Sleep Disturbance During the Menopausal Transition in a Multi-Ethnic Community Sample of Women

Study Design and Participants

SWAN is a community-based, multisite cohort study of the menopausal transition. A total of 3,302 women were enrolled at 7 sites: Boston, MA, Chicago, IL, Detroit area, MI, Los Angeles and Oakland, CA, Newark, NJ, and Pittsburgh, PA. The study design and cohort recruitment have been described in detail previously.²⁵ Recruitment strategies varied due to differences in resources in each of the seven communities. Each site recruited Caucasian women and a minority group sample. Women were of Caucasian, African American (Boston, Chicago, Detroit, Pittsburgh), Chinese (Oakland), Japanese (Los Angeles), and Hispanic (Newark) origins. SWAN cohort eligibility included the following: age 42–52 years, premenopausal or early perimenopausal, intact uterus and ≥1 ovary, ≥1 menstrual period in the previous 3 months, no sex steroid hormone use in the previous 3 months, and not pregnant. Each site's institutional review board approved the study, and all women gave written informed consent to participate.

Procedures and Measures

From the study cohort of 3,302 participants, 257 women were excluded for having no follow-up data or missing all sleep measures at baseline, leaving a total of 3,045 women. Baseline and annual visit data through the first 7 annual assessments were used. Except for participants at the New Jersey site, for whom no data were available for year 7, each woman could contribute up to 8 observations. The average number of observations was 6.7 (6.9 when excluding the New Jersey site). At the seventh annual visit, the overall SWAN retention rate (including deceased participants) was 72.8% (80.8% when excluding the New Jersey site). Also, at the sixth assessment only 54 observations were available from New Jersey participants. Because these data did not differ statistically from those for the 378 New Jersey participants who dropped out before year 6 in terms of ethnicity, marital status, education, age, body mass index (BMI), difficulty paying for basics, smoking status, or for the menopausal transition variables or all sleep measures at baseline or the fifth annual assessment, these 54 observations were included in the analyses. Data from all observations are reported because the conclusions were the same after removing the New Jersey data. Most model estimates changed less than 10% and only estimates for late perimenopausal women in trouble falling asleep and waking up early changed around 15%. The P-value for the pairwise comparison of late perimenopausal versus premenopausal became more significant (P < 0.001).

Dependent Variables

Trouble falling asleep (sleep initiation difficulty), waking up several times a night (sleep maintenance difficulty), and waking up earlier than planned and unable to fall asleep again (early morning awakening) were self-reported at baseline and at each annual assessment. Participants were asked about the frequency of sleep problems (“your sleep habits”) in the past 2 weeks: (1) none, (2) less than once a week, (3) one or two times a week, (4) three or four times a week, (5) five or more times a week. These 3 sleep measures were dichotomized as “no” (<3 times a week) or “yes” (≥3 times a week), which is consistent with the frequency criterion most commonly used in insomnia research.^26,27

Independent Variables

Covariates

Participants provided extensive health, psychosocial, lifestyle and biologic data at baseline and at annual assessments. Variables selected have been related to trouble sleeping or mediated the association between reproductive hormones and trouble sleeping in previous research.^20,28

Data Analysis

The analytic sample included a total of 20,407 observations from 3,045 women, with women contributing 2–8 data points from baseline through the seventh annual assessment. Baseline summary statistics across race/ethnicity for the dependent (sleep measures) and independent (menopausal transition-related) variables and covariates were computed by using a chi-square analysis or one-way analysis of variance approach. The dependent variables consisted of 3 individual sleep symptoms: trouble falling asleep, waking up several times a night, and early morning awakening. These symptoms were analyzed separately so that we could examine relationships between menopausal transition characteristics and specific types of sleep continuity disturbances.

Longitudinal random effects logistic regression models^39,40 were used to evaluate the association between each of the 3 self-reported sleep difficulties and menopausal transition status (approximated by bleeding patterns, frequency of vasomotor symptoms, serum reproductive hormones E2 and FSH) over time. This approach allowed women to contribute different numbers of observations, thus fully using all of the available data. In modeling, both E2 and FSH were log-transformed to reduce the skewness, and the log-transformed hormone values at each visit were separated into 2 components: baseline and change over follow-up after baseline for distinguishing between the cross-sectional effect and the longitudinal effect.³⁹

We first estimated the age-adjusted association by including only concurrent age as a covariate in each model. Then, we constructed multivariable models by adjusting for additional potential confounders, which were identified from the literature, as described above. After checking for collinearity, covariate variables that were suggestive of a possible bivariate association (i.e., P < 0.15) were included in the initial multivariable models. Backward elimination was used to identify suitable multivariable models, with selection criteria set at P < 0.05 for those covariates remaining in the final models. Ethnicity, site, age, and current HT use were forced into all the final multivariable models because of their importance as potential confounders and so that the models could be compared; also, the blood draw in window variable was forced into the hormone models. Due to a specific SWAN study interest in ethnic differences, interactions of race/ethnicity with the independent variables were included to test whether demonstrated associations varied significantly among racial/ethnic groups. Finally, to assess the impact of HT use on models of vasomotor symptoms and hormones (note, the impact of HT use on models of menopausal status was evaluated by using an 8-level menopausal status/HT use variable), we first reran our multivariable analysis by omitting the subgroup (13.3% of the sample) that reported HT use since last study visit. The results (data not shown, all model estimates changed less than 7%) were very similar to results presented here that include HT users. We also tested the interaction terms between HT use and vasomotor symptoms as well as change in hormones since baseline. None of the interaction terms we tested was statistically significant, and therefore the results are not presented.

P-values, odds ratios (ORs), and 95% confidence intervals (CIs) for independent variables were computed and reported across sleep dependent variables. We used Stata Release 8 xtlogit procedure⁴¹ to estimate the longitudinal random effects logistic regression models, and the remaining analyses were conducted using SAS Version 8.⁴²

Participant Baseline Characteristics

Of our total sample, 47.5% were Caucasian, 28% were African American, 9% were Japanese, 8% were Chinese, and 7.5% were Hispanic (Table 1). Racial/ethnic groups differed significantly (P < 0.05) on all baseline characteristics except age and FSH. Using menstrual bleeding criteria, 52.6% of the sample were premenopausal and 45.2% were early perimenopausal at baseline; 2.3% (n = 69) were missing bleeding pattern regularity information. Most were married or living with a partner (65.9%; 1.5% missing) and had at least some college, (75.7%; 0.9% had no information).

At baseline, 30.8% of the cohort reported at least one of the 3 sleep difficulties (28.0% of premenopausal women and 34.3% of early perimenopausal women). Sleep maintenance, the most prevalent problem, was reported by 25.9% of the baseline cohort. Waking early (12.4%) and trouble falling asleep (10.1%) were notably less prevalent. Over the course of the transition, sleep difficulties increased with the exception that early morning awakening decreased somewhat during the transition from late perimenopause to postmenopause (Figure 1). The highest percentages of each of the 3 sleep problems were reported by those who were surgically menopausal.

Menopausal Transition State and Sleep Difficulties

Age-adjusted results, displayed in Table 2, revealed a progressive and statistically significant increment in the odds of trouble falling asleep as women transitioned to postmenopause over the 7 years. A similar trend was observed for waking up several times and for waking up early, except that the highest ORs were observed in late perimenopause, followed by a small decrement during the transition from late perimenopause to postmenopause.

After adjusting for all other covariates the general trends described above tended to persist and remained statistically significant, particularly for trouble falling asleep and waking up several times (Table 3). However, some of the sequential pairwise menopausal transition effects (not indicated in the table) were no longer statistically significant. Women transitioning from premenopause to early perimenopause, compared to women who remained premenopausal, had higher odds of reporting waking up several times (OR = 1.26; 95% CI = 1.08, 1.47; P = 0.003); women transitioning from early to late perimenopause had higher ORs of reporting both waking up several times (OR = 1.55; 95% CI = 1.30, 1.86; P < 0.001) and early morning awakening (OR = 1.45; 95% CI = 1.17, 1.79; P = 0.001); and women transitioning from late perimenopause to postmenopause had a lower OR of reporting early morning awakening (OR = 0.77; 95% CI = 0.61, 0.98; P = 0.03). Among the 8 menopausal status/HT use categories, women who were surgically menopausal without current HT use had the highest ORs for trouble falling asleep and waking up several times. In naturally postmenopausal women, compared to HT non-users, women who were HT users had lower ORs for trouble falling asleep (OR = 0.53; 95% CI = 0.34, 0.85; P = 0.007) and waking up several times (OR = 0.64; 95% CI = 0.48, 0.85; P = 0.002). Similarly, surgically postmenopausal women with HT use had lower ORs for trouble falling asleep than did those without HT use (OR = 0.51; 95% CI = 0.26, 0.99; P = 0.047). Because the “unknown” category includes a mix of pre-, early, and late perimenopausal HT users, we could not differentiate the HT effect for any of these 3 groups to permit interpretable “HT–no HT” comparisons within each group.

Vasomotor Symptoms and Sleep Difficulties

For all 3 types of sleep difficulties, a significant dose-response relationship was observed with VMS in the age-adjusted models (Table 2), which persisted, although of lesser magnitude (ORs for ≥ 6 days = 2.16–2.93) after adjusting for the additional covariates (Table 3). More frequent VMS were associated with a significantly higher OR of each sleep difficulty as compared with reports of either no VMS or 1–5 days of VMS. These effects of VMS (and menopausal status) were each independently related to sleep difficulties in multivariable models (Table 3).

E2 and FSH Levels

In modeling the hormone effects, the log-transformed E2 and FSH levels in the follow-up visits were split into baseline (cross-sectional) and change since baseline (longitudinal) values. Age-adjusted results (Table 2) were attenuated modestly in the multivariable-adjusted models, which also were adjusted for blood draw cycle day (Table 3). We observed negative associations between E2 levels and sleep problems and positive associations between FSH and sleep problems. Decreases in annual E2 levels were significantly associated with higher odds of trouble falling asleep and of waking up several times. Increases in annual FSH levels were significantly associated with higher odds of waking up several times.

Race/Ethnicity and Sleep Difficulties

Pairwise comparisons between race/ethnic groups were performed for all multivariable models. For trouble falling asleep, no differences were found between ethnic groups. For waking up several times, the estimates for Caucasian women were significantly higher than for all the other race/ethnic groups, and the estimates for Hispanic women were significantly lower than for all the other race/ethnic groups. For early morning awakening, the estimates for Caucasian women were significantly higher than those for Hispanic women in all models and were marginally significantly lower than those for Chinese women in menopausal status and VMS models. The estimates for Hispanic women were significantly lower than for African American women and for Chinese women in all models. The lower estimates for Japanese women compared with those for Chinese women were of only marginal statistical significance in all models.

In addition to the mean difference between race/ethnic groups, we also examined the potential interaction effect between race/ethnic groups and each of the menopausal transition-related variables in all multivariable models. Statistically significant interactions were found in 2 of 12 models. For waking up several times, we found variation in race/ethnic effects across menopausal status (χ² = 61.22, df = 28, P = 0.0003); however, because of the many cells (number of cells = 40), some of them with few participants, further interpretation is ambiguous. After removing HT users, surgically menopausal women, and New Jersey participants to reduce model fitting complexity, we observed no statistically significant interaction (χ² = 14.56, df = 9, P = 0.1038), indicating that the adjusted associations between sleep difficulties and menopausal status did not vary across race/ethnic groups. In the full sample, the VMS-by-race/ethnicity interaction for early morning awakening was significant (χ² = 24.52, df = 8, P = 0.0019). Within each racial/ethnic group, ORs for early morning awakening increased with increasing VMS frequency compared with no VMS, reaching statistical significance for all groups with VMS ≥ 6 days (P ≤ 0.001). Groups differed in their pattern and magnitude of increase. ORs were highest for Chinese (OR = 3.97) and Japanese (OR = 4.22) women reporting VMS ≥ 6 days, and Hispanic women had similar ORs for 1–5 days (OR = 2.87) and ≥ 6 days (OR = 2.96). Within each of the 3 VMS levels, comparisons between the Caucasian group and other race/ethnic groups revealed only two statistically significant differences – in those without VMS, Hispanic women were less likely to report early morning awakening (OR = 0.37, P = 0.002) and, for VMS ≥ 6 days, Chinese women were more likely (OR = 2.87, P = 0.001) to report early morning awakening. After removing HT users, surgically menopausal women, and New Jersey participants, the interaction remained statistically significant (χ² = 12.71, df = 6, P = 0.0479).