Sleep in Bipolar Disorder: A Systematic Review and Meta-Analysis of Case Control Drug Free Patients

1. Introduction

Bipolar disorder (BD) is a chronic psychiatric condition [1,2] that is estimated to affect 0.5-3% of the population [2,3,4]. Periods of depressed mood, with hallmark loss of interest and pleasure, alternate with periods of hypomania (type II BD) or mania (type I BD), presenting with symptoms that range from unusually elated mood, disinhibition and restlessness to extreme irritation and anger [1,2,3,4,5]. Depressive episodes are nonetheless reported to account for 75% of the symptomatic time [1]. Both manic and depressive phases can present psychotic features [6], especially in those diagnosed with type I BD [7]. Psychotic symptoms can be congruent or incongruent with the mood phase, with incongruence being associated with a poorer outcome of the condition [6]. These abnormal mood periods are separated by “euthymic” intervals where mood appears normal, although residual symptoms are present in most cases [8,9]. Periods of euthymia can last weeks to years [10]. Such periods can be virtually absent in rapid-cycling patients, defined as experiencing at least four mood episodes in a year [11,12,13] and associated with poorer treatment response [14]. Similarly, BD patients with mixed features are associated with poorer treatment response and outcomes [15], experiencing overlapping symptoms of both manic/hypomanic and depressive phase [16,17,18]. Euthymia is often achieved with pharmacological treatment: despite lithium salts being the gold standard choice [19], other mood stabilizers like valproic acid [20] and lamotrigine [21] and antipsychotics like quetiapine [22,23] are widely used for the treatment of BD as well [24]. Antidepressants are generally avoided as monotherapy given their risk to trigger mania [24,25]. Nonetheless, poor compliance with pharmacological treatment is common in BD [1]. Patients with BD have the highest suicide rate among psychiatric disorders [26], with approximately 15-20% of patients dying by suicide [1,27].

Sleep disturbances in BD have been reported since early studies on the “manic-depressive illness” [28,29,30] and associated with poorer quality of life [31,32,33] and cognitive functioning [34,35], suicide attempts [36,37] and higher mood episode relapse [38,39]. The circadian rhythm disturbances observed in BD are believed to play a pivotal role in the genesis and outcome of BD [40,41,42,43,44]. Sleep reduction has been proposed to be the common pathway of various biological and psychosocial factors triggering the onset of mania, creating a self-reinforcing cycle [45]. Decreased need for sleep and insomnia is, in fact, observed in hypomania and mania [46], whereas the depressive phase is associated with both insomnia and hypersomnia [38,47,48], although the latter appears to be more related to anergia rather than to increased sleep propensity [49]. High prevalence of sleep disturbances was nonetheless observed prior to a mood episode [50].

1.1. Polysomnographic Studies of Bipolar Disorder and Healthy Controls

Polysomnography (PSG) case-control investigations of sleep in BD and in healthy controls (HC) have been conducted in the last decades with heterogeneous results. Total sleep time (TST) was consistently found to be decreased in drug-free manic patients [51,52,53] and normal in drug-free depressed BD patients [54,55,56,57,58,59]. Nonetheless, a decreased TST in bipolar depression was also found [51,60,61]. Sleep onset latency (SOL) was found to be prolonged in bipolar depression [51,54,60,61], although other studies reported normal SOL in both the depressive [55,57,58,59] and manic/hypomanic phase [52,53,62,63]. Sleep efficiency (SE) was reported to be decreased in mania [51,53], hypomania [63] and bipolar depression [51,54,61], although several studies reported normal SE in mania [52,62] and bipolar depression [55,57,58,60]. Wake after sleep onset (WASO) was mostly reported to be normal in both mania [51,52,53,62] and bipolar depression [51,54,55,57,58,59,60].

Reports on NREM sleep in bipolar disorder are not consistent. Some studies reported no abnormalities in drug-free BD [52,62,64], while others reported abnormalities in either or both sleep stage 1 and 2 duration (ST1, ST2) and percentage (%ST1, %ST2) [53,56,57,59,60,61,63]. Regarding slow-wave sleep (SWS), several studies found no differences in either delta sleep time (DELTA) or percentage (%DELTA) regardless of the mood phase [51,52,54,58,59,61], while few studies reported reduced stage 3 sleep duration (ST3) [62] or increased percentage (%ST3) [53]. %DELTA was found to be reduced in hypomania [63].

Reports on REM sleep in bipolar disorder are similarly heterogeneous: REM sleep time (REMT) and percentage (%REM) were found to be normal regardless of the mood phase [51,53,54,56,57,58,59,60,62,63,65]. REM latency (REML) was mainly reported to be shortened in the depressive phase [51,54,55,58,60,61,66] and in mania [51,52,53] or hypomania [63], although not different in other studies [57,59,62,65,67]. REM density (REMD) was reported to be increased [52,53,58] in mania, hypomania [63] and in bipolar depression [60,61], although other studies reported normal REMD in each phase [51,54,55,57,59,65].

PSG studies in drug-free euthymic patients are scarce, reporting increased %REM and REMD in the first sleep cycle [68], increased ST1 [69] or no differences compared to HC [70]. One study reported reduced TST and increased SOL in euthymic BD [71].

1.2. Polysomnographic Studies of Bipolar and Unipolar Depression

Since early studies, patients with bipolar depression have been reported to have longer TST [54,55,58,60,72,73,74] and increased SE [54,55,58,73] than patients with unipolar depression (UD). Reports on SOL and WASO [51,54,55,60,73] in BD depression compared to UD, conversely, provided very heterogeneous results.

A similar heterogeneity of findings concerns both NREM sleep [51,60,73,74,75], including SWS [51,54,56,58,72,73,76], and REM sleep parameters [51,54,55,56,58,59,60,61,64,67,72,73,74,77]

A recent study reported reduced %ST2, %DELTA and REML and increased %REM in minimally treated, first-episode BD compared to UD [78].

1.3. Polysomnographic Studies of Pharmacological Treatment in Bipolar Disorder

The effect of lithium on BD sleep parameters has been investigated in a limited number of studies, most of which reported an increase in TST and SE, a reduction in SOL but no consistent beneficial effect on WASO. NREM sleep increased, especially through an increase in ST2 and SWS, as REMT increased and REML was prolonged [79,80,81]. Some studies investigated the effect of tricyclic antidepressant amitriptyline or monoamine oxidase inhibitor tranylcypromine on bipolar depression sleep, both showing a marked REM-suppressing effect [82,83]. Regarding the effect of antipsychotic treatment on sleep during the manic phase, olanzapine was found to be more effective than haloperidol in stabilizing sleep, increasing SE and decreasing WASO [84], and quetiapine was reported to increase TST and REML [85].

When BD patients receiving stable lithium treatment were compared to HC, some studies found no effect of lithium on sleep parameters [86] or that TST, SE and NREM sleep were still decreased in BD [87]. Nonetheless, another study found increased NREM and significant REM suppression after long-term lithium treatment [88]. PSG studies of BD patients on stable combination pharmacotherapy (mood stabilizers, antipsychotics, antidepressants, benzodiazepines) reported an overall stabilization of sleep parameters, although with an increase of REMD and a longer SOL, compared to HC [89,90,91,92,93,94].

The present study aims to systematically review the available literature on sleep parameters in bipolar disorder patients, in the different phases of the illness, when drug free or under drug treatment, and in comparison, with normal controls and unipolar depression. A meta-analysis was also conducted on the available case-control studies.

2. Methods

The literature search was conducted querying PubMed, Web of Science (WoS) and PsycINFO databases from inception to 2026. The main keywords combination used to ensure the identification of all the relevant studies was: “(sleep OR sleep∗) AND bipol*”, although different other keywords were combined to perform a thorough research: “mania OR manic OR manic-depressive”, “PSG OR polysomnogra∗”, “EEG OR electroencephalogra∗”, “REM OR REMS OR rapid-eye-movement OR rapid-eye-movements”, “”NREM or non-REM”, “SWS OR slow-wave”, “paradox OR paradox∗”, “print OR print∗”, “D-phase OR dream OR dream∗”, “night OR night∗”, “monit∗”, “architect∗”. Specific attention was devoted to identifying studies regarding the effect of drugs on BD patients, therefore further keywords also based on both molecule and brand name were implemented in the search, such as: “mood stabilizer”, “anticonvulsant”, “lithium”, “antipsychotic”, “Depakote”, “Lamictal” and so on. These queries provided 3702 studies from PubMed, 4543 studies from WoS and 72 studies from PsycINFO. Out of these 8317 studies, 4541 resulted in duplicates present on multiple databases. The remaining 3776 studies were entirely read to assess their usefulness: 3670 studies were excluded due to lack of sleep data provided. 106 studies were assessed as eligible and all of them were retrieved, although 5 of these were excluded due to presenting a patient's sample (and their sleep data) already entirely presented in a previous study by the same author or co-author. 107 studies were included in the study: 101 studies were included in database research, and 6 other eligible studies were manually researched and found (due to not being indexed in any database previously searched). 43 studies were assessed as eligible for meta-analysis, while the remaining 64 studies were assessed as eligible for the systematic review only. The eligibility decision was made after a full-text article reading by the authors (GB, BT and DM), all agreeing on the inclusion or exclusion of each study. The meta-analysis was conducted in order to investigate sleep parameters differences between the following groups: drug-free BD and HC (Table 1), drug-free BD and drug-free UD (Table 2), treated BD and HC (Table 3), drug-naive/drug-free BD and treated BD (pre and post-treatment) (Table 4). Case-control studies included in the systematic review only are reported in Table 5 (with the relative reason for exclusion) [93,95,96,97,98,99,100,101,102,103,104,105,106,107]. All the other non-case-control studies included in the systematic review only are reported in Table 6 [29,30,49,85,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152,153]. Two studies [85,93] provided separate datasets, one eligible for inclusion in the meta-analysis and the other only for the systematic review. Figure 1 presents the PRISMA (Preferred reporting items for systematic reviews and meta-analyses) diagram of the systematic review.

2.1. Eligibility Criteria

Studies were considered eligible for inclusion in the meta-analysis if they met the following criteria:

1) Case-control studies including at least one sample of patients with BD, whether drug-naive, drug-free, or treated, and one HC sample.

2) Studies including at least one sample of patients with BD and one sample of patients with UD, both of whom were drug-naive or drug-free.

3) Studies including at least one sample of patients with BD assessed before (while drug-naive or drug-free) and after treatment with at least one antipsychotic, antidepressant or mood stabilizer.

4) Studies in which the drug-free period was clearly reported.

5) Studies in which BD was diagnosed according to DSM and/or ICD criteria.

6) Studies in which sleep was assessed using PSG and scored according to the AASM or Rechtschaffen & Kales criteria.

Studies were excluded based on the following criteria:

1) Studies without an HC sample.

2) Studies in which the BD sample included more than 5% of participants with diagnoses other than BD (e.g., a mixed BD and UD sample).

3) Studies in which more than 5% of the BD sample differed from the majority of the sample with respect to treatment status (e.g., 90% drug-free and 10% treated).

4) Studies with a BD patient sample with comorbidities (e.g., ADHD, GAD).

5) Studies in which the mood phase of the BD sample was not specified.6) Single-case studies.

7) Studies describing PSG measurements in BD patients without reporting clear data.

The excluded studies were included in the systematic review.

As specified above, studies were excluded if the BD sample was not sufficiently homogeneous with respect to diagnosis and medication status. A conservative threshold (>5% of the sample) was applied to minimize potential confounding (154,155) resulting from the inclusion of participants with different diagnoses and/or exposure to psychotropic medications, both of which may substantially influence sleep architecture and polysomnographic parameters.

The quality of each study included in the systematic review and/or meta-analysis was assessed using the Newcastle-Ottawa Scale, and only studies with a score of at least 6 were included in the meta-analysis.

2.2. Data Extraction

Data extracted from each study were: sample size, age, gender, diagnosis, medication status, type of medication (if reported), mean and standard deviation (SD) of TST, SOL, SE, WASO, REMT, %REM, REML, REMD, ST1, ST2, ST3, ST4, %ST1, %ST2, %ST3, %ST4,

DELTA, %DELTA. Some studies did not report SD for their sleep data; therefore, SD has been estimated as follows. In each meta-analysis (e.g. REM density, BD vs HC), the average SD was estimated by calculating the weighted average (based on the number of participants of each study) of all the available SDs. To the studies participating in the very same meta-analysis and whose sleep parameters were reported without SD, this estimated, weighted average SD was attributed. This estimation method for SD is regarded as reliable (Higgins et al., 2024) and is the only estimation process regarding missing data conducted in this study. In a limited number of studies some sleep parameters were not reported clearly but could be extractable by calculating them from other parameters (e.g., REMT = TST/100 x %REM). All these cases of data extraction have been properly highlighted in the supplementary material, where meta-analysis and systematic review data is displayed in extended form.

Some studies were reported multiple times in several tables for they included multiple independent patient samples or diagnostic subgroups that constituted separate datasets to the analyses. Therefore, repeated citations do not represent duplicate inclusion of the same sample, but distinct patient samples analyzed separately within the original study.

2.3. Statistical Analysis

Jamovi 2.6.44 (desktop Windows x64 solid version) was used to conduct the meta-analysis. R 4.6.0 (desktop Windows x64 version) with meta power 0.2.2 package was used to compute statistical power for each meta-analysis. A minimum of three data sets from studies was required to perform a meta-analysis for each parameter. The analysis was carried out using the standardized mean difference as the outcome measure. A random-effects model was fitted to the data. The amount of heterogeneity (i.e., tau2), was estimated using the restricted maximum-likelihood estimator. In addition to the estimate of tau2, the Q-test for heterogeneity and the I2 statistic are reported. In case any amount of heterogeneity is detected (i.e., tau2 > 0, regardless of the results of the Q-test), a prediction interval for the true outcomes is also provided. Studentized residuals and Cook's distances are used to examine whether studies may be outliers and/or influential in the context of the model. Studies with a studentized residual larger than the 100 x (1 - 0.05/(2 X k)) th percentile of a standard normal distribution are considered potential outliers (i.e., using a Bonferroni correction with two-sided alpha = 0.05 for k studies included in the meta-analysis). Studies with a Cook's distance larger than the median plus six times the interquartile range of the Cook's distances are influential. The rank correlation test and the regression test, using the standard error of the observed outcomes as predictor, are used to check for funnel plot asymmetry. The Knapp-Hartung correction was applied in meta-analyses including fewer than 10 studies to reduce the risk of underestimating standard errors and producing overly narrow confidence intervals.

3. Results

A total of 43 case-control studies with 670 bipolar patients, 520 healthy controls, and 280 patients with unipolar depression satisfied the criteria to be eligible for the meta-analysis [51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94]. The results of the extended meta-analysis are reported in Table 7. Parameters relating to heterogeneity, publication bias, the presence of outliers or particularly influential studies, and tests for possible funnel plot asymmetry are reported in Table 8. Figure 2 shows Forest plot of sleep parameter differences between drug-free BD and HC.

Figure 2.

Figure 3 shows Forest plot of sleep parameter differences between drug-free depressed BD and UD.

Figure 3.

Figure 4 shows Forest plot of sleep parameter differences between treated BD and HC.

Figure 4.

Figure 5 shows Forest plot of sleep parameter differences between pre-post treatment BD

Figure 5.

3.1. Sleep Duration and Continuity

The TST in drug-free BD patients, either in the depressive (369.24 ± 54.57 VS 403.39 ± 31.39; p=0.007) and manic (311.88 ± 84.76 VS 391.24 ± 42.93; p=0.001) patients were reduced compared to HCs. Unipolar drug-free depressive patients showed a reduced TST compared to drug-free BD depressive patients (372.28 ± 61.40 VS 348.66 ± 63.33; p=0.012). Pharmacological treatment appeared to normalize the TST in BD patients when compared to HCs (386.97 ± 73.72 VS 391.01 ± 66.82; p=0.712). The SOL of drug-free BD depressive patients did not significantly differ compared to HCs (32.53 ± 19.95 VS 19.49 ± 10.06; p=0.097) or drug-free UD patients (37.11 ± 22.09 VS 36.23 ± 13.71; 0.520). The SE of drug-free BD patients was reduced in depressive (81.22 ± 8.18 VS 88.60 ± 5.32; p=0.036) and manic (74.76 ± 14.66 VS 87.90 ± 5.20; p=0.034) patients compared to HCs, while it did not appear to differ from that of drug-free UDs (p=0.227). Pharmacological treatment appeared to normalize the SE of BD patients (p=0.169). The WASO parameter was not different in drug-free BD patients compared to HC or UD patients.

3.2. NREM Sleep

ST2 was reduced in drug-free BD manic patients compared to HCs (176.11 ± 95.00 VS 229.25 ± 41.00; p=0.002), DELTA and % DELTA were reduced in drug-free BD depressive patients compared to HCs, respectively (44.10 ± 34.00 VS 68.87 ± 30.33; p=0.003) and (11.98 ± 7.74 VS 16.86 ± 6.16; p=0.007). Drug treatment appeared to normalize the DELTA of treated BD patients (59.75 ± 35.64 VS 63.91 ± 32.15; p=0.946); comparison of pre- and post-treatment DELTA values ​​in BD patients showed higher post-treatment values ​​(37.01 ± 16.83 VS 48.84 ± 25.88; p=0.048). Treated BD patients had also increased ST1 (47.80 ± 30.73 VS 39.95 ± 21.53; p=0.009), and %ST1 (12.16 ± 8.54 VS 9.63 ± 5.21; p=0.002) compared to Healthy controls.

3.3. REM Sleep

REMT was reduced in drug-free BD manic patients (61.48 ± 29.62 VS 75.79 ± 18.29; p=0.011) compared to HCs. Drug-treated BD patients showed a reduced %REM compared to HCs (16.47 ± 6.66 VS 19.79 ± 7.04; p=0.020). REML in drug-free BD patients, compared to HCs, was reduced in manic patients (54.58 ± 15.35 VS 75.76 ± 20.75; p<0.001). Treated BD patients showed reduced %REM (16.47 ± 6.66 VS 19.79 ± 7.04; p=0.020) increased REM latency (108.43 ± 53.54 VS 90.27 ± 40.95; p=0.050) and increased REM density (12.06 ± 7.91 VS 7.37 ± 4.92; p=0.001), compared to healthy controls. Post-treatment values ​​showed increased REML in BD patients compared to pre-treatment values ​​(61.67 ± 31.46 VS 126.89 ± 61.83; p=0.002).

REMD was increased in drug-free BP manic patients compared to HCs (8.14 ± 5.07 VS 7.24 ± 0.44; p=0.048). Drug-free BD depressive patients had reduced REMD compared to drug-free UD patients (1.84 ± 0.50 VS 2.14 ± 0.49; p<0.001).

4. Discussion

Studies on polysomnographic EEG in bipolar patients have been frequently inconclusive, due to small sample size, with most of the studies referring to single case reports, clinical heterogeneity of patients, different phases of the illness and medication status.

A review (156) on EEG sleep alterations of bipolar patients across different clinical stages and of high-risk individuals, including a total of 22 studies, described as the most consistent findings across all stages of the disorder, increased sleep onset latency and increased REM density. Of the total sample included, six studies had patients during drug treatment and 14 studies during a medication-free period.

A recent meta-analysis (157) on sleep of bipolar patients across mood phases has reported that during depressive phase, patients had higher percentage of REM sleep, while in manic/mixed phase patients exhibited shorter total sleep time, lower sleep efficiency, and longer sleep onset latency. The study did not differentiate patients according to their drug status and included sleep parameters assessed by either actigraphy, polysomnography or sleep diary, for several of EEG sleep architecture variables, meta-analysis was not applicable, while for others the number of studies considered was between 2 and 3.

In our study, which included a total of 43 polysomnographic studies, alterations of sleep duration and of the different sleep stages were found in each phase of the bipolar disorder, but with some differences according to the phase of the illness.

Drug free bipolar depressive patients showed reduction of total sleep time, sleep efficiency, and of delta sleep duration and percentage compared to healthy controls. They also showed differences with drug-free patients with unipolar depression that had reduced total sleep time and increased REM density compared to patients with bipolar depression.

Reduction of SWS is a consistent feature in major depression (158), Borbely and Wirz Justice (159) have proposed that process S is deficient in depression and that rebound of SWS following sleep deprivation might restore SWS and improve depression.

Drug free manic patients showed reduction in total sleep time and sleep efficiency. A reduced need for sleep in manic patients was confirmed by both reduced stage 2 time and reduced REM time; they also had shortened REM latency and increased REM density. Early and seminal work on rapid cycling bipolar patients (111)described that the shifts to an elevated mood were accompanied by increased REM activity (number of REMs in the REM period) during the second half of the night while the shifts to depressed mood were associated with decreased REM activity during the second half of the night, furthermore the REM index (ratio of REM activity to REM time) was decreased on “depressive” nights. The presence of these REM alterations, suggestive of an increased activity of REM in the manic phase, might reflect enhanced cholinergic activity. possibly contributing or correlating to mood changes and switch mechanisms.

Treated bipolar patients showed no differences compared to healthy controls, except for reduced % REM, increased REM latency and REM density. Although most of the patients were on mood stabilizers, the use of antidepressants was not negligible in the sample analyzed, thus a reduced REM % and increased REM latency can be attributed to the well-known REM suppressing effect of antidepressants. Results of the comparison between pre and post treatment also showed increased REM latency following treatment.

REM density has been considered as a possible trait marker in depressed patients, also present in high-risk individuals. It has also been suggested that REM density may reflect sleep satiety (160), Aserinsky found that in extended sleep REM density increased with each successive REM episode, approaching a maximum value, with no further changes with additional sleep. According to Aserinsky, REM density reflects the satisfaction of a sleep need or the build-up of a pressure to awaken. Feinberg et al. (161) first proposed that REM density may also be related to level of arousal, with REM density being lower when sleep is deeper. A recent paper has proposed REM density as a physiological marker which indicate during sleep the “time to wake”. (162)

Patients with unipolar depression showed reduced total sleep time and increased REM density compared to patients with bipolar depression, suggesting that EEG sleep might serve as a possible tool to differentiate the two diagnostic categories. At clinical level unipolar patients show more frequently insomnia, whereas bipolar patients show either insomnia or (not restorative) hypersomnia. A recent meta-analysis (163) has reported that depression in unipolar disorder was associated with a shorter total sleep time and poorer sleep efficiency than patients diagnosed with bipolar depression.

Although the present study addressed all the available data in literature some aspects should be considered in interpreting the results. The meta-analyses concerning manic and euthymic bipolar disorder samples were based on a small number of datasets, reflecting limited availability of eligible studies and resulting in imprecise and unstable effect estimates. Consequently, statistical power was limited, with wide confidence intervals and reduced reliability of the pooled results. Several analyses also exhibited substantial heterogeneity, which constrains the robustness of quantitative inferences, although this is a frequent issue in all psychiatric samples. Also, the imputation of missing standard deviations, applied in a limited number of analyses, may have affected effect size precision and between-study variability, representing a potential source of bias. It should also be considered that a major limitation of studies assessing a biological variable in a clinical population is the medication status, drug treatment can clearly influence the measures addressed. Drug treatments in bipolar patients include mood stabilizers such as lithium and valproate, antidepressants, antipsychotics and benzodiazepine. All these classes of drugs have been shown to affect sleep. In the data analyzed, most of the drug free patients had a wash-out period of at least two weeks, with very few studies with a shorter wash-out period. Although two weeks is the gold-standard when studying a drug-free patient, a residual effect of the drugs cannot be excluded, especially when considering rebound effects on the sleep variables from withdrawal of previous treatment (i.e. REM rebound following REM suppressing effects from antidepressants).