Sleep and Circadian Rhythm Disturbance in Remitted Schizophrenia and Bipolar Disorder: A Systematic Review and Meta-analysis

Abstract Background Sleep and circadian rhythm disturbances in schizophrenia are common, but incompletely characterized. We aimed to describe and compare the magnitude and heterogeneity of sleep-circadian alterations in remitted schizophrenia and compare them with those in interepisode bipolar disorder. Methods EMBASE, Medline, and PsycINFO were searched for case–control studies reporting actigraphic parameters in remitted schizophrenia or bipolar disorder. Standardized and absolute mean differences between patients and controls were quantified using Hedges’ g, and patient–control differences in variability were quantified using the mean-scaled coefficient of variation ratio (CVR). A wald-type test compared effect sizes between disorders. Results Thirty studies reporting on 967 patients and 803 controls were included. Compared with controls, both schizophrenia and bipolar groups had significantly longer total sleep time (mean difference [minutes] [95% confidence interval {CI}] = 99.9 [66.8, 133.1] and 31.1 [19.3, 42.9], respectively), time in bed (mean difference = 77.8 [13.7, 142.0] and 50.3 [20.3, 80.3]), but also greater sleep latency (16.5 [6.1, 27.0] and 2.6 [0.5, 4.6]) and reduced motor activity (standardized mean difference [95% CI] = −0.86 [−1.22, −0.51] and −0.75 [−1.20, −0.29]). Effect sizes were significantly greater in schizophrenia compared with the bipolar disorder group for total sleep time, sleep latency, and wake after sleep onset. CVR was significantly elevated in both diagnoses for total sleep time, time in bed, and relative amplitude. Conclusions In both disorders, longer overall sleep duration, but also disturbed initiation, continuity, and reduced motor activity were found. Common, modifiable factors may be associated with these sleep-circadian phenotypes and advocate for further development of transdiagnostic interventions that target them.


Introduction
Difficulties in the initiation, maintenance, and timing of sleep are common complaints among individuals living with serious mental illness, and are associated with disability, distress, and poorer quality of life. [1][2][3] However, these symptoms often remain neglected, 4,5 and advancing their characterization and treatment are a priority. 6 In schizophrenia, poorer sleep quality, sleep onset and maintenance insomnia, and fragmented and irregular sleep have been described during periods of remission 7,8 and relapse. 9, 10 Sleep disruption has been associated with greater symptom severity, 7,8,11 and targeting sleepcircadian dysfunction may also ameliorate psychotic symptoms. 12 However, our understanding of sleep in schizophrenia lags behind that of many psychiatric disorders, and no previous meta-analyses of actigraphy data have been performed. The primary objective of this systematic review and meta-analysis of case-control actigraphy studies is therefore to gain a deeper understanding of sleep-circadian phenotypes in people with schizophrenia, who are treated with medication and in remission. Actigraphy enables rest-activity profiles to be captured across the entire 24-hour cycle under free-living conditions, over several days or weeks, and is particularly valuable for examining sleep and circadian variables concurrently. 13 By contrast, sleep problems have received greater attention in bipolar disorder, with previous meta-analyses of actigraphy studies demonstrating greater sleep latency, fragmentation and duration, and poorer sleep efficiency in patients compared with controls, that persist into the remission phase. [14][15][16] Given the ubiquity of sleep dysfunction across psychiatric disorders, 17,18 interest has grown in conceptualizing sleep as a transdiagnostic process, 19 with shared cognitive, neurobiological, and treatment mechanisms that may underpin both the sleep dysfunction and psychiatric disorder. By extension, interventions that target a transdiagnostic sleep process may demonstrate benefit across a range of psychiatric disorders. Psychopathology in schizophrenia and bipolar disorder overlap 20 -approximately 50% of those with bipolar disorder experience psychotic symptoms, 21 which together with evidence for shared genetic liability 22 and neurotransmitter dysfunction, 23 support a dimensional model. However, sleep parameters have not previously been compared between these disorders. Our second objective is therefore to comprehensively update previous meta-analyses of actigraphic parameters in bipolar disorder, and compare this with schizophrenia.
Finally, there is growing interest in the heterogeneity of sleep phenotypes in psychiatric disorders. For example, latent class analysis of sleep duration in schizophrenia 24 suggested that 3 distinct subtypes exist: those with short-, normal-, and long-sleep durations. A greater diversity of sleep phenotypes in patients would be expected to be reflected in greater variability of sleep parameters in patients than controls. To examine this hypothesis, we employed a novel meta-analytic approach, as recently applied to ecological data 25 and immune parameters in schizophrenia, 26 which examines variability by computing the mean-scaled coefficient of variation ratio (CVR) between groups. Greater variability in cases versus controls implies greater heterogeneity, suggesting the existence of subtypes of sleep pathology.
We hypothesized that effects in a comparable direction and magnitude would be found, which would argue for comparable sleep and circadian pathology across disorders, and stimulate the development of interventions targeting sleep and circadian dysfunction across the psychosis spectrum.

Methods
Analysis was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) 27 and Meta-Analysis of Observational Studies in Epidemiology (MOOSE) 28 guidelines, following an a priori protocol (supplementary information).

Study Selection
Two investigators (N.M. and T.P.) searched EMBASE, Ovid Medline, and PsycINFO databases independently for studies published in English, from inception until 18 November 2018, using the following search terms: ((schizophreni* OR schizoaffective OR psychosis OR psychotic) OR (bipolar OR manic OR mania)) AND (circadian OR diurnal OR actigraph* OR actimet* OR accelerometer). Specialist sleep journals, gray literature, and conference abstracts were also hand-searched. Abstracts were then screened independently by 2 investigators, and relevant full-text reports retrieved.
Inclusion criteria were as follows: (a) adults with schizophrenia-spectrum disorders (schizophrenia, schizoaffective disorder, delusional disorder) or bipolarspectrum disorders (bipolar disorder types I, II, or BD-NOS subtypes) meeting operationalized DSM-IV or ICD-10 criteria; (b) cases in nonacute phase of illness (defined below); (c) a healthy control group with no psychiatric or sleep disorder; (d) actigraphy or accelerometry recording for at least 24 hours, with description of methodology and analysis, reporting accepted actigraphic sleep, circadian, or motor activity parameters (see supplementary methods).
Studies in populations experiencing an illness episode (acute psychosis, mania, depression) and those in children or adolescents were excluded. In bipolar studies, ascertainment of remission/interepisode status either through validated instruments or clinical interview, was sought, and criteria are reported in table 1. In schizophrenia, although remission criteria have been proposed, 29 these are in less widespread use than in bipolar disorder, and selection of schizophrenia studies therefore relied on statements of stable/chronic psychosis, the absence of statements of relapse, and statements of stable treatment with antipsychotic medication.
It was decided a priori that any parameter reported by 2 or more studies would be eligible for meta-analysis. Six actigraphic sleep (time in bed; total sleep time; sleep latency; wake after sleep onset; awakenings; sleep efficiency) and 5 circadian (motor activity; amplitude; interdaily stability; intradaily variability; acrophase) parameters were included, as defined in figure 1.

Statistical Analysis
Due to the range of demographic characteristics and actigraphic devices used, it was expected that studies would be heterogeneous. To account for this, the standardized mean difference (SMD) and 95% confidence interval (CI) between case and control groups were calculated for schizophrenia and bipolar disorder groups using Hedges' g, 45 and a random-effects model fitted using a restricted maximum-likelihood estimator and inverse-variance method. Effect sizes between schizophrenia and bipolar disorder subgroups were then compared using a Waldtype test. Mean differences (MD) were also calculated for sleep duration parameters and reported in minutes.
Comparison of variability between the 2 groups was assessed by computing the relative mean-scaled coefficient of variation ratio (CVR) 25 Healthy unemployed controls from same local area, with no history of psychiatric illness.
Clozapine (2) Amisulpride (3) Olanzapine (7) Risperidone (3) Flupenthixol (1) Zuclopenthixol (1) Trifluoperazine (1) Combination (2) Actiwatch  are associated with greater variance. Estimates of relative variability between groups may therefore reflect betweengroup differences in the mean. Hence, the CVR quantifies variability differences while accounting for the differences in the mean by calculating the natural logarithm of the ratio of unbiased estimates of population coefficients of variation for each group. 25 After transforming back to a linear scale, a CVR of 1 indicates equal variability in patient and control groups, and a CVR of >1 would indicate greater relative variability in the patient groups. Distinction should be made between variability in within-individual night to night variability in actigraphic indices, as reported in some studies, 44,47,48 from group-level variability, as reported here. To investigate the influence of age, sex, and antipsychotic medications on outcomes, the mean age across schizophrenia and bipolar studies, the proportion of males in the patient groups, and the mean chlorpromazine equivalent for the total sample were computed using published estimates, 49,50 and the proportion of each study sample prescribed sedative antipsychotics (clozapine, olanzapine, quetiapine) was calculated. Random-effects meta-regression was then undertaken, using these data as the moderator variables.

Assessment of Study Quality, Inconsistency, and Publication Bias
Study quality was assessed by 2 independent investigators (N.M. and S.F.) using the Newcastle-Ottawa Scale (NOS) for case-control studies 51 (supplementary methods). Sensitivity analyses were performed excluding studies that (a) were deemed as poor quality using the NOS, (b) included a subset of participants not meeting full remission criteria, and (c) studied participants with a mean age > 50.
Inconsistency between studies was assessed using the I 2 statistic, 52 where values of >50% conventionally indicating moderate-high inconsistency, and <50% indicating lowmoderate inconsistency. Publication bias was assessed by testing for asymmetry of funnel plots using Egger's regression test 53 and by trim-and-fill imputation of missing studies, with recomputation of summary estimates based on imputed data.
All analyses were performed using the metafor package 54 in the R statistical programming language. 55 A 2-tailed P value of <.05 was deemed statistically significant.

Results
Three thousand five hundred and forty-five abstracts were screened, from which 93 full-text papers were assessed for inclusion (see PRISMA flow diagram, supplementary figure 1). Excluding overlapping studies, this yielded 15 schizophrenia and 15 bipolar studies (tables 1 and 2) for the quantitative synthesis comprising a total of 360 schizophrenia participants (and 319 healthy controls) and 607 bipolar participants (and 484 healthy controls).

Circadian variables
Motor activity = mean activity counts per 24-h period Relative amplitude = a measure of the delineation between periods of rest and activity. A higher amplitude is consistent with greater activity during day, and lower activity at night.
Interdaily stability = invariability of the 24-h rhythm between days, with a lower value indicating more stability between days Intradaily variability = fragmentation of the rhythm within a 24-h period, with a lower value indicating less fragmentation and a more consolidated, structured rhythm Acrophase = timing of peak of intensity of activity profile.

Sleep variables
Wake after sleep onset = minutes awake within sleep period   (2) Antidepressant (11) Antipsychotic (10) Hypnotic (2) No medication (1) Actiwatch-R AW2, nondominant wrist, Sleep duration, SE, SL, time awake.    , P = .0014, with no difference between groups. No significant mean differences in relative amplitude, interdaily stability, intradaily variability, and acrophase (average timing of activity peak) were found for either schizophrenia or bipolar disorder, nor was there a significant difference between groups. Findings are summarized in table 3 and figure 2, and forest plots for each variable are presented in supplementary figures 2-12.

Coefficient of Variation Ratio in Sleep and Circadian Parameters
After scaling for the mean, significantly elevated variability was found for total sleep time in both schizophrenia (CVR = 1. 43

Meta-regression of Chlorpromazine Equivalent, Sedative Medication, Age, and Sex
Calculation of chlorpromazine (CPZ) equivalence and the proportion of the sample on sedative antipsychotics were only possible for the schizophrenia studies, as none of the bipolar studies reported these data in adequate detail (supplementary table 1). Greater CPZ equivalent dose predicted significantly longer total sleep time (z = 2.74, P = .006), but also longer sleep latency (z = 2.19, P = .03). A greater proportion of the sample being prescribed sedative antipsychotic medication predicted shortened sleep latency (z = −2.15, P = .03); however, the test for total sleep time was nonsignificant. All other associations with medication, including motor activity, were nonsignificant. In both schizophrenia and bipolar groups combined, increasing age predicted decreasing relative amplitude Note: ***p < .001; **p < .01; *p < .05. a bold text = statistical significance at P < .05 level.

Study Quality and Sensitivity Analyses
Compared with studies in bipolar disorder, schizophrenia studies tended to be of lower quality, with shortcomings around the representativeness, selection, definition, and comparability of control groups. Six studies (including 2 using the same data set) were rated as being of poor quality using the Newcastle-Ottawa Scale (supplementary  table 3). However, excluding these studies in a sensitivity analysis did not significantly alter findings with respect to the analyses of mean differences (supplementary table 4 and supplementary figure 13), except for the comparison of effect size for sleep latency between schizophrenia and bipolar disorder no longer reaching statistical significance. Repeating the CVR analysis after removing these studies resulted in total sleep time and sleep efficiency in schizophrenia studies no longer reaching statistical significance (supplementary figure 14).
Excluding 3 schizophrenia studies 36-38 with mean participant age >50 resulted in acrophase reaching statistical significance in this group (SMD [95% CI] = 0.47 [0.05, 0.89], P < .05), but did not alter outcomes for the other variables. Excluding the 2 bipolar disorder studies that included a subset of participants not meeting age 39 or remission criteria 58 did not significantly alter mean differences in sleep and circadian parameters, except for sleep efficiency in the bipolar disorder group no longer reaching statistical significance. After excluding these studies in the Coefficient of Variation Ratio analysis, sleep efficiency was significantly more variable in the bipolar disorder group (CVR = 1.32 [1.09, 1.59], P = .004), and variability in relative amplitude in both disorders no longer reached statistical significance.

Inconsistency and Publication Bias
Study inconsistency was in the moderate-high range in the schizophrenia studies in comparison to bipolar disorder meta-analyses, which were mostly in the low-moderate range (supplementary table 5). Assessment of publication bias using Egger's test indicated the possibility of bias in 3 bipolar disorder variables; however, in these cases, trimand-fill analysis did not impute any potentially missing  mentary table 6).

Discussion
In individuals with treated, remitted schizophrenia, a sleep-circadian phenotype characterized by longer total sleep time and time in bed, longer sleep latencies, elevated wake after sleep onset, and decreased motor activity was observed. This phenotype was comparable with that of remitted bipolar disorder, although effect sizes were generally greater in schizophrenia compared with bipolar disorder, and significantly so for total sleep time, sleep latency, and wake after sleep onset. Our third main finding was that group-level variability in total sleep time, time in bed, and relative amplitude was significantly elevated in both disorders, indicating greater heterogeneity in these parameters in schizophrenia and bipolar populations compared with controls.
Sleep disturbance is an intrinsic feature of many psychiatric disorders, and previous meta-analyses of polysomnographic studies have suggested similar disturbances are observed across a range of disorders. 17,18 However, these analyses did not include bipolar disorder, nor have actigraphic studies been meta-analyzed previously in a transdiagnostic context. Polysomnography offers the advantage of accurate assessment of sleep stages, but is in general conducted in a sleep-clinic environment and usually for only one night. Actigraphy allows objective assessments of sleep over longer periods in the home environment. Our analysis provides a first comparison of actigraphically assessed sleep parameters in schizophrenia and bipolar disorder, and extends our understanding of sleep and circadian disturbances in these disorders by showing common features of actigraphic sleep and circadian dysregulation. Taken together, these point to interacting disturbances in sleep initiation, maintenance, and daytime activity levels.

Interpretation
The elevated sleep latency and wake after sleep onset suggest that in both disorders, patients took significantly longer than controls to fall asleep and experienced more fragmented, poorly consolidated rest periods, consistent with reduced sleep propensity. This aligns with studies demonstrating difficulties with sleep initiation and maintenance in 36%-44% in schizophrenia, 8 and 55% of patients with remitted bipolar disorder, 60 which may be related to several factors. First, residual psychotic symptoms including hallucinations and paranoia, and subthreshold symptoms of mania and depression, interfere with sleep induction. In turn, accumulating evidence suggests that disrupted sleep precipitates and maintains affective and psychotic symptoms, in a mutually reinforcing cycle. 69,70 Second, as supported by the finding of reduced daytime motor activity, sleep pressure may be attenuated as a result of naps and daytime inactivity, 60 which together with circadian misalignment (particularly delayed sleep phase 71 ) can compromise sleep initiation and maintenance. Third, dysfunctional attitudes about sleep have been reported in schizophrenia 72 and were robustly associated with greater severity of sleep disturbance in bipolar disorder. 60 However, it is important to recognize that although disruption in sleep initiation and continuity variables was statistically significant in both disorders, mean differences were small and less likely to be of clinical significance, at group level, in bipolar disorder.
A second important finding was that, despite evidence for difficulties in sleep initiation and maintenance, which is consistent with insomnia, total time spent asleep and in bed were also significantly increased with large effect sizes in both disorders, consistent with hypersomnia. Though definitions are imprecise, hypersomnia is receiving increasing attention as a core sleep phenotype in bipolar disorder, 73 but remains under-recognized and rarely addressed in schizophrenia. 24 Importantly, the elevated sleep duration parameters in schizophrenia argue that this phenotype may be equally, if not more prevalent, in schizophrenia. Causes of hypersomnia are poorly understood, but may include greater circadian predisposition for long-sleep duration 74 ; reduced drive for wakefulness arising from longer time in bed; anergia and fatigue associated with depression and the negative symptom dimension 75 ; fewer scheduled daytime occupational and social activities, and use of sleep as a means to escape from distressing symptoms. 5 Additionally, firstand second-generation antipsychotics bind to sleep-wake regulating receptor families, and increase sleep duration and continuity in patients and controls. 76 Agents such as clozapine and olanzapine have a particularly pronounced sleep-promoting and consolidating action 32,76 : the significantly greater sleep duration in schizophrenia in comparison to bipolar disorder may therefore follow from the more frequent use of sedative antipsychotics, and at higher doses. 77 The medication meta-regression provided support for this hypothesis, where higher doses of antipsychotic were associated with increased total sleep time, and greater sedative antipsychotic prescription predicted reduced sleep latency. Mood stabilizers are less sedative, although lithium has been shown to lengthen and slow down the circadian period, 78 increasing sleep duration.
Reduced mean motor activity over the 24-hour period in patients, and the associated trend toward an attenuation of the relative amplitude in both disorders, signifies a flattening of the rest-activity profile. This is consistent with an interplay between the 2 contrasting phenotypes described above: decreased daytime activity (consistent with greater sedentary behavior and the effects of sedative medication) and increased activity within the main sleep episode (consistent with fragmented sleep). A vicious cycle can therefore be established, where a reduced drive for wakefulness, secondary to the factors discussed in the previous paragraph, leads to longer time in bed, longer total sleep time, but also reduced sleep propensity, which in turn drives elevated sleep latency and sleep fragmentation (supplementary figure 16). Such a sleep phenotype is consistent with clinical experience and has been induced in volunteers in laboratory studies, where extension of the sleep opportunity promotes not only greater total sleep time, but also longer sleep latency and poorer sleep efficiency. 79 The elevated variability in sleep duration parameters suggests that in addition to the overall group effect toward longer mean sleep times in both disorders, greater heterogeneity with respect to sleep duration is also observed in clinical populations. This concurs with recent studies demonstrating subtypes differing with respect to sleep duration in schizophrenia patients with insomnia symptoms 24 and in bipolar disorder patients with hypersomnia, 73,80 and suggests that insomnia and hypersomnia-type patterns can coexist in some individuals.

Implications
First, these results suggest the presence of common signatures of sleep-circadian dysfunction in schizophrenia and bipolar disorder, and advocate for the development of transdiagnostic interventions that target core difficulties, particularly with sleep initiation, maintenance, and hypersomnia. Interventions that address this objective are emerging: Cognitive Behavioural Therapy for insomnia has been adapted for bipolar disorder, 81 schizophrenia, 82 and transdiagnostically in severe mental illness. 83,84 However, they are yet to be widely established in clinical practice. The present findings also indicate that hypersomnia is common yet under-recognized, and merit greater attention in both clinical and research contexts, particularly in light of its association with elevated risk of relapse in bipolar disorder. 65,85 Second, we suggest that a subgroup of individuals with serious mental illness experience diverse sleep disturbances that include both "insomnia" and "hypersomnia" phenotypes, which can be conceptualized as a dynamic interaction between the drive for wakefulness and drive for sleep, which are differentially influenced by a range of factors associated with the psychiatric disorder and its treatment. Clinicians managing both schizophrenia and bipolar disorder should maintain awareness of the sleep or wake-promoting effects of different psychotropic medications and consider how these can be harnessed in tailoring treatments to each patient's sleep-circadian phenotype.
Finally, sleep-circadian disturbances have been associated with cognitive dysfunction, 86 psychotic and affective symptoms, 69 relapse, 87 and suicidality. 88 Actigraphy may therefore serve not only as a trait marker for quantifying stability of remission and risk of adverse outcomes, but also as a state marker for predicting dynamic changes in mental state, including relapse. Novel approaches to the longitudinal measurement of rest-activity profiles over remission and relapse in schizophrenia are currently being explored. 89 Increasing evidence suggests that shortened, interrupted, and misaligned sleep disrupts cognitive and neurobiological systems that also underlie psychotic and affective phenomena 19 ; sleep and circadian disruption therefore represents a valuable mechanism for understanding and treating these disorders.

Strengths and Limitations
This is the first study of which we are aware that examines actigraphic sleep parameters in both schizophrenia and bipolar disorder, and also that investigates variability using the CVR. It includes a large number of studies, mostly from populations living in their home environment, and uses standardized measures that allow comparison of data from different actigraphic devices.
Some limitations must also be acknowledged. Although most studies attempted to match for age and sex, a key consideration remains the selection of cases and control groups. First, the majority of patients with schizophrenia, and to a lesser extent bipolar disorder, are unemployed, 90 and lead markedly different daily schedules to employed controls. However, only one study 44 explicitly accounted for this by selecting unemployed controls. Second, some schizophrenia studies examined exclusively inpatient, or mixed inpatient and outpatient groups, which limits comparability with studies examining outpatient populations that are not governed by ward schedules. Third, remission and relapse are not as clearly operationalized in schizophrenia compared with bipolar disorder, and schizophrenia populations may therefore manifest greater residual psychopathology and sleep disturbance. Future studies should therefore define clinical status more stringently, using accepted criteria. Fourth, primary sleep disorders that interfere with sleep initiation and maintenance including obstructive sleep apnea, nightmares, and restless leg syndrome are over-represented in schizophrenia 91,92 and bipolar disorder, 93 yet were only screened for in a small number of studies, potentially biasing findings. Finally, the sedative effects of many antipsychotic medications likely explain a significant proportion of the longer sleep duration found in schizophrenia. However, it is important to note that schizophrenia patients nonetheless experience poorer sleep continuity, as evidenced by greater sleep latency and wake after sleep onset, in comparison to bipolar patients. This suggests a greater overall degree of sleep disturbance in schizophrenia in comparison to bipolar disorder, which is not resolved by the sleep-promoting effects of many antipsychotic agents. Some less sedating antipsychotic agents may in some circumstances contribute to this disturbance, and future research should focus on understanding this relationship.
There were also a number of broader methodological limitations. Although actigraphy has been validated in populations with schizophrenia 94 and bipolar disorder, 62 its specificity for sleep remains low, and the tendency to overestimate sleep time may be significant in individuals with sedentary behaviors. Few studies reported circadian variables such as the timing of the rest-activity cycle, nocturnal sleep versus daytime naps, or nonparametric measures, limiting the statistical power of meta-analysis of these important metrics. In view of evidence for late chronotype 95 and phase delay in the rest-activity and melatonin rhythms, 44,58 future studies should report circadian parameters, where possible comparing these with endogenous markers of circadian phase. Finally, only one 68 of the bipolar disorder studies reported the proportion of patients with a history of psychosis. A comparison between bipolar disorder I patients with psychosis and schizophrenia patients would be an interesting question for future comparative studies across the psychosis spectrum.

Conclusions
Individuals with schizophrenia and bipolar disorder in the remission phase demonstrate sleep-circadian dysfunction that is characterized by both greater sleep latency and fragmentation, but also an increase in sleep duration. In some individuals, insomnia and hypersomnia-type patters may overlap and maintain one another. Further development of transdiagnostic interventions that assess and target core dimensions of sleep-circadian disturbance is a priority.

Supplementary Material
Supplementary data are available at Schizophrenia Bulletin online. acted as consultant and received research support from several pharmaceutical companies, but this work was not related to bipolar disorder or schizophrenia.