Overall, 61 participants who initiated treatment (84.7%) completed the study (77.8, 83.9, and 100% of those on ET, zolpidem, and placebo, respectively). Four subjects (n = 2 on ET, n = 2 on zolpidem) withdrew because of side effects. Other causes of early withdrawal were time commitment (n = 2), loss to follow-up (n = 4), and concurrent enrollment in another study (n = 1).

Fig. 1.

Study participation by treatment assignment.

Subjects

There were no differences between groups in demographic, menopause, current or prior depression, or sleep characteristics at baseline (Table 1). The mean (±sd) age of study participants overall was 51.1 ± 5.0 yr. Two thirds of participants were Caucasian and one quarter were African-American. Menopause status included 48.6% postmenopausal women, 37.5% perimenopausal women, and 13.9% with hysterectomy only. The duration of time from the final menstrual period varied among the postmenopausal women (median 48 months, interquartile range 16–84 months), but time from final menstrual period did not differ between treatment groups. Body mass index did not differ between groups. Baseline serum FSH and estradiol levels were consistent with clinical determinates of peri/postmenopause status. Frequency of hot flashes did not differ between groups (median 5.2, interquartile range 3.4–9.1, hot flashes reported per 24 h).

Table 1.

Baseline characteristics by treatment assignment

	All (n = 72)	Estradiol (n = 27)	Zolpidem (n = 31)	Placebo (n = 14)	P value
Age (yr)	51.1 ± 5.0	50.0 ± 4.8	51.3 ± 5.1	52.6 ± 4.9	0.28
Race					0.45
Caucasian (%)	48 (67.6%)	17 (63.0%)	22 (73.3%)	9 (64.3%)
African-American (%)	19 (26.8%)	9 (33.3%)	6 (20.0%)	4 (28.6%)
Othera	4 (5.6%)	1 (3.7%)	2 (6.7%)	1 (7.1%)
Marital status					0.79
Never married	18 (25.4%)	17 (63.0%)	7 (23.3%)	9 (64.3%)
Married/living with partner	18 (25.4%)	9 (33.3%)	10 (33.3%)	4 (28.6%)
Divorced/separated/widowed	35 (49.2%)	1 (3.7%)	13 (43.3%)	1 (7.1%)
Educational status					0.08
At least some high school	14 (19.7%)	9 (33.3%)	3 (10.0%)	2 (14.3%)
At least some college	48 (67.6%)	16 (59.3%)	24 (80.0%)	8 (57.1%)
Graduate school	9 (12.7%)	2 (7.4%)	3 (10.0%)	4 (28.6%)
Full- or part-time employment	55 (77.5%)	20 (74.0%)	23 (76.7%)	12 (85.7%)	1.0
Body mass index (kg/m2)					0.45
Normal (<25)	18 (27.3%)	5 (20.8%)	8 (28.6%)	5 (35.7%)
Overweight (25–29.9)	25 (37.9%)	9 (37.5%)	13 (46.4%)	3 (21.4%)
Obese (≥30)	23 (34.8%)	10 (41.7%)	7 (25.0%)	6 (42.9%)
Current depression disorder					0.46
Major depression	47 (65.3%)	17 (63.0%)	22 (71.0%)	8 (57.1%)
Minor depression	9 (12.5%)	4 (14.8%)	3 (9.7%)	2 (14.3%)
Dysthymia	16 (22.2%)	6 (22.2%)	6 (19.4%)	4 (28.6%)
Depression symptom scores
MADRS score	20.9 ± 4.7	21.6 ± 4.4	20.6 ± 4.7	20.1 ± 5.6	0.58
BDI score	17.9 ± 8.9	20.2 ± 6.6	19.3 ± 8.6	15.5 ± 5.8	0.16
Proportion with prior depression					0.32
No prior episodes	20 (27.8%)	7 (25.9%)	9 (29.0%)	4 (28.6%)
One prior episode	26 (36.1%)	12 (44.4%)	7 (22.6%)	7 (50.0%)
2+ prior episodes	25 (34.7%)	8 (29.6%)	14 (45.2%)	3 (21.4%)
Menopause status		10 (37.0%)	11 (35.5%)	6 (42.9%)	0.39
Perimenopausal	27 (37.5%)	14 (51.9%)	13 (41.9%)	8 (57.1%)
Postmenopausal	35 (48.6%)	3 (11.1%)	7 (22.6%)	0 (0%)
Hysterectomyb	10 (13.9%)
Number of months from final menstrual period (postmenopausal only; median, interquartile range)	48 (16–84)	36 (19–120)	48 (20–72)	46 (13.5–85.5)	0.97
FSH (IU/liter)	101.7 ± 58.9	81.0 ± 60.2	114.7 ± 58.3	112.9 ± 49.0	0.07
Estradiol (pg/ml)c	20 (19–48)	23 (19–52)	19 (19–48)	21 (19–48)	0.88
Vasomotor symptoms
Number reported per 24 h	5.2 (3.4–9.1)	4.8 (3.2–7.0)	5.7 (2.9–10.0)	5.4 (3.7–10.1)	0.88
Number reported during night	2.3 (1.5–3.4)	2.6 (1.6–3.3)	2.0 (1.5–3.8)	2.6 (1.0–3.3)	0.98
Number measured during night	2.0 (1.0–4.0)	1.5 (1.0–4.0)	2.3 (0.8–3.8)	3.0 (1.0–3.5)	0.81
Sleep quality (PSQI score)	11.9 ± 3.1	11.7 ± 3.2	12.0 ± 2.9	11.9 ± 3.5	0.95
Actigraphy-based sleep measures
Sleep latency (min)	20.5 (9.0–34.5)	23.0 (16.0–39.0)	18.0 (8.0–26.5)	13.0 (8.5–41.5)	0.16
Sleep efficiency (%)	87.1 (80.1–89.8)	84.8 (78.1–88.9)	87.7 (80.8–91.6)	87.1 (83.7–89.4)	0.18
Wake-time after sleep onset (min)	25.8 (17.3–42.0)	26.5 (17.3–49.0)	25.5 (18.5–44.5)	26.8 (16.0–31.5)	0.74
Total sleep time (min)	373.3 (333.0–426.5)	372.5 (348.0–407.0)	412.3 (339.3–446.3)	355.8 (285.5–397.5)	0.44
Number of awakenings per hour	2.1 (1.7–2.7)	2.1 (1.6–3.0)	2.1 (1.8–2.7)	2.2 (1.8–2.6)	0.75

Normally distributed data are presented as mean ± sd, nonnormal data are presented as median (interquartile range), and categorical data are presented as n (%).

^aOther race/ethnicities: Asian (n = 1), Hispanic (n = 2), other (n = 1).

^bEligible if FSH is greater than 20 IU/liter.

^cEstradiol levels assigned as 19 pg/ml if less than 20 pg/ml.

Two thirds of participants met criteria for a major depressive episode, whereas the remainder had dysthymia or minor depression (Table 1). There were no differences between groups in the severity of depressive symptoms at baseline, with MADRS and BDI scores reflecting moderate levels of depressive symptoms (20.9 ± 4.7 and 17.9 ± 8.9, respectively). The majority of women had a prior history of depression.

Baseline measures of perceived sleep quality and objectively assessed sleep disturbance did not differ between groups (Table 1). The mean PSQI score was 11.9 ± 3.1; 96% had a PSQI score greater than 5, indicating poor sleep quality (34). Actigraphic measures revealed that participants slept an average of 6¼ h per night and had a median sleep efficiency of 87.1%.

Screening PSG studies were completed by 23 randomly selected participants (31.9%). Of 22 who had interpretable PSG studies (seven assigned to zolpidem, eight to ET, and seven to placebo), eight (36.4%) met criteria for sleep apnea (four assigned to zolpidem, one to ET, and three to placebo). No one met criteria for PLMS.

Treatment effects

Depression improved in all three treatment groups, without significant differences between groups (Fig. 2). The overall decrease in MADRS scores was 11.8 ± 8.6 over the 8-wk treatment period. On average, estradiol levels increased in the ET group (28.3 ± 56.7 pg/ml) and decreased in the zolpidem and placebo groups (−12.4 ± 83.8 and −13.5 ± 47.9 pg/ml, respectively), but these changes did not differ statistically between groups (P = 0.12). Sleep quality improved more (P = 0.049) in women treated with zolpidem (PSQI decreased by 5.7 ± 3.8) than with ET (2.5 ± 4.8) or placebo (3.8 ± 4.8), whereas objectively measured sleep patterns showed minimal change in any treatment group. For the group overall, perceived hot flashes were reduced by 3.3 ± 4.2 (37.9% ± 59.9%) per day. There were no statistically significant differences between groups in the extent of hot flash suppression, regardless of assessment method. However, ET tended to improve hot flashes more than other treatments, with perceived hot flashes reduced more by ET (4.2 ± 4.4) than zolpidem (3.3 ± 4.0) or placebo (2.2 ± 4.1). Women with the highest overall symptom burden (defined as greater than the median split on the MADRS, PSQI, and perceived hot flash frequency) had a greater improvement in depressive symptoms when treated with ET than the less symptomatic group (MADRS reduction 17.5 ± 7.3 vs. 9.4 ± 7.6, P = 0.04), but the two groups did not differ in response to zolpidem or placebo.

Fig. 2.

Mean (±sd) changes in depression symptoms on the MADRS by treatment assignment. Change in depression symptoms did not differ between treatment groups (P = 0.12).

Improvement in mood, as defined by a decreased MADRS score, correlated significantly with an increase in serum estradiol (r = −0.36, P = 0.007, Fig. 3A) and improved perceived sleep quality (PSQI; r = 0.51, P < 0.001, Fig. 3B), but there was no correlation of mood improvement with suppression of hot flashes (Fig. 3C) or changes in objectively measured sleep patterns. Mood improvement was not associated with menopause status, baseline levels of hot flashes, sleep quality, or objective sleep patterns or baseline or study-end levels of estradiol or FSH. Changes in estradiol, PSQI scores, and hot flashes did not correlate with each other.

Fig. 3.

Correlations between improvement in depression (change in MADRS score) and change in serum estradiol level (A), change in perceived sleep quality (B), and change in subjectively measured hot flashes (C) across all study subjects. All change values represent the difference of baseline subtracted from study end.

In adjusted regression models, increases in estradiol levels (P = 0.009) and improvement in perceived sleep quality (P < 0.001) both remained statistically significant predictors of improvement in depression (Table 2). Suppression of hot flashes was not associated with improvement of depression in adjusted models, regardless of how hot flashes were assessed.

Table 2.

Effect of changes in perceived sleep quality, serum estradiol levels, and hot flashes on improvement in depression symptoms (MADRS scores) using unadjusted and adjusted^a linear regression models

	Unadjusted			Adjusteda
	β	se	P	β	se	P
All (n = 72)
↑ sleep quality	0.99	0.22	<0.001	0.90	0.22	<0.001
↑ serum estradiolb	0.04	0.02	0.007	0.04	0.01	0.009
↓ hot flashes	0.23	0.28	0.42	−0.01	0.23	0.99
Perimenopausal (n = 27)
↑ sleep quality	0.73	0.31	0.03	0.65	0.23	0.02
↑ serum estradiolb	0.09	0.02	0.001	0.06	0.02	0.009
↓ hot flashes	0.08	0.34	0.82	−0.14	0.22	0.53
Postmenopausal (n = 35)
↑ sleep quality	0.89	0.23	0.001	0.89	0.30	0.008
↑ serum estradiolb	0.02	0.03	0.47	0.04	0.04	0.27
↓ hot flashes	0.33	0.42	0.45	−0.18	0.49	0.72

^aModel covariates include baseline MADRS, treatment assignment, and change from baseline to study end in sleep quality (PSQI), serum estradiol level (picograms per milliliter), and hot flashes (per 24 h period, as measured subjectively on a daily diary).

^bChange in serum estradiol levels relative to lowest quartile of change.

Differences by menopause status

Post hoc analyses of predictors of treatment response revealed that predictors of depression symptom improvement varied by menopause status (Table 2). Reduction in depression symptoms correlated with increasing estradiol levels in perimenopausal (r = −0.68, P = 0.001, Fig. 4A) but not postmenopausal women (r = −0.15, P = 0.47, Fig. 4A), whereas improvement in sleep quality correlated with improvement of mood in both groups (r = 0.47, P = 0.03, and r = 0.61, P < 0.001, respectively, Fig. 4B). These associations were unaffected by adjustment for key covariates in menopause status-specific models (Table 2), with improvement in sleep as the only significant predictor in postmenopausal women (P = 0.008) and increases in estradiol levels (P = 0.009) and improved sleep quality (P = 0.02) as predictors of mood improvement in perimenopausal women. The association between increasing estradiol levels and improvement in mood persisted (P = 0.024) among the subgroup of perimenopausal women who continued to have sleep disturbance [PSQI reduction <3 (34)]. For postmenopausal women, duration of time from the final menstrual period did not correlate with improvement in mood (r_s = 0.27, P = 0.16), even among the subgroup treated with ET (r_s = 0.46, P = 0.18). Objective measures of hot flashes and sleep disturbance did not predict mood improvement in adjusted models for either menopause status group.

Fig. 4.

Correlations between improvement in depression (change in MADRS score) and change in serum estradiol levels (A) and between change in MADRS score and in perceived sleep quality (B) separated by perimenopausal (□) and postmenopausal (▾) status. All change values represent the difference of baseline subtracted from study end.

It is notable that increases in estradiol levels correlated with mood improvement in perimenopausal women, regardless of whether ET or another treatment was administered, although estradiol levels did not increase from baseline for the perimenopausal group as a whole (−1.7 ± 61.4 pg/ml). In contrast, there was no association between increasing estradiol levels and mood improvement in postmenopausal women, even among those treated with ET whose estradiol levels increased (57.4 ± 51.6 pg/ml). Estradiol levels increased more on ET than other treatments among postmenopausal (P < 0.001), but not perimenopausal (P = 0.33), women. Therefore, although serum estradiol increased in postmenopausal women on ET, this elevation in estradiol did not result in improved mood. Mood improvement did not correlate with baseline or study-end levels of estradiol or FSH for peri- or postmenopausal women.

Medication tolerability

Study interventions were well tolerated. Of four women who withdrew because of side effects, two on ET (7.4%) withdrew because of persistent insomnia and mood worsening (n = 1) and medical complications related to a previous surgery (n = 1), and two on zolpidem (6.5%) withdrew because of daytime sedation (n = 1) and transient neurological symptoms (n = 1).

Discussion

Results of this intervention study show that treatment of depression is predicted by improved sleep quality, but not suppression of hot flashes, in peri- and postmenopausal women with co-occurring hot flashes and insomnia and by increased serum estradiol levels in perimenopausal women only. Improvement of depression did not differ based on random assignment to treatment with estrogen, zolpidem, or placebo. However, the range of changes in serum estradiol levels, sleep quality, and hot flashes that were generated using this intervention study provided insight into the independent impact of these predictors on depression in this population. We observed that predictors of mood improvement differed by menopause status. Better perceived sleep quality, but not increased estradiol levels, predicted mood improvement in postmenopausal women despite the increase in estradiol seen in those receiving ET. In contrast, in perimenopausal women, an increase in both estradiol levels and sleep quality predicted improvement in depressive symptoms, suggesting that perimenopausal women have a heightened sensitivity to mood changes related to estradiol. Our results indicate that targeting insomnia in peri- and postmenopausal women, and increasing estradiol levels in perimenopausal women, may be critical in the overall management of depression, whereas suppression of hot flashes is not a priority when treating depression.

The current study was designed to generate changes in menopause-related symptoms and estradiol to disentangle the pathways through which these commonly co-occurring symptoms interrelate and determine how ET treats menopause-associated depression. We had initially hypothesized that ET would have a greater effect than zolpidem on depression because we expected that treatment of sleep disturbance alone would not be an effective way to treat depression in this population. However, the selected treatments in this study did not treat depression differentially nor did the nature of the expected symptom responses align perfectly with each intervention. Estradiol levels increased only in the group on ET, but hot flashes improved in all treatment groups, although there was a trend toward hot flashes improving more on ET. Similarly, perceived sleep quality improved in all three treatment groups but more robustly in the zolpidem group. The strong therapeutic effect of zolpidem therapy on depression may be secondary to its potent effect on sleep quality, which exceeded that of ET.

The strong placebo effect on mood complicates interpretation of the comparison between ET and zolpidem. Placebo response rates are known to be high in randomized trials targeting depression (35) and hot flashes (36). Others have observed that co-occurring anxiety is an important factor that differentially affects response to placebo and active treatments in trials for women with hot flashes and depression, such that higher baseline anxiety predicts better response to placebo and worse response to the active intervention (36). Although anxiety commonly co-occurs with depression and hot flashes (7, 37), we did not measure anxiety and are therefore unable to assess its contribution to our findings.

Consistent with our hypothesis, we observed that an increase in estradiol levels predicted improvement in depressive symptoms in perimenopausal women. It is notable that this association was seen despite the observation that estradiol levels did not increase on average from baseline for perimenopausal women treated with ET. The absence of an overall increase is likely due to the variable and higher baseline estradiol levels in perimenopausal women, who intermittently produce estradiol. Nonetheless, at the level of the individual perimenopausal woman, the greater the increase in serum estradiol, the more robust was her mood improvement. These data are consistent with placebo-controlled trials showing benefit of ET specifically in perimenopausal women with depression (17, 18) and observational data linking depression to greater variability of serum estradiol in this population (1, 12). Our current findings support previous studies showing that ET treats depression in perimenopausal women likely because estradiol regulates central nervous system modulators that influence mood in the perimenopausal context of endogenous fluctuation of estradiol (38).

In contrast to perimenopausal women, elevation of estradiol levels did not predict mood improvement among postmenopausal women despite an increase in serum estradiol on ET. These results are consistent with trials showing that a higher dose of 17β-estradiol (0.1 mg/d transdermal) does not improve depression in postmenopausal women (19). The dose of ET used in both the previous (19) and current study may not have been high enough to ensure that the serum estradiol increased above an unknown threshold required for a therapeutic effect in postmenopausal women. The divergent association of estradiol levels and mood improvement between peri- and postmenopausal women supports the notion that perimenopausal women have a differential sensitivity to changing estradiol levels (38). In contrast to perimenopausal women (1, 12), changes in estradiol may not drive risk for depression in postmenopausal women, thereby limiting the potential therapeutic benefit of ET in the postmenopausal population. Alternatively, similar to the critical period hypothesis showing differential cognitive response to ET according to time from menopause (39–41), the absence of a beneficial mood response to increasing estradiol levels in postmenopausal women may result from a longer elapsed time in an estrogen-deficient state. We did not observe a correlation between time from menopause and mood improvement on ET in postmenopausal women but have a small number of women with which to examine this association. Given the variability in time from menopause in our postmenopausal group, it remains plausible that the absence of an association between an increase in estradiol levels and improved mood may in part be explained by the more prolonged interval among some postmenopausal women.

Results of our study demonstrate that treatment of insomnia improves depression in both peri- and postmenopausal women, supporting the notion that sleep disturbance has an independent effect on depression. These findings are consistent with trials showing that therapies targeting sleep disturbance optimize treatment of depression (42) and improve quality of life in women with hot flashes (5) and observational studies that have found that sleep disturbance has an independent negative effect on mood (20). Taken together, these findings emphasize the importance of close monitoring of sleep when treating depression. Given the beneficial effect of treatments targeting insomnia, future studies should examine whether combination therapy with ET and hypnotic agents like zolpidem might work best for multiply symptomatic peri/postmenopausal women because two pathways are targeted simultaneously.

As hypothesized, suppression of hot flashes did not predict improvement in depression. This important finding indicates that treatment of hot flashes is not required to treat depression and provides evidence that hot flashes are not, by themselves, central to the etiology of menopause-associated depression. Our results are consistent with other trials showing that ET treats depression among a subgroup of women without hot flashes (17). Other studies indicate that, although hot flashes and sleep disturbances are both predictive of poor mood, the effect of hot flashes on mood is not mediated by sleep disturbance (20).

The majority of study participants had experienced additional episodes of depression before study enrollment. The perimenopause is a period of risk for women who have (6) and who have not (1, 2) previously experienced depression, suggesting that this period of widely fluctuating estradiol levels confers a specific vulnerability to depression for both those with and without a prior depression history. However, it is not possible to definitely establish that a particular depression episode during the menopause transition is specifically menopause induced, regardless of whether it is a first or recurrent episode. Inclusion of women with depression that is either menopause induced or coincidental with the menopause transition may have introduced variability in mood responsivity to the specific interventions used in this study.

This study has notable strengths and several limitations that are intrinsic to the study population and trial design. Although the trial is the largest of its kind to date, it remains a small trial, and the relatively small size of the placebo group may have overestimated the true placebo response. Although restricting the sample to postmenopausal women without endogenous estradiol production would have been optimal to isolate the effect of exogenous ET, inclusion of perimenopausal women enabled us to make important observations that mood improves as estradiol increases in perimenopausal women, regardless of whether estradiol increased because of an exogenous or endogenous source. Finally, because women with OSA identified on the PSG were not excluded, inclusion of women with OSA may have influenced the therapeutic effect of our treatment interventions.

In summary, results of this intervention study in peri- and postmenopausal women with depression indicate that mood improves when sleep quality improves, regardless of the specific treatment used. Our results suggest that interventions targeting sleep disturbance should be prioritized in the treatment of menopause-associated depression. Our results also demonstrate that for perimenopausal, but not postmenopausal, women depression improves in concert with increases in serum estradiol, whether from an endogenous or exogenous estrogen source. Some perimenopausal women may benefit from estradiol therapy if the benefits outweigh the risks, although the use of ET for depression should be considered carefully because it is an off-label use. The absence of an association between suppression of hot flashes and improvement in depression indicates that the association between sleep and depression is independent of hot flashes. Our findings suggest that sleep disturbance may play a mechanistic role in menopause-associated depression and that clinicians caring for symptomatic peri/postmenopausal women should prioritize and closely monitor relief of insomnia as an integral part of depression treatment.