Introduction
Neuropsychiatric diseases are significant causes of disability and death throughout the world (GBD 2016 Neurology Collaborators, 2019; GBD 2015 Neurological Disorders Collaborator Group, 2017; Vigo, Thornicroft, & Atun, Reference Vigo, Thornicroft and Atun2016) and they take a large toll on individuals, families and health-care systems (GBD 2019 Diseasesand Injuries Collaborators, 2020; The Lancet, 2017). Sleep disturbances are frequent complaints in patients with neuropsychiatric diseases. Historically, sleep disturbances were viewed as clinical symptoms which result from the pathology of neuropsychiatric diseases. However, increasing evidence suggests a complex inter-relationship and potential bidirectional causality between sleep disturbances and these diseases (Krystal, Reference Krystal2020). Sleep disturbances longitudinally predict the development of psychiatric diseases and neurological disorders (i.e. in depression, anxiety, and neurodegeneration) (Galbiati, Verga, Giora, Zucconi, & Ferini-Strambi, Reference Galbiati, Verga, Giora, Zucconi and Ferini-Strambi2019; Hertenstein et al., Reference Hertenstein, Feige, Gmeiner, Kienzler, Spiegelhalder, Johann and Baglioni2019; Shi et al., Reference Shi, Chen, Ma, Bao, Han, Wang and Lu2018). Some treatments for sleep disturbances improve the symptoms of neuropsychiatric conditions, and vice versa, treating neuropsychiatric diseases may also affect sleep (Krystal, Reference Krystal2020). These findings suggest that clarifying the relationships between sleep and neuropsychiatric diseases may be helpful for understanding the pathology of the diseases and for improving their clinical management (Krystal, Reference Krystal2020).
Polysomnography (PSG) is the gold standard method for objectively assessing sleep features in clinical and non-clinical settings. PSG measured sleep reflects neurophysiological functioning in humans. For instance, evidence supports slow wave sleep's (SWS) role in energy restoration, clearing metabolites, hormone release, immunity, and memory consolidation (Leger et al., Reference Leger, Debellemaniere, Rabat, Bayon, Benchenane and Chennaoui2018). Rapid eye movement (REM) sleep helps maintain neuronal homeostasis in the brain as disturbances of REM sleep can affect brain excitability, synaptic pruning, and neurogenesis, and loss of REM sleep can lead to neurodegeneration (Chauhan & Mallick, Reference Chauhan and Mallick2019). Thus, investigating and comparing PSG sleep variables across neuropsychiatric diseases has the potential to reveal neurobiological mechanisms of specific disorders and to reveal neural commonalities and differences that may help refine diagnostic categories and may have implications for more effective clinical management (Baglioni et al., Reference Baglioni, Nanovska, Regen, Spiegelhalder, Feige, Nissen and Riemann2016a).
Many case–control studies have reported various PSG changes for different neuropsychiatric diseases, and meta-analyses of PSG changes in some neuropsychiatric diseases have been published. Meta-analytic approaches are typically considered as the highest rank of evidence and can provide a more accurate ‘big picture’ for disease characteristics. However, they can also introduce confusion into the literature due to the low methodological standards of some published meta-analyses and, perhaps more importantly, of their included studies (Solmi, Correll, Carvalho, & Ioannidis, Reference Solmi, Correll, Carvalho and Ioannidis2018). Thus, poorly conducted meta-analytic studies with their potentially flawed findings may obscure rather than clarify the state of science for a particular question (Ioannidis, Reference Ioannidis2016; Ioannidis, Reference Ioannidis2017). Specifically, meta-analyses are susceptible to reporting bias, publication bias, and residual confounding bias, and other types of problems which can result in inflated estimates (Ioannidis, Reference Ioannidis2008) or false positives (Ioannidis, Reference Ioannidis2005) for examined data parameters. These types of flaws have resulted in an excess of significant associations (p < 0.05) in psychological science and other medical fields (Boffetta et al., Reference Boffetta, McLaughlin, La Vecchia, Tarone, Lipworth and Blot2008; Ioannidis, Munafo, Fusar-Poli, Nosek, & David, Reference Ioannidis, Munafo, Fusar-Poli, Nosek and David2014) that may have obscured the most important or distinguishing characteristics for a given disorder. Thus, it is important to comprehensively evaluate evidence from meta-analyses to minimize such quality concerns (Ioannidis, Reference Ioannidis2009, Reference Ioannidis2016).
An umbrella review, which summarizes, assesses, and grades the findings of multiple meta-analyses, is a standardized and systematic collection of data from studies on a specific topic (Fusar-Poli, Hijazi, Stahl, & Steyerberg, Reference Fusar-Poli, Hijazi, Stahl and Steyerberg2018; Ioannidis, Reference Ioannidis2009). This approach to data review allows a higher-level synthesis of the evidence and a better recognition of the uncertainties, weaknesses, various kinds of bias, and strengths of the available evidence (Bougioukas et al., Reference Bougioukas, Bouras, Apostolidou-Kiouti, Kokkali, Arvanitidou and Haidich2019). Compared with the meta-analytic approach, which is usually restricted to one single topic, umbrella reviews have advantages because they can examine evidence across a broad and high-quality database and provide a comprehensive overview of a specific topic (Aromataris et al., Reference Aromataris, Fernandez, Godfrey, Holly, Khalil and Tungpunkom2015; Ioannidis, Reference Ioannidis2009). This capability has led to an increasing emphasis being placed to umbrella reviews to best address the extensive literature of complex neuropsychiatric science and other medical fields (Barbui et al., Reference Barbui, Purgato, Abdulmalik, Acarturk, Eaton, Gastaldon and Thornicroft2020; Hailes, Yu, Danese, & Fazel, Reference Hailes, Yu, Danese and Fazel2019; Ioannidis, Reference Ioannidis2017).
To our knowledge, to date, no umbrella review has been conducted on the topic of PSG changes in neuropsychiatric diseases. Given the role that sleep plays in essentially all these diseases, such a review may provide unique insight into sleep changes across diseases. Therefore, we performed this first umbrella review of relevant meta-analyses of case–control studies and attempted to provide a comprehensive overview and examination of the strength of evidence, precision of the estimates, presence of biases, and robustness of the published PSG changes in patients with neuropsychiatric diseases compared with healthy controls (HCs).
Methods
This umbrella review was done following the PRISMA reporting guidelines (Moher, Liberati, Tetzlaff, & Altman, Reference Moher, Liberati, Tetzlaff and Altman2009) and its protocol was registered (PROSPERO ID: CRD42020202318).
Search strategy, study selection, and eligibility criteria
The following terms were searched for in abstract or title: (‘meta-analy*’ or ‘metaanaly*’ or ‘meta-analysis’ or ‘meta analy*’) AND (‘polysomnogra*’ OR ‘PSG’ OR ‘sleep architect*’ OR ‘sleep monit*’ OR ‘sleep stage*’ OR ‘electroencephalogra*’ OR ‘EEG’). The detailed search strategies used for each literature database are provided in online Supplementary Tables S1–S5. We initially searched MEDLINE, EMBASE, PsycINFO, and CINAHL, and All EBM databases from inception to 26 Nov 2020, to identify systematic reviews and meta-analyses of case–control studies exploring PSG changes in patients with neuropsychiatric diseases compared with non-neuropsychiatric HCs. We updated the literature search using the same search strategies on 28 Mar 2022, to find any newly published meta-analyses. Two investigators (YZ and RR), with a good inter-rater agreement for potentially eligible studies (Kappa = 0.837), independently selected the potential eligible articles. The references of relevant studies were manually screened to identify eligible articles. Any disagreements were discussed by three authors (YZ, RR, and XDT) to reach a final decision.
The included studies meet the following eligibility criteria: (1) the participants were patients with mental illnesses (including but not limited to depression, generalized anxiety disorder, schizophrenia, bipolar disorder, etc.) or neurological diseases [including but not limited to stroke, epilepsy, Parkinson's disease (PD), Huntington's disease (HD), etc.]. The diagnosis of mental illnesses was according to any edition of the Diagnostic and Statistical Manual of Mental Disorders or International Classification of Diseases criteria or a structured psychiatric diagnostic interview. The diagnosis of neurological disease was also according to established criteria (e.g. diagnosing PD according to Brain Bank criteria); (2) differences in PSG parameters (i.e. total sleep time (TST), wake time after sleep onset, sleep efficiency (SE), sleep latency (SL), and percentage of N1, N2, SWS and REM sleep, REM latency, periodic limb movement index, apnea hypopnea index, arousal index, cyclic alternating pattern (CAP) parameters, or power spectral data) between patients with neuropsychiatric diseases and non-neuropsychiatric HCs were explored by meta-analysis. The eligible articles were published in peer-reviewed journals with no language restrictions. The exclusion criteria are provided on online Supplementary Appendix pp3.
Data extraction
Data extraction was done independently by two investigators (YZ and RR) with a high inter-rater percentage agreement (99.5%). In the case of discrepancies, three investigators (YZ, RR and XDT) discussed the concerns and made the final decision. From each eligible article, we recorded the first author, year of publication, disease names, and number of comparisons included. If a quantitative synthesis was done, we extracted the study-specific estimated effect size of differences in PSG parameters between cases and HCs together with their corresponding 95% confidence intervals (CIs) and the number of cases and HCs in each study. If the eligible article only reported the pooled effect sizes and did not report the study-specific effect size, we extracted the study-specific effect size from the included individual component studies of each eligible article and then re-estimated their effect sizes. In one eligible article (Cox & Olatunji, Reference Cox and Olatunji2020) which integrated various PSG parameters into three variables (sleep continuity, sleep depth, and REM pressure) but did not report detailed data on sleep continuity and sleep architecture (i.e. TST, SL, SE, N1, N2, SWS, and REM sleep), we also extracted the study-specific effect size from the individual component studies. Metrics followed those of the original meta-analyses [i.e. mean difference, standardized mean difference (SMD), or Hedge's g].
Quality assessments
AMSTAR 2 (A Measurement Tool to Assess Systematic Reviews), which has good inter-rater agreement, content validity, and test-retest reliability was used to assess the methodological quality of the meta-analyses (Shea et al., Reference Shea, Reeves, Wells, Thuku, Hamel, Moran and Henry2017). The domains which AMSTAR 2 evaluates and the detailed methods for use of AMSTAR 2 are provided on online Supplementary Appendix pp6. Two reviewers (YZ and RR) independently used AMSTAR 2 to assess the meta-analyses and the inter-rater agreement was good (Kappa = 0.82). Any disagreements were discussed by three authors (YZ, RR, and XDT) to reach a final decision.
Data analysis
Summary SMDs with 95% CI were re-estimated using common metric random effects methods (DerSimonian & Laird, Reference DerSimonian and Laird1986). The heterogeneity between studies was evaluated using Cochran's Q test (Cochran, Reference Cochran1954) and the I 2 statistic (I 2 > 50% indicates high heterogeneity) (Higgins, Thompson, Deeks, & Altman, Reference Higgins, Thompson, Deeks and Altman2003). We estimated the 95% prediction interval, the range in which we expect the PSG differences between groups will lie for 95% of future studies (Higgins, Thompson, & Spiegelhalter, Reference Higgins, Thompson and Spiegelhalter2009).
We noted when prediction intervals excluding the null value (0 in the case of SMDs) suggest that the statistically significant PSG changes in patients with neuropsychiatric diseases are likely to persist in future studies. We assessed whether there was evidence for small-study effects (i.e. whether smaller studies tend to give substantially larger estimates of effect size compared with larger studies) with the regression asymmetry test proposed by Egger et al. (Egger, Davey Smith, Schneider, & Minder, Reference Egger, Davey Smith, Schneider and Minder1997). A p value less than 0.1 occurring in conjunction with more conservative effect sizes in larger studies compared with that found in the in random effects meta-analysis was judged to be evidence for small-study effects.
We evaluated the existence of excess significance bias to examine whether the observed number of studies with statistically significant results (positive studies, p < 0.05) in each meta-analysis was larger than their expected number (Ioannidis & Trikalinos, Reference Ioannidis and Trikalinos2007). For each meta-analysis, the expected number was calculated as the sum of the statistical power estimates for each study in the meta-analysis. The power of each original case–control study was calculated by an algorithm using a non-central t distribution (Lubin & Gail, Reference Lubin and Gail1990), which is necessary for evaluating excess significance bias. The estimated power depends on the plausible SMD. Because the true SMD for any meta-analysis is unknown, we assumed that the most plausible effect is given by the largest study (smallest standard error) (Ioannidis, Reference Ioannidis2013). Excess significance bias for each meta-analysis was determined at a p value less than 0.10 (Ioannidis & Trikalinos, Reference Ioannidis and Trikalinos2007).
Statistical analyses were conducted using Comprehensive Meta-Analysis software version 2.0 and STATA version 14.0. Power calculations were done in R version 3.5.1 and the pwr package. All p values were two tailed.
Credibility of evidence
As with earlier umbrella reviews (Barbui et al., Reference Barbui, Purgato, Abdulmalik, Acarturk, Eaton, Gastaldon and Thornicroft2020; Belbasis, Bellou, Evangelou, Ioannidis, & Tzoulaki, Reference Belbasis, Bellou, Evangelou, Ioannidis and Tzoulaki2015; Kim et al., Reference Kim, Son, Son, Radua, Eisenhut, Gressier and Fusar-Poli2019, Reference Kim, Kim, Lee, Jeong, Lee, Lee and Fusar-Poli2020), we classified the strength of PSG changes in each neuropsychiatric disease as convincing (class I), highly suggestive (class II), suggestive (class III), weak (class IV) or not significant (NS). Convincing evidence required p values in random effects models below 10−6, number of cases > 1000, the largest study nominally significant (p < 0.05), no evidence of small-study effects, no large heterogeneity (i.e. I 2 < 50%), no evidence of excess of significance bias, and 95% prediction intervals not including the null value. Highly suggestive evidence required p values < 10−6, number of cases > 1000, and the largest study nominally significant (p < 0.05). Suggestive evidence required p values < 10−3 and number of cases > 1000. Weak evidence required no specific number of cases and p < 0.05. For PSG comparisons classified as convincing, highly suggestive, or suggestive, we attempted further assessment for the robustness of the evidence by subset analyses limited to individual component studies that excluded patients taking medications impacting sleep, studies excluding patients with other psychiatric comorbidities, and studies using different PSG scoring methods [Rechtschaffen and Kales (R&K) v. American Academy Sleep Medicine (AASM)].
Results
Study selection
Our search identified 3537 publications. After removing duplicates and screening titles and abstracts, 64 full-text articles were assessed for eligibility. Twenty-seven systematic reviews (Baglioni et al., Reference Baglioni, Regen, Teghen, Spiegelhalder, Feige, Nissen and Riemann2014, Reference Baglioni, Nanovska, Regen, Spiegelhalder, Feige, Nissen and Riemann2016a, Reference Baglioni, Nissen, Schweinoch, Riemann, Spiegelhalder, Berger and Sterr2016b; Bertrand et al., Reference Bertrand, d'Ortho, Reynaud, Lejoyeux, Bourgin and Geoffroy2021; Biancardi, Sesso, Masi, Faraguna, & Sicca, Reference Biancardi, Sesso, Masi, Faraguna and Sicca2021; Chan, Chung, Yung, & Yeung, Reference Chan, Chung, Yung and Yeung2017; Chen et al., Reference Chen, Liu, Wu, Xuan, Zhao and Sun2021; Cox & Olatunji, Reference Cox and Olatunji2020; D'Rozario et al., Reference D'Rozario, Chapman, Phillips, Palmer, Hoyos, Mowszowski and Naismith2020; Díaz-Román, Hita-Yanez, & Buela-Casal, Reference Díaz-Román, Hita-Yanez and Buela-Casal2016; Keenan, Sherlock, Bramham, & Downes, Reference Keenan, Sherlock, Bramham and Downes2021; Lugo et al., Reference Lugo, Fadeuilhe, Gisbert, Setien, Delgado, Corrales and Ramos-Quiroga2020; Mantua et al., Reference Mantua, Grillakis, Mahfouz, Taylor, Brager, Yarnell and Simonelli2018; Ng et al., Reference Ng, Chung, Ho, Yeung, Yung and Lam2015; Plante, Reference Plante2018; Stanyer, Creeney, Nesbitt, Holland, & Hoffmann, Reference Stanyer, Creeney, Nesbitt, Holland and Hoffmann2021; Winsor et al., Reference Winsor, Richards, Bissell, Seri, Liew and Bagshaw2021; Winsper et al., Reference Winsper, Tang, Marwaha, Lereya, Gibbs, Thompson and Singh2017; Xu et al., Reference Xu, Deng, Qin, Vgontzas, Basta, Xie and Li2020; Yeh et al., Reference Yeh, Lin, Li, Chien, Wu, Liou and Hsu2022a, Reference Yeh, Lin, Li, Chien, Wu, Liou and Hsu2022b; Zhang et al., Reference Zhang, Ren, Sanford, Yang, Zhou, Zhang and Tang2019a, Reference Zhang, Ren, Yang, Zhou, Li, Shi and Tang2019b, Reference Zhang, Ren, Sanford, Yang, Zhou, Tan and Tang2020a, Reference Zhang, Ren, Yang, Zhang, Shi, Sanford and Tang2021, Reference Zhang, Ren, Yang, Zhang, Shi, Okhravi and Tang2022; Zhang, Ren, Yang, Sanford, & Tang, Reference Zhang, Ren, Yang, Sanford and Tang2020b), including 465 case–control studies, met inclusion criteria (Fig. 1). Details of the reviews excluded, and the reasons for exclusion, are provided in online Supplementary Table S6.
Flow chart of literature search.
Description of the included systematic reviews and meta-analyses
From these 27 included systematic reviews, we extracted information on 321 pooled analyses exploring sleep macrostructure changes in 27 neuropsychiatric diseases compared with HCs (Table 1). Of the 321 pooled analyses of sleep macrostructure, there were 10 on schizophrenia, 9 on bipolar disorder, 12 on major depressive disorder (MDD), 10 on generalized anxiety disorder, 9 on obsessive compulsive disorder, 10 on panic disorder, 6 on social anxiety disorder (SAD), 10 on borderline personality disorder, 9 on insomnia, 10 on adult attention deficit hyperactivity disorder (ADHD), 12 on childhood ADHD, 9 on adult autism spectrum disorder (ASD), 10 on childhood ASD, 8 on anorexia nervosa, 10 on posttraumatic stress disorder (PTSD), 10 on stroke, 12 on mild cognitive impairment, 11 on traumatic brain injury, 12 on idiopathic REM sleep behavior disorder, 11 on idiopathic hypersomnia, 12 on HD, 13 on PD, 12 on Wilson's disease (WD), 12 on narcolepsy, 12 on Alzheimer's disease, 7 on seasonal affective disorder, 11 on adult migraine, 11 on child migraine, 10 on child and adolescent epilepsy, 12 on adult epilepsy, and 9 on persistent tic disorder. The 321 pooled analyses of sleep macrostructure were based on 27 neuropsychiatric diseases, 191 061 total participants, a median 135 neuropsychiatric cases per pooled analysis (interquartile range (IQR) 77–355, range 27–1663), and a median 285 total participants per pooled analysis (IQR 152–862, range 50–2975). As shown in Fig. 2, the overall patterns of sleep changes varied widely across different diseases. Furthermore, there were a total of 35 pooled analyses exploring sleep microstructure changes (CAP parameters) within three neuropsychiatric diseases (7 on narcolepsy, 23 on ADHD, and 5 on epilepsy; see descriptions in online Supplementary Table S7). The means for polysomnographic parameters in patients with neuropsychiatric diseases and HCs are provided in online Supplementary Table S8. The quality assessments of included systematic reviews and meta-analyses are provided on online Supplementary Appendix pp6.
Change patterns (standardized mean differences) of sleep parameters in 27 neuropsychiatric diseases. Abbreviations of disease names: AD, Alzheimer's disease; ADHD, attention deficit hyperactivity disorder; ASD, autism spectrum disorder; BPD, borderline personality disorder; GAD, generalized anxiety disorder; HD, Huntington's disease; IH, idiopathic hypersomnia; iRBD, idiopathic rapid eye movement sleep behavior disorder; MCI, mild cognitive impairment; MDD, major depressive disorder; OCD, obsessive compulsive disorder; PD, Parkinson's disease; PTD, persistent tic disorder; PTSD, posttraumatic stress disorder, SAD, social anxiety disorder; TBI, traumatic brain injury, WD, Wilson's disease. Abbreviations for sleep parameters: AHI, apnea hypopnea index; AI, arousal index; PLMI, Periodic limb movement index; REM, rapid eye movement sleep; REMD, rapid eye movement sleep density REML, rapid eye movement sleep latency; SE, sleep efficiency; SL, sleep latency; SWS, slow wave sleep; TST, total sleep time; WASO, wake time after sleep onset.
Characteristics, quantitative synthesis, and bias assessment of the eligible articles
Abbreviations of disease names: AD, Alzheimer's disease; ADHD, attention deficit hyperactivity disorder; ASD, autism spectrum disorder; BPD, borderline personality disorder; GAD, generalized anxiety disorder; HD, Huntington's disease; IH, idiopathic hypersomnia; iRBD, idiopathic rapid eye movement sleep behavior disorder; MCI, mild cognitive impairment; MDD, major depressive disorder; OCD, obsessive compulsive disorder; PD, Parkinson's disease; PTD, persistent tic disorder; PTSD, posttraumatic stress disorder, SAD, social anxiety disorder; TBI, traumatic brain injury, WD, Wilson's disease. Other non-disease related abbreviations: AMSTAR 2, A Measurement Tool to Assess Systematic Reviews; ESB, excess significance bias (which was used to examine whether the observed number of studies with statistically significant results (positive studies, p < 0.05) in each pooled analysis was larger than their expected number); LS, largest study with significant effect (which was used to reflect whether the largest study (study with smallest standard error) of a pooled analysis attend a statistically significant level). NA, applicable. Abbreviations for sleep parameters: AHI, apnea hypopnea index; AI, arousal index; PLMI, Periodic limb movement index; REM, rapid eye movement sleep; REMD, rapid eye movement sleep density REML, rapid eye movement sleep latency; SE, sleep efficiency; SL, sleep latency; SWS, slow wave sleep; TST, total sleep time; WASO, wake time after sleep onset.
Main analyses
For the main analyses of sleep macrostructural data, one hundred and forty-seven (45.8%) of 321 pooled analyses were statistically significant with p < 0.05, 73 (22.7%) with p < 0.001, and 30 (9.3%) with p < 0.000001. 21 (14.3%) of 147 statistically significant pooled analyses included more than 1000 neuropsychiatric cases per disease. 146 (45.5%) of 321 comparisons showed large heterogeneity (I 2 > 50%). In 95 of the 321 pooled analyses (29.6%), the effect sizes of the largest study were nominally statistically significant at p < 0.05. The 95% prediction interval excluded the null in only 22 (6.9%) of 321 pooled analyses. Small-study effects were found for 37 pooled analyses (11.5%), and excess significance bias was identified for 62 pooled analyses (19.3%) (Table 1). For the main analyses of sleep microstructural data (CAP parameters), please see online Supplementary Table S7.
Credibility of evidence
Of the 321 pooled analyses none had convincing strength of PSG differences according to quantitative umbrella review criteria (see Fig. 3). Only seven (2.2%) were supported by highly suggestive evidence; increased SL and decreased SE in MDD, increased N1 percentage, and decreased N2 percentage, SL and REML in narcolepsy, and decreased REM sleep percentage in PD. Five (1.6%) were supported by suggestive evidence; decreased REML in MDD, decreased SE in PTSD and in narcolepsy, and decreased TST and SE in PD. There were 136 (42.4%) pooled analyses supported by weak evidence and 174 (54.2%) showing no significant changes in sleep parameters in neuropsychiatric diseases compared with HCs. The findings of subset analyses are listed in online Supplementary Table S9 and Appendix pp16.
Credibility of polysomnographic alterations in 27 neuropsychiatric diseases. Abbreviations of disease names: AD, Alzheimer's disease; ADHD, attention deficit hyperactivity disorder; ASD, autism spectrum disorder; BPD, borderline personality disorder; GAD, generalized anxiety disorder; HD, Huntington's disease; IH, idiopathic hypersomnia; iRBD, idiopathic rapid eye movement sleep behavior disorder; MCI, mild cognitive impairment; MDD, major depressive disorder; OCD, obsessive compulsive disorder; PD, Parkinson's disease; PTD, persistent tic disorder; PTSD, posttraumatic stress disorder, SAD, social anxiety disorder; TBI, traumatic brain injury, WD, Wilson's disease. Abbreviations for sleep parameters: AHI, apnea hypopnea index; AI, arousal index; PLMI, Periodic limb movement index; REM, rapid eye movement sleep; REMD, rapid eye movement sleep density REML, rapid eye movement sleep latency; SE, sleep efficiency; SL, sleep latency; SWS, slow wave sleep; TST, total sleep time; WASO, wake time after sleep onset.
Discussion
To our knowledge, this is the first umbrella review of alterations in PSG parameters in neuropsychiatric diseases. Our umbrella review has the particular strength of including a robust hierarchical classification of the published evidence. We reviewed 27 systematic reviews of 321 pooled analyses of studies of PSG alterations in neuropsychiatric diseases compared with HCs. Overall, available experimental evidence shows that patients with neuropsychiatric diseases show altered PSG characteristics compared with HCs, but strength of these findings varied considerably. Seven of the 147 statistically significant pooled analyses were supported by highly suggestive evidence: increased SL and decreased SE in MDD, increased N1 percentage, and decreased N2 percentage, SL and REML in narcolepsy, and decreased REM sleep percentage in PD, while five pooled analyses were supported by suggestive evidence: decreased REML in MDD, decreased SE in PTSD and in narcolepsy, and decreased TST and SE in PD.
Overall, our umbrella review shows that, although alterations in multiple PSG characteristics in various neuropsychiatric diseases have been evaluated in multiple studies, reviews and meta-analyses, the number of changes of PSG characteristics that have suggestive or stronger support is limited. In addition, no significant pooled analyses concerning PSG changes are supported by convincing evidence. Consistent with umbrella review criteria, high between-study heterogeneity, random effects p value > 10−6, sample size of cases < 1000, prediction intervals including the null value, and small-study effects bias are common contributors that downgrade the overall confidence of published meta-analyses. Our umbrella review finds that small sample sizes in the individual studies and meta-analyses are the main factor downgrading PSG findings in neuropsychiatric diseases. This may be attributable to the relatively low incidence of some diseases (i.e. HD, WD, and SAD) in the general population, and the methodological challenges of putting patients who exhibit complex combinations of neurological symptoms (i.e. motor and cognitive impairments) and psychiatric features through the relatively intense protocols required for PSG research. Furthermore, sleep problems in some neuropsychiatric diseases tend to go undiagnosed by physicians and underreported by patients, possibly due to a lack of insight or perceived relative unimportance of sleep disturbances compared with the motor, cognitive and psychiatric features that are recognized as key features of the diseases (Videnovic, Lazar, Barker, & Overeem, Reference Videnovic, Lazar, Barker and Overeem2014). This may result in PSG examinations not being prescribed for many patients with neuropsychiatric diseases.
Nevertheless, from a clinical perspective, exploring PSG characteristics in neuropsychiatric diseases can provide valuable information and insight. Sleep comprises approximately one third of human life and is a critical state for basic brain function and neuropsychiatric health (Baglioni et al., Reference Baglioni, Nanovska, Regen, Spiegelhalder, Feige, Nissen and Riemann2016a; Harvey, Murray, Chandler, & Soehner, Reference Harvey, Murray, Chandler and Soehner2011; Regier, Kuhl, Narrow, & Kupfer, Reference Regier, Kuhl, Narrow and Kupfer2012). Our umbrella review revealed that increased SL and decreased SE in MDD, increased N1 percentage in narcolepsy, and decreased REM sleep percentage in PD ranked as highly suggestive evidence. These findings could be seen in other neuropsychiatric diseases (i.e. PTSD, schizophrenia, and HD), although the level of credibility of evidence varied; suggesting that single PSG parameter changes should be considered as transdiagnostic sleep characteristics across various neuropsychiatric diseases rather than disease-specific sleep features. Still lacking is robust evidence supporting that any single sleep variable alteration is specific for a single disease, as suggested by Benca, Obermeyer, Thisted, & Gillin (Reference Benca, Obermeyer, Thisted and Gillin1992) (Benca et al., Reference Benca, Obermeyer, Thisted and Gillin1992) and Baglioni et al. (Baglioni et al., Reference Baglioni, Nanovska, Regen, Spiegelhalder, Feige, Nissen and Riemann2016a). By comparison, looking across PSG variables reveals that no two diseases have the same sleep profile (Fig. 2). This suggests that specific profiles of sleep alterations may best define distinct disorders rather than alterations in a single sleep variable.
A great amount of research has been conducted on genes, proteins, and neural circuits to try to find biomarkers which could identify or predict neuropsychiatric diseases; however, to date, no specific marker has been found which confidently identifies or distinguishes different neuropsychiatric diseases. In addition, psychomotor activity, mood, cognition, suicidal ideation, psychotic symptoms, and neurological symptoms, have been traditionally considered basic dimensions in neuropsychiatric diseases (Cuthbert & Kozak, Reference Cuthbert and Kozak2013; Morris, Rumsey, & Cuthbert, Reference Morris, Rumsey and Cuthbert2014; Sanislow et al., Reference Sanislow, Pine, Quinn, Kozak, Garvey, Heinssen and Cuthbert2010). Our results suggest that the overall change patterns of PSG parameters should be comprehensively evaluated as an important basic dimension and potential disease-specific biomarker for neuropsychiatric diseases (Lim et al., Reference Lim, Mazzotti, Sutherland, Mindel, Kim, Cistulli and Penzel2020). However, the umbrella review method we employed did not allow a statistical analysis that would test the ability of specific sleep profiles to identify or distinguish different neuropsychiatric conditions. This hypothesis could be potentially tested using machine learning methodology in a large sample study consisting of various neuropsychiatric diseases.
Existing evidence shows that successful treatment of sleep disturbances has a positive impact on the course of neuropsychiatric diseases (Gee et al., Reference Gee, Orchard, Clarke, Joy, Clarke and Reynolds2018; Krystal, Reference Krystal2020). Traditionally, pharmacologic (i.e. hypnotics) and psychosocial interventions (i.e. cognitive behavioral therapy for insomnia) are the main options for treating sleep disturbances in various neuropsychiatric diseases (Qaseem, Kansagara, Forciea, Cooke, & Denberg, Reference Qaseem, Kansagara, Forciea, Cooke and Denberg2016; van der Zweerde, Bisdounis, Kyle, Lancee, & van Straten, Reference van der Zweerde, Bisdounis, Kyle, Lancee and van Straten2019). It has been suggested that these therapies may improve some altered PSG determined variables, such as TST and SE, which are seen in different neuropsychiatric diseases (Monti, Torterolo, & Pandi Perumal, Reference Monti, Torterolo and Pandi Perumal2017; Talbot et al., Reference Talbot, Maguen, Metzler, Schmitz, McCaslin, Richards and Neylan2014). Thus, from the prospective of improvements of PSG variables, traditional pharmacologic and psychosocial interventions have the properties of transdiagnostic treatments (one treatment that could improve sleep disturbance across patients with different diagnosis). Harvey and colleagues have proposed that the use of transdiagnostic treatment protocols could decrease the burden on clinicians, who currently must learn multiple specific treatment protocols that often share many common theoretical underpinnings and components (Harvey et al., Reference Harvey, Murray, Chandler and Soehner2011). On the other hand, given the different PSG patterns across different neuropsychiatric diseases seen in our umbrella review, it would appear that ‘one size fits all approach’ treatment protocols may be insufficient to improve sleep in all neuropsychiatric diseases. Rather more targeted treatment approaches should emphasize disease-specific altered sleep patterns in developing new sleep intervention protocols across different neuropsychiatric diseases. This idea was also proposed by Harvey and colleges (Harvey, Reference Harvey2009; Harvey et al., Reference Harvey, Murray, Chandler and Soehner2011) who suggested that new sleep intervention protocols should include core treatment modules that would be delivered regardless of diagnosis, in addition to optional modules to cover treatment of disorder-specific symptoms.
Clinically, serious psychiatric and neurological symptoms and sleep disturbances in some neuropsychiatric patients do not allow the withdrawal of treatment. When performing PSG examinations, some medications (i.e. anti-depressants and hypnotics) may affect sleep measures. Additionally, one neuropsychiatric disease may co-occur with other neuropsychiatric diseases (i.e. depression in PTSD, schizophrenia, and PD) which may interact and produce either over- or under-estimations of PSG changes in patients with neuropsychiatric diseases. Thus, we limited analyses to studies excluding patients with comorbidities, and studies excluding patients taking antidepressant and hypnotics, which revealed that only decreased SE and increased SL in MDD remained as highly suggestive evidence. Majority of other comparisons in the subset analyses were downgraded to weak evidence or changed to no significant PSG differences between cases and HCs. In fact, stratifying the analysis by the aforementioned factors inevitably decreased the power of the analysis. As shown in Table 1 and online Supplementary Table S9 except for MDD, the sample size in other subset analyses were largely decreased compared with the whole sample analysis, which may be the main factor that decreased the power of the analysis. Nevertheless, it should be noted that majority of the patients in the whole sample analysis were drug-naïve or had a washout period before PSG examination and that most of the component studies had excluded patients with other comorbid neuropsychiatric diseases, minimizing or alleviating these potential confounds to accurate PSG measurement.
This umbrella review has some limitations. First, some meta-analyses were excluded from predictive intervals and excess significance tests because they did not provide adequate data necessary to conduct the respective analyses. Second, we did not assess the quality of component studies of each of the meta-analyses as it was beyond the scope of our umbrella review. Third, biases that might have been caused by the respective method characteristics of individual component studies, such as sex, age, race/ethnicity, socioeconomic status effects, and genetic causes of diseases, were not fully assessed in our umbrella review, due to insufficient information (i.e. not performing analyses stratified by sex or other factors) reported in the majority of the component studies. Fourth, our umbrella review did not include all neuropsychiatric diseases. For instance, PSG changes in multiple sclerosis and multiple system atrophy are lacking because meta-analyses for these topics were not found in our literature search. Thus, the evidence map of PSG characteristics is still incomplete. Fifth, in our study selection process, we encountered more than one meta-analysis on the same topic that included some, but not all, of the same studies. In these instances, we included the most up-to-date meta-analysis that contained the most studies. It also should be noted that, in addition to newly identified original case–control studies, different meta-analyses on the same topic may use different eligibility criteria and different search terms that results in differences in included studies. This means that not all relevant data across meta-analytic studies were considered. However, we cannot offer a way to address this concern, though it has been previously noted (Correll et al., Reference Correll, Cortese, Croatto, Monaco, Krinitski, Arrondo and Solmi2021; Dragioti et al., Reference Dragioti, Solmi, Favaro, Fusar-Poli, Dazzan, Thompson and Evangelou2019; Kim et al., Reference Kim, Kim, Lee, Jeong, Lee, Lee and Fusar-Poli2020) and may be resolved in future as the methodology for umbrella reviews continues to evolve.
Despite these limitations, this umbrella review mapped PSG characteristics across 27 neuropsychiatric diseases. Out of 321 identified PSG comparisons, evidence from the pooled analyses was highly suggestive for increased SL and decreased SE in MDD, increased N1 percentage, and decreased N2 percentage, SL and REML in narcolepsy, and decreased REM sleep percentage in PD. Evidence from the pooled analyses was suggestive for decreased REML in MDD, decreased SE in PTSD and in narcolepsy, and decreased TST and SE in PD. We cannot state that other PSG comparisons supported by weak evidence are not meaningful, but they have uncertainties that need to be resolved. Although the credibility of evidence of PSG characteristics in the 27 neuropsychiatric diseases varied across different PSG variables and different diseases, the current findings provide a starting point that may guide advances in sleep research and improve the understanding of sleep features in neuropsychiatric diseases. Critically, no two diseases had the same altered sleep patterns, suggesting that specific sleep profiles may be an important dimension for marking distinct disorders. Further well-designed studies with large sample sizes and accurate assessment of potential biases are needed to confirm and expand these findings.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291722001581
Acknowledgements
This work was supported by the Ministry of Science and Technology of the People's Republic of China (2021ZD0201900) and the National Natural Science Foundation of China (82120108002, 82170099, 82170100).
Author's contributions
XDT designed the study. YZ drafted the manuscript. YZ, RR, YS and HPZ contributed to database preparation. YZ and RR contributed to accuracy checks, did data analyses and assessed the quality of meta-analyses using AMSTAR-2. LDS, MVV and LHY provided important suggestions for improving the manuscript and critically revised the manuscript. All authors commented on and approved drafts and the final manuscript.
Conflict of interest
We declare no competing interests.
