Results
Description of the Studies
Of the 1,478 articles identified, 24 studies with a total of 63,542 midlife women were reviewed and meta-analyzed. Among those women, 21,389 were classified as premenopausal, 17,219 were classified as perimenopausal, 22,439 were classified as postmenopausal, and 2,495 were classified to have had surgical menopause. All of the studies were published in English and in peer-reviewed journals.
As shown in Table 1 , adjusted effect sizes were reported in 10 studies comparing perimenopause with premenopause, and postmenopause with premenopause, taking into account confounders such as vasomotor symptoms, physical health, and lifestyle factors. Variations in the confounders included were noticeable. Although all study samples met the inclusion criterion of being recruited from the community, five studies were still found to be nonrepresentative of the community population. With regard to ethnicity, six studies were carried out with predominantly Asian women, seven studies were carried out with predominantly white samples, and four studies were carried out with predominantly Hispanic samples; the remaining studies included multiethnic backgrounds or women of other ethnic backgrounds (eg, Iranian).
A validated sleep measure was used in 7 studies, whereas the remaining 17 studies used nonvalidated sleep measures that included both single-item measures and multiple-item measures. Apart from the validity of the sleep measure, the recall time frame during which women reported sleep problems may also impact the estimation of the effect size, as the prevalence rate of an event increases as the recall time frame increases. In the studies included, women were asked to report their sleep experience from the previous night to the previous 12 months. To enable comparison, we adopted an arbitrary cutoff period of 1 month, as this is the standard period used in validated sleep questionnaires (eg, Pittsburgh Sleep Quality Index and Athens Insomnia Scale). Among 24 studies, 16 studies assessed the prevalence of sleep disturbance during a 1-month period, and 2 studies failed to specify a clear recall time frame.
Owing to the great difference in influence on sleep between natural and surgical postmenopause, the risk of sleep disturbance for these two groups, as contrasted with premenopause, should be assessed separately. Based on the reports of the included articles, 4 studies combined natural and surgical postmenopause and 14 studies performed separate analyses, whereas the combination could not be determined in 6 studies.
Risk of Bias in Included Studies
The key features of risk of bias in the 24 included studies are summarized in Figure 2, in addition to Table 2 . Although the included studies were assessed to have met the criteria for inclusion in this review, there were important methodological limitations in all studies, as discussed below. Agreement on assessment of risk of bias was reached by the authors.
(Enlarge Image)
Figure 2.
Methodological quality graph.
Calculation of Effect Size
The pooled ORs for each pair of menopausal groups are listed in Table 3 , and the corresponding forest plots are presented in Figures 3 to 7. With premenopause as the reference group, the ORs of sleep disturbance for perimenopause, postmenopause, and surgical menopause were all statistically significant and higher than 1, indicating an increased odds of experiencing sleep disturbance when women reach these stages. The ORs for perimenopause and postmenopause were of small magnitude, and the OR for surgical menopause was of moderate magnitude. Minimal change in pooled effect sizes was observed after removing from the final calculation studies that used a nonrepresentative sample of the community population, studies without a recall time frame for the measure of sleep disturbance, or studies that combined natural and surgical menopause. The percentages of change in pooled effect sizes from the total sample to a subset sample were less than 10%, with most below 5%. As a result, all of the studies were included in the final meta-analysis (percentages of change in effect sizes available upon request).
(Enlarge Image)
Figure 3.
Forest plot of pooled odds ratios (ORs) of sleep disturbance for perimenopause versus premenopause. ES, effect size.
(Enlarge Image)
Figure 4.
Forest plot of pooled odds ratios (ORs) of sleep disturbance for postmenopause versus premenopause. ES, effect size.
(Enlarge Image)
Figure 5.
Forest plot of pooled odds ratios (ORs) of sleep disturbance for surgical menopause versus premenopause. ES, effect size.
(Enlarge Image)
Figure 6.
Forest plot of pooled odds ratios (ORs) of sleep disturbance for postmenopause versus perimenopause. ES, effect size.
(Enlarge Image)
Figure 7.
Forest plot of pooled odds ratios (ORs) of sleep disturbance for surgical menopause versus postmenopause. ES, effect size.
Heterogeneity
The analysis showed the existence of heterogeneity among studies for all of the comparison groups. The degree of heterogeneity ranged from 66.9% to 87.9%, suggesting a substantial amount of inconsistency between studies.
Subgroup Analysis
Using a random-effects model, we conducted a subgroup analysis to explore potential sources of variation. Factors including adjustment for confounders, representativeness of the sample, ethnicity, sleep measure, recall time frame, and combination of natural and surgical postmenopause were analyzed as potential moderators. None of the identified moderators impacted on any of the effect sizes, except for ethnicity/culture ( Table 4 , Table 5 , Table 6 , Table 7 , and Table 8 ). The pairwise comparison of ethnic groups for each significant effect size is listed in Table 9 . For the ORs of sleep disturbance for perimenopause versus premenopause, the subgroup analysis by sleep measure did not fully explain the heterogeneity. It was interesting to see a still high level of inconsistency among studies that used a validated sleep measure, but not among studies that used nonvalidated measures. Likewise, when subgroup analysis was performed for the pooled ORs of sleep disturbance for postmenopause versus perimenopause, the level of inconsistency remained high in studies that used a nonrepresentative community sample and in those that used a recall time frame of 1 month or less. A high level of heterogeneity was again evident when we performed a subgroup analysis of the ORs for surgical menopause using natural postmenopause as the comparison group (explained in "Discussion").
Publication Bias
As revealed by Eggers' regression test,P values for intercepts varied from 0.174 to 0.568, suggesting no publication bias in any comparison group. Funnel plots were also constructed, and symmetry was visually inspected (available upon request). No apparent asymmetry was observed, except for the effect size representing the relationship between surgical menopause and sleep disturbance, possibly because of the smaller number of studies.