Disruption of Sleep Continuity During the Perimenopause: Associations with Female Reproductive Hormone Profiles

Abstract

Context

Nocturnal vasomotor symptoms (nVMS), depressive symptoms (DepSx), and female reproductive hormone changes contribute to perimenopause-associated disruption in sleep continuity. Hormonal changes underlie both nVMS and DepSx. However, their association with sleep continuity parameters resulting in perimenopause-associated sleep disruption remains unclear.

Objective

We aimed to determine the association between female reproductive hormones and perimenopausal sleep discontinuity independent of nVMS and DepSx.

Methods

Daily sleep and VMS diaries, and weekly serum assays of female reproductive hormones were obtained for 8 consecutive weeks in 45 perimenopausal women with mild DepSx but no primary sleep disorder. Generalized estimating equations were used to examine associations of estradiol, progesterone, and follicle stimulating hormone (FSH) with mean number of nightly awakenings, wakefulness after sleep onset (WASO) and sleep-onset latency (SOL) adjusting for nVMS and DepSx.

Results

Sleep disruption was common (median 1.5 awakenings/night, WASO 24.3 and SOL 20.0 minutes). More awakenings were associated with estradiol levels in the postmenopausal range (β = 0.14; 95% CI, 0.04 to 0.24; P = 0.007), and higher FSH levels (β [1-unit increase] = 0.12; 95% CI, 0.02 to 0.22; P = 0.02), but not with progesterone (β [1-unit increase] = −0.02; 95% CI, −0.06 to 0.01; P = 0.20) in adjusted models. Female reproductive hormones were not associated with WASO or SOL.

Conclusion

Associations of more awakenings with lower estradiol and higher FSH levels provide support for a perimenopause-associated sleep discontinuity condition that is linked with female reproductive hormone changes, independent of nVMS and DepSx.

Disruption of sleep continuity is common during the perimenopause (1) and characterized by more nightly awakenings and greater wakefulness after sleep onset (WASO), while reduced ability to fall asleep (sleep-onset latency, SOL) is less common during this reproductive transition. Approximately 56% of perimenopausal women report sleep discontinuity compared with 32% of premenopausal women of a similar age (2, 3). Sleep discontinuity results in reduced quality of life, and also an elevated risk of developing cardiovascular disease and depression, diseases that become more prevalent in women during the perimenopause. These data underscore the need to examine the impact of key physiologic changes during perimenopause on sleep parameters representing sleep discontinuity.

Hot flashes and night sweats, or vasomotor symptoms (VMS), increase in both prevalence and severity during the perimenopause (4), occurring in up to 70% of women across the menopause transition (5). Nocturnal VMS (nVMS) associate with greater subjective ratings of sleep difficulty (6) and with insomnia disorder (4, 7), involving nighttime awakenings and WASO when measured by either self-report (7, 8) or polysomnography (PSG) (7, 9), providing robust evidence to support a link between nVMS and disruption of sleep continuity. Despite this association, perimenopause-associated sleep continuity disruption can manifest in the absence of nVMS (2, 3). Notably, we have shown that, among women with VMS, the number of PSG-assessed nightly awakenings exceeds the number of nVMS (7). Similarly, others demonstrated that nVMS account for only 27.2% of PSG-assessed WASO in perimenopausal women (9). Together, these findings highlight that sleep discontinuity is not explained entirely by nVMS and suggest that other factors besides nVMS may drive the increased prevalence of perimenopause-associated disruption of sleep continuity.

During the perimenopause, diminished follicle development, reduced ovulation, and perturbation of the hypothalamic-pituitary-ovarian axis translate to greater variability in estradiol (E2) levels interspersed with periods of low and unchanging levels of E2, as well as less frequent production of progesterone, together with elevation of follicle stimulating hormone (FSH) (10-12). Changes in reproductive hormones have been suggested to explain perimenopause-associated disruption in sleep continuity independent of nVMS, but results across studies remain mixed (6, 8, 13). For example, while some population-based studies using self-reported sleep conclude that increasing FSH and/or decreasing E2 levels associate with higher likelihood of trouble falling and staying asleep (6) and with more nightly awakenings (8), others report no association (13, 14). The role of progesterone is less defined, although one cross-sectional study observed increasing levels of the urinary progesterone metabolite pregnanediol glucuronide associated with “trouble sleeping” (13). When utilizing PSG, studies report positive associations of higher and more rapidly increasing FSH levels with more WASO, awakenings, and arousals in perimenopausal women without sleep complaints (15, 16). However, associations of absolute levels or changing levels of serum E2 and progesterone with PSG-derived sleep parameters have not been observed to date (15, 16).

The mechanism through which gonadal steroids modulate sleep is not well understood, but these findings point to the potential of female reproductive hormones to directly modulate sleep. Animal models show estrogen and progesterone receptor expression in several central sleep/wake regulatory nuclei (17-19). Exogenous E2 administration modulates sleep deprivation-induced neuronal activation of these areas (20, 21) and alters sleep architecture (22, 23) and sleep/wake consolidation.

Despite the potential of the changing female reproductive hormone dynamics in the perimenopause to influence sleep continuity, their contribution to perimenopause-associated sleep disruption independent of nVMS remains largely undefined. Thus, our primary aim was to investigate whether perimenopause-associated sleep discontinuity is linked to either variability in or absolute levels of E2, FSH, and progesterone independent of nVMS. To accomplish this, we analyzed data from an 8-week observational analysis in perimenopausal women with mild to moderate depressive symptoms in which weekly measures of female reproductive hormones (E2, FSH, and progesterone), subjective sleep continuity parameters (e.g., nightly awakenings, WASO, SOL) and nVMS were obtained. We hypothesized that disruption in sleep continuity would occur independent of nVMS and depressive symptom severity.

Methods

Study Overview

Data were derived from a study on female reproductive hormones and depressive symptoms in 50 perimenopausal women (24). Unmedicated women with mild to moderate depressive symptoms were enrolled in an 8-week observational study involving 9 weekly assessments (see Fig. 1) during which sleep and VMS diaries were completed continuously and serum reproductive hormones (estradiol, FSH, progesterone), and depressive symptoms were assessed weekly. All subjects provided written informed consent for study procedures, which were approved by the Mass General Brigham Human Research Committee (Institutional Review Board) for data collected at Massachusetts General Hospital and Brigham and Women’s Hospital.

Figure 1.

Study overview showing data sampling collected over the 9 weekly timepoints for the 45 perimenopausal women during the 8-week study period. At each visit, serum levels of estradiol (E2), progesterone (P4), and follicle stimulating hormone (FSH) were measured, severity of sleep interruption and vasomotor symptoms (VMS) were assessed using daily diaries, and depressive symptoms were assessed using a clinician-rated mood scale for the preceding 7 days.

Participants

Detailed description of the inclusion and exclusion criteria have been previously published (24). Briefly, participants for the study were 50 women who were: 1) between the ages of 35 and 56 years; 2) perimenopausal per Stages of Reproductive Aging Workshop (STRAW) criteria (25) (menstrual cycle length varying > 7 days in consecutive cycles or > 60 days without menses but amenorrhea < 12 months in the prior 12 months reflecting early or late menopause transition, respectively); 3) not using centrally active medications or systemic hormones; and 4) absent of any primary sleep and/or psychiatric disorder other than unipolar depression (25). Mild to moderate depressive symptoms were defined in this study as a score from 10 to 25 on the clinician-rated Montgomery-Åsberg Depression Rating Scale (MADRS) (24). For the current analysis, 5 participants were excluded for missing sleep diary data (< 6 of 8 weekly diaries with < 4 of 7 nights per week of data), resulting in 45 participants (90% of the sample) being included in the analysis.

Sleep Assessments

Daily subjective sleep diaries were completed for the duration of the 8-week study period (see Fig. 1). Participants were instructed to complete diaries each morning, answering questions related to their sleep patterns and continuity from the previous night (26). Questions captured when participants reported getting into and out of bed, the time it took to fall asleep, number of nightly awakenings, total minutes spent awake throughout the night, and their time of final awakening. Answers to these questions were then utilized to calculate the primary sleep continuity parameters each week: number of awakenings per night, WASO, and SOL. Missing data were imputed for diaries providing only 4 to 6 nights of data within a given week, using the average of available data for that individual woman during that week.

Hormone Assays

Serum levels of estradiol were determined with liquid chromatography-mass spectrometry (LC-MS) (Mayo Clinic, Rochester, NY), with 10 pg/mL as the lower limit of quantitation (27, 28). Visits with estradiol levels below the detectable limit (n = 81, 20.8%) were assigned a value of 9 pg/mL, below the limit of detection. Serum progesterone and FSH were measured by chemiluminescence immunoassays (progesterone by Abbott Architect ci8200 and Beckman Coulter, Fullerton, CA; FSH by Abbott Architect ci8200). The lower analytical sensitivities of these assays were 0.08 to 0.1 ng/mL for progesterone, and 0.05 IU/L for FSH, respectively. All 3 interassay coefficients of variations (CV) were estimated ≤ 10%.

Vasomotor Symptoms

VMS were assessed daily to calculate a weekly average for the 8-week study period using a self-report VMS diary as previously described (24). VMS diaries were completed each morning to capture the number of VMS occurring between lights-out and lights-on (nVMS) and again at the end of the day to capture separately the number of VMS occurring during the daytime.

Depressive Symptom Severity

Depressive symptom severity was measured weekly using the clinician-rated 10-item MADRS (range 0-60, higher score worse) (29). This questionnaire has good internal consistency, test-retest reliability, and validity (30).

Statistical Analyses

Population characteristics are presented as means ± SD, medians (interquartile range [IQR]) or frequencies with percentages, depending on the distribution and type of variable.

To determine the associations of reproductive hormones with sleep outcomes (number of awakenings, WASO, SOL), data were analyzed using General Estimating Equation (GEE) models with robust (sandwich estimator) errors (31). In order to compare effect sizes across female reproductive hormones, serum levels of estradiol, progesterone, and FSH were standardized prior to data analysis. Given the repeated nature of the data, we used GEE models to examine 2 groups of associations. First, we calculated the association of serum absolute levels of female reproductive hormones (predictor) with each sleep continuity parameter (outcome). To obtain one measure for each sleep parameter, the daily sleep diary measures were averaged across each particular week. Therefore, hormone measurements at the start of each week were linked to the sleep measure from the entire corresponding week (see Fig. 1). Second, we examined the association of the woman-specific coefficient of variation (CV) for each female reproductive hormone as a measure of within-woman variability over the 8-week study period (predictor), with the same sleep continuity parameters (outcome) as used in the first groups of associations. The woman-specific coefficient of variation was calculated by dividing the within-subject SD of hormone level_{1 to 9} by the within-subject mean hormone level_{visit 1 to 9} multiplied by 100%. Additional analysis involved dichotomization of estradiol in the postmenopausal range, based on the lower limit for the premenopausal range for this assay (15 pg/mL) (32). A sensitivity analysis was performed using the cutoff of 10 pg/mL.

GEE models with different covariance structures were fitted for each sleep continuity parameter and the optimal covariance structure was chosen based on the Quasi-Akaiki Information Criterion for model selection (33). Sleep continuity parameters, including WASO and SOL, followed a negative binomial distribution and were modeled accordingly. In contrast, number of nightly awakenings was modeled with Poisson distribution. GEE models were run in 3 stages. First, by minimally adjusting in order to account for variability in both the number of days between visits (mean = 7.3 days, SD = 4.1 days) and time spent in bed (median = 486 minutes, IQR = 445-535 minutes). Adjusting for the interval between visits is important because hormone parameters may depend on cycle patterns. Adjusting for time spent in bed is important because the opportunity for sleep and therefore the presentation of sleep discontinuity are influenced by the amount of time spent in bed. Second by further adjustment for time-variant depressive symptom severity (MADRS score) and nVMS frequency and, lastly, by additional adjustment for time-invariant baseline sociodemographics including age (years), race (Caucasian or not), obtainment of college degree (yes/no), and body mass index (BMI, kg/m²). Data were analyzed using STATA 15 (College Station, Texas). An alpha of < 0.05 was considered statistically significant.

Results

Participant Characteristics

Baseline characteristics of the 45 subjects included in the analysis are depicted in Table 1. Briefly, the mean age of study participants was 48.8 ± 4.0 years with a mean BMI of 27.3 ± 6.7 kg/m². The majority were either Caucasian (57.8%) or African American (37.8%). At baseline, the median (IQR) number of awakenings, WASO, and SOL reported per night were 1.5 (0.9-2.5), 24.3 (8.6-47.1) minutes, and 20.0 (10.0-33.6) minutes, respectively, reflecting a mild level of disruption in sleep continuity. Across all study visits, 88.9% of the sample reported the occurrence of nVMS at least once and 44.4% had at least one ovulatory cycle indicated by a distinct progesterone peak > 6 ng/dL. At study entry, women were mildly depressed (median [IQR] MADRS score 9.8 [7.4-13.9]), with one-fifth meeting criteria for a current episode of major depression and one-third reporting have received a diagnosed of depression over their lifespan prior to study enrollment. The median levels of serum estradiol, progesterone, and FSH at baseline were 87.8 (IQR 30.3-122.1) pg/mL, 0.2 (IQR 0.1-1.0) ng/mL, and 23.0 (IQR 7.5-52.5) IU/L, respectively. The median woman-specific CVs across study participation for serum estradiol, progesterone, and FSH were 75.2% (IQR 58.6%-93.7%), 114.8% (IQR 37.3%-155.0%), and 55.2% (IQR 27.9%-67.1%), respectively.

Table 1.

Characteristics of 45 perimenopausal women at baseline and across the 8-week study period

	Perimenopausal women (n = 45)
Baseline characteristics
Age, mean ± SD, y	48.8 ± 4.0
Caucasian, no. (%)	26 (57.8%)
College graduate, no. (%)	20 (44.4%)
Body mass index, kg/m2, mean ± SD	27.3 ± 6.7
Nocturnal vasomotor symptoms, median (IQR)	0.7 (0.05 – 1.75)
Severity of depressive symptoms (MADRS), median (IQR)	9.8 (7.4 – 13.9)
No. participants with current major depressive episode (MDE), n (%)	9 (20.0%)
No. participants with lifetime history of MDD, n (%)	15 (33.3%)
Number of ovulatory cycles, no. (%)
No ovulatory cycles	25 (55.6%)
1-2 ovulatory cycles  3 + ovulatory cycles	14 (31.1%) 6 (13.3%)
Predictors
Estradiol (pg/mL), median (IQR)	87.8 (30.3 – 122.1)
Estradiol in postmenopausal range, no. (%)a	118 (30.3%)
FSH (IU/L), median (IQR)	23.0 (7.5 – 52.5)
Progesterone (ng/mL), median (IQR)	0.2 (0.1 – 1.0)
CV estradiol, median (IQR)	75.2 (58.6 – 93.7)
CV progesterone, median (IQR)	114.8 (37.3 – 155.0)
CV FSH, median (IQR)	55.2 (27.9 – 67.1)
Outcomes
Awakenings (IQR)	1.5 (0.9 – 2.5)
Wake-time after sleep onset (min), median (IQR)	24.3 (8.6 – 47.1)
Sleep-onset latency (min), median (IQR)	20.0 (10.0 – 33.6)
Time spent in bed (min), median (IQR)	485.8 (445.4 - 535.0)
Total sleep time (min), median (IQR)	406.5 (367.1 - 448.6)

Estradiol in the postmenopausal range dichotomized at < 15 pg/mL. Ovulatory cycles were defined by a distinct progesterone level > 6 ng/mL. All variables were assessed at every visit, except for age, race, and educational status, which were determined at study entry.

Abbreviations: CV, coefficient of variation; FSH, follicle stimulating hormone; IQR, interquartile range; MADRS, Montgomery-Åsberg Depression Rating Scale.

Associations of Sleep Continuity Parameters With nVMS and Depressive Symptom Severity

We first tested associations of each sleep continuity parameter with nVMS and severity of depressive symptoms. Number of nightly awakenings associated with nVMS in minimally and fully adjusted models (data not shown: beta [β] = 0.14 with 95% CI [0.02 to 0.26], P = 0.02, and β = 0.14 [95% CI] [0.02 to 0.26], P = 0.03, respectively). In contrast, there was no association of WASO or SOL with nVMS, nor of any sleep continuity parameter with depressive symptom severity (data not shown, all P > 0.05).

Associations of Sleep Continuity Parameters With Female Reproductive Hormone Levels

Associations of each sleep continuity parameter with levels of each female reproductive hormone are shown in Fig. 2 and Tables 2 and 3. Modeled as a continuous measure, lower levels of estradiol were associated with a greater number of nightly awakenings (minimally adjusted standardized β = −0.07; 95% CI, −0.13 to −0.02, P = 0.01), as were estradiol concentrations in the postmenopausal range (β = 0.21; 95% CI, 0.11 to 0.32; P < 0.001, Fig. 2A and 2B and Table 2). Consistent with the estradiol findings, higher levels of FSH associated with a greater number of nightly awakenings (β = 0.17; 95% CI, 0.06 to 0.28; P = 0.002, Fig. 2C and Table 2). Estradiol in the postmenopausal range and higher FSH levels (β = 0.14; 95% CI, 0.04 to 0.24; P = 0.007; and β = 0.12; 95% CI, 0.02 to 0.22; P = 0.02, respectively) remained significantly associated with more nightly awakenings in fully adjusted models, while the association with lower levels of estradiol as a continuous measure became marginally significant (β = −0.05; 95% CI, −0.11 to 0.01; P = 0.07, Table 2). Thus, having estradiol levels in the postmenopausal range was associated with 0.18 more awakenings per night on average relative to those with higher E2 levels, and a 1-SD increase in FSH (37.2 IU/L) was associated with an average increase of 0.16 awakenings per night after accounting for all other contributing factors using the fully adjusted models. In contrast, progesterone levels were not associated with nightly awakenings in minimally (β = −0.03; 95% CI, −0.06 to 0.01; P = 0.11) or fully adjusted (β = −0.02; 95% CI, −0.06 to 0.01; P = 0.20) models (Fig. 2D and Table 2). There was no association of female reproductive hormone levels with WASO or SOL (Tables 2 and 3, all P > 0.05). Results were unchanged after substituting daytime for nighttime VMS and after excluding 1 subject who underwent menopause before age 40 years.

Figure 2.

Associations of serum reproductive hormone levels with the mean number of awakenings per night on a daily sleep diary across the 8-week study period. Figures show the mean number of awakenings per night in relation to A) serum estradiol levels, B) dichotomized by estradiol in the hypo-estrogenic postmenopausal range vs the estrogenized range, C) serum follicle stimulating hormone (FSH) level, and D) serum progesterone, using adjusted beta-coefficients. Models were adjusted for the number of nocturnal vasomotor symptoms (VMS), depressive symptom severity, age, race, obtainment of graduate degree, body mass index, and number of days between study visits.

Table 2.

Associations of serum levels of estradiol (modeled continuously and categorically dichotomized at the postmenopausal range), FSH, and progesterone in relation to number of awakenings and WASO on daily sleep diary across the 8-week study period in 45 perimenopausal women

	# Awakenings		WASO (min.)
	B (95% CI)	P	B (95% CI)	P
Estradiol
Minimally adjusteda	-0.07 (-0.13, -0.02)	.01	-0.01 (-0.09, 0.07)	.86
Moderately adjustedb	-0.06 (-0.11, -0.001)	.04	0.0002 (-0.08, 0.08)	1.00
Fully adjustedc	-0.05 (-0.11, 0.01)	.07	0.03 (-0.08, 0.14)	.58
Estradiol in postmenopausal range
Minimally adjusteda	0.21 (0.11, 0.32)	< 0.001	0.02 (-0.16, 0.19)	.85
Moderately adjustedb	0.16 (0.05, 0.26)	.003	0.003 (-0.17, 0.18)	.97
Fully adjustedc	0.14 (0.04, 0.24)	.007	-0.04 (-0.21, 0.13)	.63
FSH
Minimally adjusteda	0.17 (0.06, 0.28)	.002	0.12 (-0.04, 0.27)	.14
Moderately adjustedb	0.12 (0.02, 0.22)	.02	0.11 (-0.04, 0.25)	.15
Fully adjustedc	0.12 (0.02, 0.22)	.02	0.06 (-0.10, 0.21)	.46
Progesterone
Minimally adjusteda	-0.03 (-0.06, 0.01)	.11	-0.02 (-0.08, 0.04)	.62
Moderately adjustedb	-0.02 (-0.05, 0.01)	.23	-0.02 (-0.08, 0.05)	.66
Fully adjustedc	-0.02 (-0.06, 0.01)	.20	-0.04 (-0.17, 0.08)	.50

Estradiol in the postmenopausal range dichotomized at < 15 pg/mL. Estimates are standardized B-coefficients determined by generalized estimating equation analyses.

Abbreviations: FSH, follicle stimulating hormone; WASO, wake-time after sleep onset.

^aAdjusted for number of days between study visits and time in bed.

^bAdjusted for number of days between study visits, time in bed, depressive symptom severity, and nocturnal vasomotor symptoms.

^cAdjusted for number of days between study visits, time in bed, depressive symptom severity, nocturnal vasomotor symptoms, age, race, BMI, and education.

Table 3.

Associations of serum estradiol, FSH and progesterone with sleep-onset latency on daily sleep diary across the 8-week study period in 45 perimenopausal women

	Sleep-onset latency (min.)
	B (95% CI)	P
Estradiol
Minimally adjusteda	0.02 (-0.04, 0.06)	.67
Moderately adjustedb	0.01 (-0.04, 0.07)	.64
Fully adjustedc	0.03 (-0.04, 0.10)	.41
Estradiol in postmenopausal range
Minimally adjusteda	0.09 (-0.07, 0.25)	.27
Moderately adjustedb	0.09 (-0.09, 0.28)	.33
Fully adjustedc	-0.001 (-0.21, 0.21)	.99
FSH
Minimally adjusteda	0.06 (-0.05, 0.17)	.31
Moderately adjustedb	0.05 (-0.06, 0.17)	.36
Fully adjustedc	0.02 (-0.09, 0.12)	.76
Progesterone
Minimally adjusteda	-0.02 (-0.07, 0.02)	.26
Moderately adjustedb	-0.02 (-0.07, 0.02)	.29
Fully adjustedc	-0.04 (-0.08, 0.01)	.09

Estimates are standardized B-coefficients determined by generalized estimating equation analyses. Estradiol in the postmenopausal range dichotomized at < 15 pg/mL. Abbreviation: FSH, follicle stimulating hormone.

^aAdjusted for number of days between study visits, and time in bed.

^bAdjusted for number of days between study visits, time in bed, depressive symptom severity, and nocturnal vasomotor symptoms.

^cAdjusted for number of days between study visits, time in bed, depressive symptom severity, nocturnal vasomotor symptoms, age, race, BMI, and education.

Associations of Sleep Continuity Outcomes With Variability of Female Reproductive Hormone Levels

Associations between variability of reproductive hormones, each integrated across the study as a coefficient of variance, with each sleep continuity parameter are shown in Supplemental Table 1 (34). Overall, there were no statistically significant associations, after adjusting for relevant covariates.

Discussion

Disruption of sleep continuity characterized by repeated awakenings that accumulate WASO time during the night is highly prevalent during the perimenopause, most commonly driven by nVMS. Using an 8-week repeated-measures approach with daily sleep diary monitoring and repeated weekly assays, our results show that the changing hormone dynamics of the perimenopause underlie awakenings independent of nVMS. Despite nVMS being reported almost universally by participants in our study, we observed that more nightly awakenings were associated with lower estradiol levels, especially when measured categorically by a threshold indicating the postmenopausal range, as well as with higher FSH levels, but not with progesterone. These results remained significant after adjusting for nVMS and other relevant covariates linked to sleep discontinuity, including DepSx, BMI, and age. Levels of female reproductive hormones were not associated with WASO or SOL, nor were there significant associations of variability in female reproductive hormone levels with sleep continuity parameters. Our findings therefore provide empiric support that reproductive hormone patterns observed in perimenopausal women—lower estradiol levels and higher FSH levels—are linked with more nightly awakenings and potentially provide a rationale for why midlife women complain that their sleep is disrupted even when they are not experiencing nighttime VMS.

Our findings that middle-of-the-night awakenings are explained by changes in reproductive hormones contribute importantly to the field given the challenge of discriminating the emergence of nVMS from the trajectory toward lower estradiol and higher FSH during the perimenopause. Our results are consistent with a physiologic study using polysomnographic (PSG) EEG monitoring to quantify sleep (15) and with 2 longitudinal cohort studies involving self-reported sleep (6-8). As we observed with diary reports of sleep continuity, more objectively defined awakenings were associated with higher levels of FSH but not with estradiol levels in a single PSG during the follicular phase (15). Similarly, the Study of Women’s Health Across the Nation (SWAN) (6) showed that reports of waking up repeatedly were associated with a decrease in serum estradiol and an increase in FSH over time when measured annually for 7 years, independent of VMS (6). Similar findings were observed in the Seattle Women’s Midlife Study (8) using urinary FSH and E1G as a proxy of estrone measured 8 to 12 times per year for 4 years and then quarterly for 4 years thereafter (8). While these findings are consistent, our approach with concurrent and repeated daily sleep assessments and weekly assays portray a richer real-time picture to explain these relationships than do a single PSG study or a monthly to annual survey assessment.

Converging lines of evidence from basic and clinical research support a role for gonadal steroids as modulators of sleep, although the exact neurobiological mechanisms remain unknown. Estrogen signaling is most impactful on sleep regulation (35). Estrogen receptors are found in various neural centers regulating sleep including in the basal forebrain, hypothalamus, dorsal raphe nucleus, and locus ceruleus (17-19) and in the sleep-active ventrolateral preoptic area. Estrogen administration alters neuronal activity in these sleep-regulating centers (17-21). In nocturnal rodents, E2 administration promotes wakefulness (22, 23, 35), contrary to the observed inverse relationship in humans between endogenous E2 levels and nocturnal awakenings during sleep in our study. Indeed, the clinical translation of our findings suggests that exogenous E2 administration may improve sleep continuity independent of VMS suppression.

Importantly, the increased wakefulness in nocturnal rodents is mainly observed when E2 is administered during the dark phase of the animal’s activity cycle, which is their typical time for wakefulness, whereas E2 administration during the light phase when nocturnal animals are asleep does not increase wakefulness. Therefore, in humans and other diurnal animals, paralleling the “time-of-day” or circadian effect (35) of E2 administration during the dark period, estradiol during sleep may reinforce sleep consolidation by reducing awakenings during the night. Supporting this time-of-day hypothesis, estrogen receptor expression in the central circadian pacemaker and centers projecting to the circadian pacemaker has been reported (36, 37), consistent with the improved sleep quality observed in trials testing the impact of hormone therapy in sleep in menopausal women (38). Another mechanism that may explain our findings may involve changes in the levels of the hypothalamic neuropeptide orexin-A, which promotes wakefulness, and changes across menopause. Estradiol influences orexin receptor 1 expression in the hypothalamus and anterior pituitary (39). Plasma orexin levels are 3 times higher in postmenopausal than premenopausal women, whereas postmenopausal women taking estrogen therapy have levels similar to premenopausal women (40). Taken together, these findings implicate several putative pathways through which estradiol may influence sleep during the perimenopausal transition to explain our findings.

While lower E2 levels were strongly linked with more awakenings, our study did not find a relationship between the number of awakenings and E2 variability across the study interval. Results of the minimally adjusted model indicate no significant association, but the association becomes statistically significant upon further covariate adjustment (Suppl. Table 1) (34). This unveiling of a significant association appears to be explained by an increase in precision (narrowing of the confidence interval) rather than a change in the magnitude of the association, as the beta coefficient changes minimally. We therefore conclude that this is not a meaningful association.

We did not observe an association of estradiol, FSH, or progesterone levels with other parameters of sleep discontinuity, namely WASO and SOL. For SOL, these findings are at odds with the SWAN study linking trouble falling asleep to lower serum estradiol levels (6), but consistent with the Seattle cohort showing no association with the urinary metabolite of estrone (8). To date, consistent with our findings, no study has observed a relationship of SOL to FSH levels (6-8). Of note, SWAN symptom data were captured annually as an ordinal indicator of how many nights per week a woman had trouble falling asleep, whereas our SOL data were quantified as a mean of the number of minutes of SOL occurring each week. Nonetheless, because middle-of-the-night awakening, rather than SOL, is the primary manifestation of sleep discontinuity in this population, it is not surprising that the changing hormone dynamics of the perimenopause are less strongly linked.

While associations with the number of awakenings were strong, our data did not reveal that reproductive hormone levels were related to WASO, the cumulative number of minutes spent awake after sleep initiation. This is in contrast to PSG studies that observed higher FSH levels associating with more WASO time, but no association with estradiol levels (15). PSG-defined WASO may provide a more robust quantification of this parameter than the diary reporting in our study. It is also plausible that, in contrast to wake initiation, WASO is driven more by other factors that maintain wakefulness, such as the sleeping environment, recovery from an nVMS event, and distress and psychological activation during the night. Indeed, in other studies, we and others have observed that nVMS (7) and stress exposures (41) contribute to more time spent awake in the middle of the night. These observations discriminating predictors of the number of awakenings vs WASO time highlight the difference between these 2 sleep continuity parameters.

Our study did not find a relationship between sleep discontinuity and mean progesterone levels nor variability of progesterone levels reflecting intermittent ovulation, either in the direction of worse or better sleep continuity. Exogenous administration of natural progesterone without exogenous estradiol improves sleep in postmenopausal (42, 43) women, and restores provoked middle-of-the-night sleep disruption (44). To date, there has been limited investigation of endogenous progesterone levels and sleep, although one study observed a counter-intuitive association between higher urinary levels of the progesterone metabolite pregnanediol glucuronide (PdG) and greater likelihood of sleep complaints in perimenopausal women (13). The role of endogenous progesterone on sleep in premenopausal women across the menstrual cycle appears to be mixed with reports of both high and low progesterone being associated with sleep disruption and poor sleep quality (45, 46). Progesterone and its metabolite allopregnanolone facilitate sleep by antagonizing the y-aminobutyric acid A receptor. In postmenopausal women, administration of natural progesterone reduces PSG-assessed WASO compared with placebo (46-48) with a smaller effect on sleep-onset latency. Integrating our findings with previous studies, progesterone therapy has a beneficial effect on menopause-related sleep disturbance, but associations with endogenous progesterone appear to be null or in the opposite direction.

Importantly, our observed associations of more nighttime awakenings with lower estradiol and higher FSH levels were robust after accounting for depressive symptoms. This finding is notable given that our study population had mild to moderate levels of depression symptoms and that depressive symptoms are strong predictors of disrupted sleep continuity in midlife women (8, 49). The bi-directional relationship between mood and sleep continuity is complex, with sleep disturbance conferring risk for depression, depression conferring risk for sleep disturbance, and sleep continuity disruption manifesting commonly as a symptom of depression. Of note, in the parent study for the current analysis (n = 50) (24), we observed that greater variability in estradiol, rather than absolute levels, was associated with severity of depressive symptoms. This is in contrast with our findings for awakenings, in which lower levels of estradiol, especially in the postmenopausal range, link with more awakenings, but there was no association with estradiol variability. Despite how commonly perimenopause-related symptoms co-occur, this discrimination of different endocrine patterns associating with specific symptoms has also been observed for irritability vs depressive symptoms (50) and for VMS vs negative mood (51), suggesting that discrepant hormone dynamics underlie each symptom manifestation. Thus, together with evidence that the perimenopause encompasses a variety of hormone profiles interspersed over time (52), these data provide empiric validation for the common clinical scenario that women experience symptoms which manifest and then resolve as they traverse the perimenopause.

Our study has several important strengths. First, we captured weekly measures of self-reported sleep continuity parameters and nVMS in parallel with weekly assessments of female reproductive hormones for 8 consecutive weeks. This yielded a robust dataset, which enables examination of associations of sleep continuity parameters with both absolute levels and variability of female reproductive hormone levels. This design and analytic approach are novel, as prior studies have not been able to capture the dynamic changes in female reproductive hormones due to infrequent monthly or annual (6, 8, 13) or, alternatively, single time-point measures (15). Second, we assayed estradiol using the gold standard of LC/MS rather than immunoassays in the low estradiol range or via urinary metabolites as a proxy of hormone levels (8, 13), improving reliability of the assays. Lastly, we confirmed that participants were free of sleep disorders and validated the presence of depressive symptom severity utilizing the clinician-rated MADRS, which contrasts with prior work that has either excluded measures of depression or assessed its presence by self-report (6, 8, 13). While depression symptoms among our participants may be considered a limitation, such mood disturbance commonly co-occurs with VMS and sleep disturbance in perimenopausal women (53). Thus, by assessing and adjusting for depressive symptoms as a potential confounding variable in addition to nVMS, we greatly enhance the generalizability of our findings.

Our study also has several notable limitations. The observational study design precludes causal inference. Additionally, the observational design of the study precludes controlling for sleep history or behavioral choices (e.g., diet, exercise, work, and social schedules) that may have influenced sleep homeostasis and circadian phase, both of which are sleep-regulatory neurobiological processes influenced by the female reproductive hormones, introducing additional variability in our observations. A second limitation is the use of self-reported sleep diaries rather than the gold standard of PSG. However, the use of PSG has constrained previous studies to a single-night recording in parallel with a single serum assay (15), thereby precluding characterization of endocrine patterns, which are dynamic and variable during the perimenopause (10-12).

Conclusion

Our finding that more nightly awakenings were associated with lower estradiol and higher FSH levels independent of nighttime VMS and depressive symptoms provides empiric support for a perimenopause-associated sleep discontinuity condition that is linked with female reproductive hormone changes. While nighttime VMS remains the primary driver of disrupted sleep continuity in this population, our data suggest that the changing hormone dynamics are also determinants that might explain problematic sleep in perimenopausal women without nighttime VMS (2). These results substantiate important validation of the perimenopausal experience in women whose sleep discontinuity cannot be explained by nighttime VMS or depressive symptoms and suggest that exogenous E2 administration might have therapeutic effects unrelated to any beneficial effect on VMS.

Glossary

Abbreviations

BMI

body mass index

CV

coefficient of variation

E2

estradiol

FSH

follicle-stimulating hormone

GEE

General Estimating Equation

IQR

interquartile range

MADRS

Montgomery-Åsberg Depression Rating Scale

nVMS

nocturnal vasomotor symptoms

PSG

polysomnography

SOL

sleep-onset latency

VMS

vasomotor symptoms

WASO

wakefulness after sleep onset

Funding Source

Supported by National Institute of Mental Health 5R01MH082922 (HJ), 5R01AG048209 (HJ), and Mary Tynan Fellowship (J.C.).

Disclosure Summary

M.N., A.W., L.F., S.C., and A.D.W. have nothing to declare. J.C. is now an employee of Novo Nordisk Inc. H.J. is a consultant for Bayer, Jazz, and Eisai, and has grant/research support from NIH, Merck, and Pfizer. Her spouse is an employee at Arsenal Biosciences and has equity in Merck Research Lab. S.A.R. holds patents for (1) Prevention of Circadian Rhythm Disruption by Using Optical Filters, and (2) Improving sleep performance in subject exposed to light at night; S.A.R. owns equity in Melcort Inc.; has provided paid consulting services to Sultan & Knight Limited, Bambu Vault LLC, Lucidity Lighting Inc.; and has received honoraria as an invited speaker and travel funds from Starry Skies Lake Superior, University of Minnesota Medical School, PennWell Corp., and Seoul Semiconductor Co. Ltd. S.A.R. has received grant/research support from Seoul Semiconductor Co. Ltd., Biological Innovation and Optimization Systems, LLC, Merck & Co., Inc., Pfizer Inc., Vanda Pharmaceuticals Inc., Lighting Science Group, NIH, and NASA. These interests were reviewed and managed by Brigham and Women’s Hospital and Mass General Brigham in accordance with their conflict of interest policies.

Data Availability

Some or all datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.