Sleep abnormalities in bipolar disorders across mood phases: A systematic review and meta-analysis

Summary

Sleep abnormalities are core features of bipolar disorders (BD), but they have not been thoroughly examined across mood phases. This meta-analysis investigated sleep disturbance prevalence and sleep characteristics differences in BD across mood phases. A systematic search through September 2024 identified 44 studies (7,614 BD cases, 3,164 controls), including 11 prevalence and 34 case-control studies. Poor sleep quality prevalence was 52% during euthymia, and insomnia prevalence was 63% during the depressive phase. Individuals with euthymic BD reported worse sleep quality and objectively measured longer total sleep time and sleep onset latency than controls. Depressive phase BD showed higher rapid eye movement percentages, while manic/mixed phase exhibited shorter total sleep time, lower sleep efficiency, and longer sleep onset latency. During euthymia, BD demonstrated greater variability in sleep duration and continuity, and more prominent sleep differences when assessed with sleep diary versus objective sleep measures, highlighting the importance of integrating objective assessments and patient-reported outcomes. Overall, these findings indicate that poor sleep quality and insomnia are highly prevalent in BD, and that some sleep parameter differences are present during euthymia, while others occur during depressive and manic phases, emphasizing the need for sleep assessments and tailored management throughout the course of BD.

1. Introduction

Bipolar disorders (BD) are characterized by episodes of hypomania or mania and depression, interspersed with periods of more stable, euthymic mood. Sleep disturbances have long been recognized to play a central role in BD [1–8], particularly during manic and depressive phases [9–11]. However, recent evidence indicates that sleep alterations are a hallmark of all phases of BD, not just acute mood episodes [12–15], and that they persist during euthymia, suggesting that sleep disruptions are not simply a reflection of mood dysregulation in BD [16–18]. Moreover, emerging research suggests that sleep pattern variations may serve as predictive markers for mood episode transitions in BD [19].

Poor sleep quality and insomnia are commonly observed in individuals with BD. The reported prevalence of sleep disturbances in these individuals varies across studies, ranging from 19% [20] to 95% [21], likely reflecting differences in assessment methods and definitions used (e.g., subjective sleep quality measures versus clinical insomnia criteria). Nonetheless, there are no meta-analyses assessing sleep disturbance prevalence in BD, including euthymia.

In BD, differences in sleep quality, duration, continuity, architecture, and EEG oscillation parameters can be assessed in case-control studies. Poorer subjective sleep quality has consistently been reported in BD relative to healthy control (HC) subjects [22, 23], suggesting that individuals with BD are aware of and complain about disrupted sleep. Differences relative to controls in sleep duration and sleep continuity parameters (i.e., total sleep time, sleep onset latency, sleep efficiency, wake after sleep onset, and their variability) have been reported both subjectively, using sleep diaries, and objectively, through actigraphy and polysomnography (PSG) [24–27]. A recent meta-analysis of actigraphy studies showed longer total sleep time (TST), sleep onset latency (SOL), and wake after sleep onset (WASO) along with lower sleep efficiency (SE) in remitted BD compared to healthy controls (HC) groups [28]. However, while some sleep studies in BD reported correlations between objective and subjective assessments of sleep parameters (particularly TST and SOL) [29, 30], other studies found differences between perceived and measured sleep, with individuals with BD typically underestimating TST while overestimating SOL [31, 32]. Notably, the severity of depressive symptoms, but not manic symptoms, was associated with greater discrepancy between objective and subjective sleep measures, with more depressed individuals underestimating their TST [29]. Combined, these findings suggest that the relationship between objective and subjective sleep measures in BD is more complex than previously assumed.

PSG and electroencephalographic (EEG) studies can also provide critical information regarding sleep architecture (e.g. proportion of time spent in different sleep stages) and sleep oscillations (e.g., sleep spindles and slow waves) [33]. Nonetheless, a case-control meta-analysis examining differences in these sleep characteristics in BD across mood phases is lacking.

Here, we performed a systematic review and meta-analysis of 1) sleep disturbance prevalence studies in individuals with BD in euthymic, depressed, and manic/mixed phases; and 2) case-control studies in participants with BD in euthymic, depressed, and manic/mixed phases relative to control groups. We included studies using various methodologies to measure sleep disturbances, ranging from subjective measures (sleep quality questionnaires, sleep diaries) to objective assessments (PSG, actigraphy) in different mood phases.

2. Methods

2.1. Search strategy and study selection

This meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) 2020 guidelines [34]. The protocol was registered in PROSPERO (CRD42023420519) [35]. We searched Web of Science and PubMed from inception to September 15^th, 2024 using the search string: (affective OR bipolar OR mania OR “mood disorder*”) AND (insomnia OR “sleep disturbances” OR “sleep problems” OR actigraph* OR actimet* OR polysomnogr* OR “sleep EEG” OR “sleep abnormalities” OR “sleep alterations” OR “sleep disorders” OR “sleep architecture” OR “sleep recordings” OR “sleep stages”). One author (MM) also manually searched the reference lists of included articles and any relevant review and meta-analyses.

We included peer-reviewed studies published in English. Diagnosis of BD (e.g., bipolar disorder I and II, cyclothymic disorder) had to be established using recognized clinical assessment tools (based on Diagnostic and statistical manual of mental disorders, DSM; International classification of disorders, ICD). Studies needed to provide measures of the prevalence of sleep disturbances in individuals with BD and/or quantification of sleep characteristics in these individuals, assessed with PSG, EEG, actigraphy, or self-reports (e.g., sleep diary). When both objective assessments were available, PSG data were preferred over actigraphy data. Mood phase of participants had to be clearly specified using validated mood rating scales, and for studies including participants in different mood phases, demographic information and sleep measures had to be reported for each mood phase group separately. Studies were excluded if they were unrelated to BD, case reports, comments, expert opinions, books, reviews, meta-analyses, animal studies, clinical trials, or intervention studies. Longitudinal studies were excluded when sleep and clinical assessments were not simultaneously obtained at baseline. Studies concerning specific populations (e.g., community dwelling elders, veterans, or affected by comorbidities such as neurological disorders, cancer, heart failure, atrial fibrillation or chronic obstructive pulmonary disease) were excluded. Studies reporting quantitative measures of sleep disturbances without a control group were excluded. Similarly, studies were excluded when no demographic information was reported or when the patient group included psychiatric disorders unrelated to BD (e.g., schizophrenia, anxiety disorders, major depressive disorder, etc.), and sleep data were not reported separately for each diagnosis. When sleep data pertained to a specific population sample, we contacted the authors to obtain detailed demographic information; when such data could not be retrieved, we excluded those studies. When several studies were performed on the same cohort, we included the study with the largest sample and most appropriate data for the purpose of the meta-analysis.

Of note, these inclusion and exclusion criteria led to the exclusion of several older studies, e.g., due to the absence of a standardized tool for the diagnosis of mental illness, a lack of sufficient information to establish clinical staging, and/or the absence of relevant sleep quality, continuity or duration parameters that were considered in the current meta-analysis.

After removal of duplicates, references were subdivided into three groups, and seven team members (CO, DF, KA, MM, MA, RJ and SC) independently and blindly screened titles, abstracts and, where relevant, the full text of studies to assess their eligibility. For each study, 2 authors conducted screening independently. When needed, a consensus was reached through the whole group of screeners or through the senior author (FF).

2.2. Data extraction

Data extraction was performed independently by six authors (CO, DF, KA, MM, MA and SC). Selected studies were divided into three groups, with two authors extracting data from each study. In addition to the sleep measures described above, the following variables were extracted: authors’ names, publication year, title, geographical location, publication journal, study design, inclusion and exclusion criteria, diagnostic assessment tools, number of participants, age (mean, SD), sex, and additional descriptive clinical data provided.

2.3. Definition of Mood Phase

In determining mood state for a given study, priority was given to clinical evaluation, followed by questionnaires administered by clinicians or other trained personnel, and/or self-reports. Three different mood states were specified: euthymic, depressive, and mania/mixed phases.

Euthymic phase:

Euthymia is the asymptomatic condition of BD. It is characterized by both the lack of clinical manifestations of a mood episode and the presence of features such as positive affect, psychological well-being, flexibility, consistency, and resistance to stress [36], and its attainment is the ultimate goal of treatment of mood disorders, including BD.

Depressive phase:

The depressive phase is characterized by at least five of the following clinical symptoms: depressed mood, anhedonia, abulia, fatigue, weight or appetite fluctuations, feelings of unworthiness or guilt, and recurrent thoughts of death [2]. In bipolar depression, depressive episodes are the predominant clinical feature [37].

Manic/mixed phase:

According to DSM-5-TR diagnostic criteria [2], BD are defined by the presence of at least one manic episode in the patient’s history, making manic/mixed phase the hallmark of the disorder. Mania is defined as a condition of pathologically elevated mood, energy status and activity [38]. It is a phase characterized by at least 3 of the following: disrupted mood with inflated self-esteem and grandiosity, pressure to keep talking, flight of ideas, distractibility, increase in goal-directed activity, and excessive involvement in possibly dangerous activities [2]. Mixed phase is defined by the presence of both manic and depressive symptoms in the same patient [39]. In DSM-5-TR, it is considered a specifier (“With mixed features”) of manic/hypomanic episodes and depressive episodes (although this can be used also in the context of major depressive disorder).

When the clinical phase was not clearly defined or reported, we classified patients as being in a mixed phase if they exhibited both manic and depressive symptoms, as indicated by elevated scores on psychometric scales or questionnaires (e.g., YMRS or HAM-D). We grouped mixed episodes with manic episodes when manic symptoms were present alongside depressive symptoms. This classification was based on the premise that mixed episodes are clinically and phenomenologically closer to manic episodes than to depressive episodes, given that the mood dysregulation in mixed states primarily involves elevated, irritable, or dysphoric mood rather than primarily depressed mood [39].

2.4. Outcome definitions

1. Poor Sleep Quality, Insomnia, and Sleep Disorder Prevalence: Different aspects of sleep problems were assessed using specific validated instruments:

   1. Poor Sleep Quality: Poor sleep quality was defined as a score >5 on the Pittsburgh Sleep Quality Index (PSQI) [40].
   2. Insomnia: Clinically significant insomnia was defined as a score >7 on the Insomnia Severity Index (ISI) [41] or a score >6 on the Athens Insomnia Scale (AIS) [42].

   We also screened for studies reporting the prevalence of sleep disorders other than insomnia, including obstructive sleep apnea, narcolepsy, periodic limb movement disorder (PLMD), nightmare disorder, delayed sleep phase disorder, and parasomnias.
2. Case-Control Sleep characteristics: the following sleep characteristics were compared in BD in different mood phases versus control groups:

   1. Self-reported sleep quality assessed by the PSQI.
   2. Sleep duration and sleep continuity parameters assessed using objective tools, including PSG and actigraphy (total sleep time; TST, sleep onset latency; SOL, sleep efficiency; SE, wake after sleep onset; WASO, number of awakenings, and variability in TST, SOL, and WASO).
   3. Sleep architecture (% in NREM and REM sleep stages, REM latency, REM density). For PSG studies, sleep stages had to be established using conventional methods (e.g., Rechtschaffen and Kales, 1968 [43]; American Academy of Sleep Medicine (AASM criteria [44]) and had to be reported for at least a full night.
   4. Sleep oscillation parameters, including sleep spindle density (number/minute), duration (sec), amplitude (μV), and frequency (Hz), as well as slow wave density.

3. Objective versus Subjective Sleep Duration and Continuity Measures: We compared objectively assessed sleep measures (PSG and actigraphy) with sleep diary data in euthymic BD. Where sufficient data were available, separate sub-analyses were conducted for PSG and actigraphy measurements.

2.5. Statistical analysis

The prevalence of sleep disturbances and differences in sleep characteristics were evaluated through mood phase-specific analyses when at least two studies were available for each mood phase. Although data were extracted from studies reporting specific sleep disorders (e.g., obstructive sleep apnea [OSA]), the inclusion of these disorders in our meta-analyses was precluded by an insufficient number of studies (fewer than 2) per mood phase. Effect sizes for sleep disturbance prevalence were logit transformed, quantifying the log odds. We used logit transformation for sleep disturbance prevalences because prevalences are bounded between 0 and 1, making direct statistical modeling difficult. The logit transformation maps those constrained values to the real line, enabling standard regression techniques and meta-analytical approaches [45, 46]. Additionally, by analyzing the logit transformed prevalences, we can directly characterize the effect of any independent variable on the log odds or odds of sleep disturbance, providing interpretable effect estimates. This procedure is equivalent to running logistic regression on the prevalence outcome and represents standard practice in meta-analysis of proportions [47, 48]. Recovery of missing or partial data from studies is elaborated in the supplementary material.

Pooled descriptive statistics (means ± standard deviations) were calculated for individuals with BD and controls using sample-size weighted means and appropriate variance pooling formulas. Sleep parameters were analyzed using both mean difference (MD) and standardized mean difference (SMD) values of Hedges’ g statistic [49]. A random-effects linear regression model was applied to each outcome, with effect sizes weighted by inverse variance to account for study variability [50]. Analyses were conducted using the rma function in the metafor package (R software v. 4.1.0; method = “PLO” for prevalence analyses; method = “MD” for mean differences; method = “SMD” for case-control comparisons). Both MD and SMD values were calculated for all sleep parameters to facilitate clinical interpretation alongside effect size comparisons. MD values are most interpretable for standardized measures (e.g., PSQI scores), whereas they should be interpreted with caution for most other sleep parameters given methodological heterogeneity across studies (e.g., different actigraphy devices, software algorithms, PSG scoring criteria), for which SMD values provide more appropriate comparisons. SMDs indicate the magnitude and direction of differences between cases and controls: positive values indicate higher outcomes in patients, while negative values indicate lower outcomes compared to controls. Effect sizes were interpreted as small (SMD = 0.2), medium (SMD = 0.5), or large (SMD = 0.8) [51]. Hypothesis testing utilized two-sided significance levels at p<.05 [52].

Moderator (meta-regression) analyses examined the effects of demographic variables (i.e., age and sex), medication (i.e., proportion of each study sample taking anticonvulsants and/or lithium), and mood phase comparisons on sleep disturbance prevalence and sleep characteristic differences using linear mixed effect meta-analysis models. We conducted additional moderation analyses to examine the effect of assessment methodology (objective versus subjective measures) on sleep duration and continuity parameters in euthymic BD. Such analyses were not possible for other mood phases due to insufficient studies. For each sleep parameter, we regressed the differences between patient and control groups from the available sample on each moderator variable. For prevalence outcomes, moderators were regressed against log-odds of sleep disturbances. A minimum of three studies was required for moderation analysis of demographic variables and medication, and two studies for each subgroup were required for analyzing the effect of mood phase or assessment methodology. A threshold of p<0.05 was used to establish statistical significance after Bonferroni correction for multiple moderator variables.

Funnel plots were generated to visualize study heterogeneity for sleep disturbance prevalence and sleep characteristic differences when a minimum of three studies was available. Hypothesis testing of Cochran’s Q statistic [53] was used to assess the heterogeneity of studies in each parameter. The I² statistic [53] was used to assess the magnitude of heterogeneity. Egger’s test [54] was used to assess funnel plot asymmetry.

Quality and risk of bias were assessed using the Agency for Healthcare Research and Quality (AHRQ) [55, 56] methodology checklist for cross-sectional/prevalence studies and a modified version of The Newcastle-Ottawa Scale (NOS) [57] for sleep architecture studies. For additional details, see Supplementary Methods.

3. Results

The initial search yielded 12,977 records (Figure 1). After removing 3,486 duplicate studies, we screened 9,491 publications and excluded 9,015 studies based on title and abstract screening, resulting in 476 studies considered for full-text review. No additional articles were identified through reference checking. After full-text review, 44 articles were included, with 11 studies assessing sleep disturbance prevalence in 5,264 individuals with BD (Table S1) and 34 studies measuring sleep characteristic differences in 2,713 participants with BD and 3,164 HC subjects (Table S2). Comprehensive descriptive statistics showing pooled means ± standard deviations for all sleep parameters across clinical states are presented in Supplementary Table S3, alongside both MD values and SMD values for clinical interpretation.

Figure 1.

PRISMA workflow of study selection. *Of note, one study [13] was included in both sleep prevalence and sleep architecture analyses.

3.1. Poor sleep quality, insomnia, and sleep disorder prevalence

Prevalence of poor sleep quality during the euthymic phase was 52% (95% CI 33%−72%, Q = 189.869; df = 5; Figure 2). Insomnia prevalence during the depressive phase was 63% (95% CI: 56%, 69%; Q = 0.0064; df = 1; Figure 2). For individual studies’ forest plot, see Figure S1. Insufficient studies were found for manic/mixed phase for sleep disturbance meta-analysis, nor were there enough studies to meta-analyze sleep disorders in any mood phases.

Figure 2.

Forest plots of prevalence of poor sleep quality and insomnia in individuals with bipolar disorders.

3.2. Self-reported sleep quality

Individuals with BD during the euthymic phase demonstrated significantly poorer sleep quality compared to controls, with pooled PSQI scores of 7.26 ± 4.27 versus 4.33 ± 2.67 in controls (n = 1,311 individuals with BD, n = 1,052 controls across 13 studies). This corresponded to a mean difference of 2.65 points [95% CI: 2.02, 3.27] and a large effect size (SMD = 0.86 [95% CI: 0.68, 1.03]; P < 0.001, Figure 3). Notably, the mean PSQI score in individuals with BD exceeded the clinical cutoff of 5, indicating clinically significant sleep disturbances, while controls scored below this threshold. Insufficient studies were available to meta-analyze data for other mood phases. Insufficient studies were available to meta-analyze data for other mood phases. Individual studies’ forest plot is available in Figure S2.

Figure 3.

Standardized mean difference (SMD) values between bipolar disorders (BD) and healthy control groups in overall sleep quality, assessed with the Pittsburgh Sleep Quality Index (PSQI) total score.

3.3. Objective sleep duration and continuity parameters

Using objective (i.e., combined actigraphy and PSG studies) sleep assessments, individuals with BD in the euthymic phase showed longer TST (mean: 476.3 ± 73.97 vs 433.42 ± 60.54 minutes; MD = 39.69 minutes [95% CI: 25.88, 53.50]; SMD = 0.62 [95% CI: 0.44, 0.81]; P < 0.001), and prolonged SOL (mean: 17.14 ± 17.35 vs 13.9 ± 12.68 minutes; MD = 2.99 minutes [95% CI: 1.28, 4.70]; SMD = 0.25 [95% CI: 0.11, 0.40]; P < 0.001) compared to controls. No significant differences were found for SE, WASO, and number of awakenings. For the depressive phase, no significant differences were found in any of the sleep duration and continuity measures, while individuals in manic/mixed phase showed shorter TST (mean: 420.04 ± 127.71 vs 453.19 ± 93.17 minutes; MD = −47.30 minutes [95% CI: −75.72, −18.89]; SMD = −0.60 [95% CI: −1.17, −0.03; P = 0.038]), longer SOL (mean: 20.97 ± 28.11 vs 14.67 ± 12.1 minutes; MD = 3.94 minutes [95% CI: 0.92, 6.97]; SMD = 0.38 [95% CI: 0.08, 0.67; P = 0.012]) and lower SE (mean: 84.40 ± 9.52% vs 92.38 ± 5.75%; MD = −7.72% [95% CI: −10.24, −5.21]; SMD = −1.04 [95% CI: −1.41, −0.67; P < 0.001]) compared to controls. Not enough studies were available to meta-analyze WASO and number of awakenings for the manic/mixed phase. Higher variability in TST (mean: 76.02 ± 37.18 vs 67.20 ± 37.43 minutes; MD = 10.63 minutes [95% CI: −3.88, 25.14]; SMD = 0.31 [95% CI: 0.01, 0.60; P = 0.043), SOL (mean: 26.83 ± 30.96 vs 14.99 ± 16.10 minutes; MD = 11.96 minutes [95% CI: 1.31, 22.60]; SMD = 0.39 [95% CI: 0.04, 0.73; P = 0.03]), and WASO (mean: 27.10 ± 24.14 vs 16.51 ± 9.27 minutes; MD = 9.27 minutes [95% CI: 2.49, 16.06]; SMD = 0.53 [95% CI: 0.19, 0.88; P = 0.003]) were also observed in euthymic BD versus controls (Figure 4). Forest plots showing individual study data for all sleep duration and continuity measures are presented in Figures S3-10. Moderator analysis revealed significantly higher TST during euthymia compared to both the manic/mixed (z = 4.62, P < 0.001) and depressive (z = 2.12, P = 0.034) phases. Additionally, the manic/mixed phase had significantly lower SE compared to euthymia (z = −2.56, P = 0.01) and depressive phase (z = −3.79, P < 0.001). No other significant differences between mood phases were established (Table S4).

Figure 4.

Standardized mean differences (SMD) values between bipolar disorders (BD) and healthy control groups in sleep duration and sleep continuity parameters, assessed by actigraphy and polysomnography. Significant differences (P < .05) in effect sizes between mood phases are indicated by brackets.

3.4. Subjective versus objective sleep measures in euthymic phase

To compare sleep diary and objectively assessed sleep measures, adequate data were available only for the euthymic phase. Consistent with previous findings, a sensitivity analysis of objective studies found longer TST with both actigraphy (SMD = 0.66 [95% CI: 0.46, 0.86; P < 0.001]) and PSG (SMD = 0.38 [95% CI: 0.05, 0.71; P = 0.023]) assessments, while sleep diary studies found no significant TST differences in euthymic BD. Furthermore, moderation analysis revealed significant differences between objective and subjective assessments for TST (z = 2.35, P = 0.019). Regarding SOL, while sleep diary (SMD = 0.71 [95% CI: 0.31, 1.11; P < 0.001]) and actigraphy studies (SMD = 0.24 [95% CI: 0.07, 0.41; P = 0.005]) yielded significant differences between groups, the only two PSG studies assessing SOL reported no differences between euthymic BD versus HC groups. As for TST, moderation analysis revealed significant differences between objective and subjective assessments for SOL (z = −2.55, P = 0.011). Objective assessments of SE failed to establish significant differences between euthymic individuals with BD and healthy controls. In contrast, meta-analysis of sleep diary studies showed significant differences between groups (SMD = −0.57 [95% CI: −0.97, −0.17]; P = 0.006). These inconsistent findings were confirmed by our moderation analysis (z = 2.54, P = 0.011). Regarding WASO, both actigraphy (SMD = 0.22 [95% CI: 0.07–0.36; P = 0.003]) and sleep diary (SMD = 0.54 [95% CI: 0.30–0.78; P < 0.001]) studies found longer WASO during the euthymic phase, but objective assessments, which included one PSG study, yielded no differences between groups, and moderation analysis revealed significant differences between objective and subjective measurements of WASO (z = −2.00, P = 0.045). Finally, the number of awakenings was higher using sleep diary studies (SMD = 0.58 [95% CI: 0.27–0.88; P < 0.001]), while two objective assessment studies found non-significant differences between groups (Figure 5), and moderation analysis showed no significant difference between objective and subjective assessments (z = −1.94, P = 0.052).

Figure 5.

Comparison between standardized mean differences (SMD) values between bipolar disorders (BD) and healthy control groups in sleep duration and sleep continuity parameters assessed by actigraphy, polysomnography, or sleep diary. Significant differences (P < .05) in effect sizes between assessment modality are indicated by brackets.

3.5. Sleep architecture and sleep oscillations

No significant differences were found between euthymic BD and control subjects in the sleep architecture parameters that were meta-analyzed (N1%, N2%, SWS%, REM%, REM Latency), although N1% showed a trend toward significantly higher values in BD (SMD = 0.41 [95% CI: −0.003, 0.82], P = 0.052). Depressive phase BD showed significantly higher REM% compared to controls (22.12 ± 6.24% vs 19.23 ± 5.12%; MD = 3.36% [95% CI: 0.49, 7.21]; SMD = 0.51 [95% CI: 0.04, 0.98]; P = 0.034), while no significant differences were found in the other parameters analyzed (N1%, N2%, SWS%, REM Latency, REM Density), and there were not enough studies available for the manic/mixed phase (Figure 6). Figures S11-S16 show forest plots of individual studies for sleep architecture parameters. Furthermore, moderator analysis showed no sleep architecture differences between euthymic and depressive phase (Table S4). Concerning sleep oscillations (sleep spindles, slow waves, and sawtooth waves), there were not enough studies examining them to conduct meta-analyses.

Figure 6.

Standardized mean differences (SMD) values between individuals with bipolar disorders (BD) and control groups in sleep architecture assessed by polysomnography.

3.6. Meta-regression of age, sex, and medications

Meta-regression of sex and age (Tables S5-S7) revealed a significant age effect only on TST in the manic/mixed phase, with older individuals with BD reporting longer TST (Table S7). No other effects of age or sex were found for any other sleep parameters (Tables S5-S7). The effects of anticonvulsant and lithium medication percentages could only be assessed in the euthymic phase for PSQI total score and main sleep duration and continuity measures, revealing no significant effects after correcting for multiple comparisons (Table S5).

3.7. Study heterogeneity

Figures S17-20 show funnel plots for sleep disturbance prevalence and sleep parameters considered in the study, including summaries of heterogeneity and asymmetry. Study heterogeneity, assessed with Cochran’s Q statistic for measures with at least 3 studies, was high for two of the sixteen measures considered in this meta-analysis (specifically, prevalence of poor sleep quality in euthymic phase and TST in manic/mixed phase), as evidenced by large I² (>75%) [53] and P<0.05. No sleep parameter showed evidence of asymmetry according to Egger’s test (all p_sym ≥ 0.05).

3.7. Study quality appraisal

Studies of sleep disturbance prevalence tended to be of higher quality, with more detailed reporting of population inclusion, assessment period, and patient response rates, resulting in nine as “good” quality and two as “weak” quality studies. Regarding sleep characteristics, 26 studies rated as “good,” while eight studies rated as “fair” (Tables S8 and S9).

Discussion

To the best of our knowledge, this is the first systematic review and meta-analysis assessing the prevalence of sleep disturbances and differences in sleep characteristics in individuals with BD in different mood phases. In individuals with BD, sleep-related problems were highly prevalent across mood phases, with 52% poor sleep quality prevalence during euthymia and 63% insomnia prevalence during the depressive phase. In case-control comparisons, individual with BD during euthymic phase showed differences in several sleep characteristics, including poorer sleep quality, longer TST and SOL, as well as greater variability in TST, SOL, and WASO versus control groups. In euthymic BD, we also found poorer subjective evaluations of sleep duration and continuity compared to objective assessments. Moreover, people with BD in the depressive phase showed higher REM%, while individuals in the manic/mixed phase had shorter TST, longer SOL, and lower SE relative to control groups. Finally, moderation analysis revealed euthymic BD had longer TST compared to both depressed and manic/mixed phase, while manic/mixed phase individuals had worse SE compared to both euthymic and depressed people with BD.

Sleep disruption is part of the diagnostic criteria of a mood episode in BD. A broad range of sleep disturbance rates has been previously reported in individuals with BD [21, 58–62], in part due to some studies not using standardized measures to assess those disturbances [59, 61, 62]. By including only studies using standardized assessments or validated measures, this meta-analysis demonstrated high prevalences of sleep-related problems: 52% for poor sleep quality during euthymia and 63% for insomnia during the depressive phase. Additionally, case-control comparisons revealed poorer subjective sleep quality in BD during the euthymic phase, which corresponded to a large effect size (SMD = 0.86) with MD of 2.65 points. While this difference falls within the range of values considered clinically relevant in the literature (2.5–3 points), the interpretation of clinical significance should be tempered, especially in light of the ongoing uncertainty regarding minimal clinically important difference (MCID) thresholds for PSQI [63]. This underscores the need to identify sleep complaints in BD in all mood states, by asking about sleep quality, continuity, and timing during routine clinical visits.

A recent meta-analysis of actigraphy data reported higher TST in remitted BD [28]. Here, we confirmed this finding in sleep studies using actigraphy, but also established higher TST in studies utilizing PSG, in individuals with BD during euthymic phase. At the same time, sleep diary studies failed to report changes in TST, suggesting that combining subjective reports with objective sleep assessments is better suited to identify changes in sleep duration in BD during euthymia. Clinical observation and patients’ self-reported experiences suggest that TST varies across different mood phases [64]. Manic phases are usually accompanied by reduced sleep need and insomnia [9, 65], while depressive phases often occur with insomnia or hypersomnia [10]. Individuals in the BD depressive phase showed no differences in overall TST compared to controls, possibly reflecting a combination of both insomnia and hypersomnia phenotypes. In contrast, patients in the manic/mixed phase had shorter TST relative to controls, as well as compared to both euthymic and depressive phase patients, consistent with reduced need for sleep. These findings are consistent with a systematic review reporting a significant association between shorter TST and mood elevation the following day [19].

Difficulties with initiating and maintaining sleep are common in BD [66]. Findings from our meta-analysis corroborate these clinical observations, with individuals with BD exhibiting longer SOL during the euthymic phase versus healthy controls. Shorter TST, longer SOL and lower SE were also observed in the manic/mixed phase, but not during the depressive phase. Together, these findings indicate that several sleep duration and continuity parameters are altered in BD regardless of the mood phase. However, differences in some of these parameters occur specifically (i.e. reduced TST) or more prominently (i.e., SE) in the manic/mixed phase, while they are not present in the depressive phase, thus showing mood phase-specific differences. Thus, by routinely assessing these sleep parameters and their relationships with mood phases, clinicians may select better pharmacological and non-pharmacological interventions for individuals with BD.

Higher variability in TST, SOL and WASO in euthymic BD individuals suggests persistent sleep disruption even during periods of mood stability, further indicating that differences in these sleep parameters do not simply reflect an acute mood episode [7, 67]. Greater sleep variability is associated with disrupted autonomic functioning, particularly reduced parasympathetic tone and altered stress reactivity [68]. The presence of heightened sleep variability during euthymia may, therefore, indicate incomplete recovery of sleep regulatory systems and represent a vulnerability marker for future mood episodes [10, 16, 22]. The findings of increased sleep variability in euthymic BD may have clinical relevance, as previous research has linked sleep variability to altered stress responses [68]. However, further research is needed to determine whether stabilizing sleep patterns represents a meaningful therapeutic target even during periods of euthymia, as irregular sleep patterns may contribute to ongoing dysregulation of biological stress response systems that, in turn, can increase risk for mood episode recurrence [69]. The persistence of sleep variability during euthymia also highlights the potential value of longitudinal tracking of sleep variability, in addition to duration and continuity measures, as a clinical marker in BD [7, 18, 24]. Another interesting finding of our meta-analysis was the discrepancy between subjective and objective assessments of sleep duration and continuity parameters in BD during euthymia, with subjective measures showing larger differences from controls than objective measures. Euthymic individuals exhibit higher levels of neuroticism, which correlates with poor sleep quality, as well as a greater number of recent stressful life events [70]. Altered sleep perception has also been reported in these individuals [71, 72]. Combined, these factors may exacerbate perceived sleep difficulties, suggesting that psychological and psychosocial variables play a crucial role in how patients evaluate their own sleep. While objective measures like actigraphy and PSG may not consistently detect severe abnormalities during euthymia, subjective reports of sleep problems often persist, potentially contributing to ongoing functional impairments and increased morbidity [10, 73]. Thus, treatment approaches should integrate both objective assessments and patient-reported outcomes. Furthermore, the high rates of sleep quality and continuity difficulties in BD calls for the widespread provision of sleep interventions, particularly cognitive behavioral therapy for insomnia (CBT-I), where possible adapted for populations with mental disorders [74]. Despite evidence for feasibility [75] and effectiveness [74, 76] in mental disorders, CBT-I remains under-resourced. Additionally, maintaining a sleep diary offers a low-cost, informative method for tracking sleep patterns and identifying discrepancies, thereby guiding more targeted clinical strategies [70]. By acknowledging the interplay between personality factors, recent stressors, and subjective perceptions of sleep, clinicians can develop comprehensive management plans aimed at optimizing both sleep and overall well-being in individuals with BD.

Concerning sleep architecture parameters, we found a small but statistically significant increase in REM% in BD depressive phase. While REM abnormalities have been consistently reported in individuals with major depressive disorder [77, 78], the clinical significance of this difference in BD requires careful interpretation. These findings are consistent with a recently hypothesized implication of REM abnormalities in the neurobiology of BD [33], but further research is needed to determine the clinical relevance of the observed differences in REM parameters.

This meta-analysis revealed several research gaps that require further investigation. Only two and three PSG studies were found for the euthymic and depressive phases, respectively, and only one study for the manic/mixed phase, which therefore could not be meta-analyzed. The limited number of PSG studies in BD is particularly concerning, given that PSG remains the gold standard for sleep assessment and provides unique information about sleep architecture parameters that cannot be obtained through any other methods. Thus, more studies are urgently needed to better understand the neurophysiological underpinnings of sleep disturbances in BD.

We also found insufficient studies across all mood phases investigating sleep oscillations, including sleep spindles, slow waves, and saw-tooth waves. Given that sleep spindle and, to a lesser extent, slow-wave deficits have been reported in psychiatric disorders [79], especially in individuals with schizophrenia [80–83], and that genetic risk and clinical phenotypes overlap between these disorders and BD [84, 85], more studies are needed to assess sleep oscillations in individuals with BD across different mood phases. Future studies should also focus on investigating REM oscillations (i.e., saw-tooth waves) in BD and their potential role in the disorder’s pathophysiology and clinical symptoms.

Limitations

Despite offering novel insights into sleep disturbances in BD, our meta-analysis has several limitations. First, for several sleep parameters we did not have enough studies for a given mood phase to perform a meta-analysis. However, our findings suggest how important it is to examine sleep disturbances in BD in different mood phases. Second, variability in mood-phase definitions and clinical assessments likely contributed to heterogeneity across studies, potentially influencing effect size estimates. To address this limitation, we applied consistent criteria for defining euthymic, manic/mixed, and depressive phases, but further research with uniform diagnostic procedures is warranted. Third, medication status was only partially documented, making it difficult to determine its full impact on sleep measures. Future investigations should therefore prioritize tracking medication regimens and controlling for their effects on sleep. Fourth, comparisons between objective and subjective sleep measures were only possible for the euthymic phase due to insufficient studies in other mood phases. Fifth, we applied relatively liberal minimum criteria for actigraphy (≥2 nights) and sleep diary (≥3 nights). However, most studies greatly exceeded those thresholds to the extent that only three actigraphy studies had <5 nights, while only one sleep diary study had <5 nights. Nonetheless, variability in the number of actigraphy/sleep diary nights across studies may have contributed to effect size heterogeneity. Furthermore, we did not systematically extract or control for the actigraphy scoring algorithms used across studies; since actigraphy sleep estimates vary significantly depending on scoring methods (e.g., Cole-Kripke, Sadeh, manufacturer-specific algorithms), this methodological heterogeneity may have contributed to our findings’ variability. Finally, most of the included studies were cross-sectional, limiting causal inferences about whether sleep alterations drive or simply reflect mood changes. Longitudinal designs with frequent mood and sleep assessments are therefore needed to clarify these relationships. Despite these challenges, our meta-analysis represents the most comprehensive effort to date to delineate sleep disruptions across distinct BD mood phases.

4. Conclusions

This systematic review and meta-analysis revealed that poor sleep quality and insomnia are highly prevalent in BD, and that differences in sleep parameters occur throughout the course of the disorder, including during euthymia. These findings highlight the need for future research studies to employ standardized, validated tools for reporting sleep measures in relation to mood phases. Research efforts should also extend beyond conventional sleep measures and harness emerging mobile and portable EEG technologies to assess sleep patterns in the patients’ home environments [86, 87]. Moreover, the presence of REM differences in BD underscores the importance of further investigating REM dysregulation using EEG (e.g., REM sawtooth waves). While future work will help establishing the implications of sleep abnormalities on the neurobiology and clinical manifestations of BD, the present findings of pervasive, persistent findings in BD across mood phases highlights the need for monitoring sleep throughout the course of the disorder.

Practical points.

1. Poor sleep quality and insomnia are highly prevalent in BD, even during euthymia;

2. Objective and subjective measurements can diverge, with subjective measures often showing larger differences between individuals with BD and controls during euthymia compared to objective measures;

3. Altered sleep patterns are present during euthymia but are also distinctively modulated during the depressive and mania/mixed phases.

Research agenda.

Future research needs to focus on:

1. Conducting more PSG studies across all mood phases, particularly during manic/mixed episodes, to better examine alterations in sleep architecture and sleep oscillations (sleep spindles, slow waves, and saw-tooth waves), given their potential role in BD pathophysiology;

2. Design longitudinal studies to understand how sleep alterations may predict or influence mood phase transitions;

3. Compare objective versus subjective measures in relation to personality traits, recent stressors, and comorbidities to better understand discrepancies in perceived versus recorded sleep;

4. Ascertain medication effects on sleep patterns in individuals with BD across mood phases;

5. Develop and test targeted sleep interventions (e.g., CBT-I, bright light therapy) in BD.

Abbreviations

AASM

American academy of sleep medicine

AHRQ

Agency for healthcare research and quality

AIS

Athens insomnia scale

BD

bipolar disorders

CI

confidence interval

DSM

Diagnostic and statistical manual of mental disorders

DSM-5-TR

diagnostic and statistical manual of mental disorders – 5 – text revision

EEG

electroencephalogram

HAM-D

Hamilton rating scale for depression

HC

healthy control(s)/comparison(s)

ICD

International classification of diseases

N1

non-REM sleep 1

N2

non-REM sleep 2

N3

non-REM sleep 3

NREM

non-rapid eye movement

PLMD

periodic limb movement disorder

PRISMA

preferred reporting items for systematic reviews and meta-analyses

PSG

polysomnography

PSQI

Pittsburgh sleep quality index

REM

rapid eye movement

SE

sleep efficiency

SMD

standard mean difference

SOL

sleep onset latency

SWS

slow wave sleep

TST

total sleep time

WASO

wakefulness after sleep onset

YMRS

Young mania rating scale

Access to data and data analysis:

The principal investigators (AM and FF) had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.