Behavioural, medical & environmental interventions to improve sleep quality for mental health inpatients in secure settings: a systematic review & meta-analysis

ABSTRACT

Despite the bidirectional relationship between sleep and mental health, the effect of interventions to improve sleep quality for inpatients from secure psychiatric settings has not been examined. The current work aimed to identify the effect of interventions for improving sleep quality for these inpatients and identify moderating factors. Eligible studies involved secure psychiatric inpatients (adult) and included a quantitative measure of sleep as primary outcome. The Cochrane Library, Scopus, PubMed and ProQuest were searched; last searched October 2021. 12,409 abstracts were screened for inclusion. The systematic review included 38 studies. The meta-analysis included 22 studies (36 trials).Sleep Quality, Insomnia Severity, Total Sleep Time and Sleep Efficiency were outcomes. The total pooled effect size for all interventions (random-effects model) was d=0.54, (p<0.0001,95% CI:0.30,0.77). Subset-analyses indicated the pooled effect sizes for behavioural interventions as d=0.65, (p<0.01,95% CI:0.16,1.14), medical interventions as d=0.58, (p<0.001,95% CI:0.25,0.91),environmental interventions as d=0.19, (p=0.38,95% CI:-0.24,0.62). Based upon duration, interventions of a five-to-ten-week period were most effective, d=0.92, (p<0.005,95% CI:0.29,1.55). Males reported a greater improvement in sleep than females. Sleep quality interventions in this population are effective. Behavioural interventions: Cognitive Behavioural Therapy for insomnia and physical activity, are most impactful. Clinical practitioners should consider implementing behavioural interventions, of a five-to-ten-week period.

KEYWORDS:

• Systematic review
• meta-analysis
• sleep quality
• insomnia
• mental health
• inpatients

Introduction

There is a well-documented bidirectional relationship between sleep and mental health (Lallukka & Sivertsen, 2017), termed the sleep-mood cycle (Talbot et al., 2012), in which poor night-time sleep and daytime mood can be mutually maintaining (Harvey, 2008). This has been reported in bipolar disorder (Talbot et al., 2012), depressive and anxiety disorders (Alvaro et al., 2013; Jansson-Fröjmark & Lindblom, 2008), and psychosis (Reeve et al., 2018). For patients diagnosed with a mental health disorder, sleep problems are very common (Freeman et al., 2017), with psychiatric disorders and sleep disorders often overlapping (Talih et al., 2018). One of the most common sleep disorders in this population is insomnia; a disorder involving prolonged difficulties in initiating or maintaining sleep, with detrimental effects on daytime functioning and wellbeing (Freeman et al., 2017). In a recent international study, the prevalence rate of acute insomnia diagnosis amongst otherwise healthy adults in the general population was 11.2% (Aernout et al., 2021). In comparison, 23% of patients with schizophrenia and 22.6% of patients with first episode psychosis were reported to meet the diagnostic criteria for insomnia (Batalla‐martín et al., 2020; Subramaniam et al., 2018). Until recently, the accepted conceptualisation was that insomnia symptoms were just symptoms associated with mental health disorders (Matteson-Rusby et al., 2010). However, the growing body of research into the predictive relationships between mental health and insomnia (Hertenstein et al., 2019) has led to insomnia itself being considered a primary disorder, not just a secondary symptom (Pigeon et al., 2017). The treatment of sleep disorders is now considered to be a preventive strategy within mental health (Pigeon et al., 2017) and recommendations now suggest insomnia should be assessed routinely in the occurrence of a mental health disorder, as well as being treated in its own right (Freeman et al., 2020).

The literature suggests, broadly, three types of intervention to improve sleep for psychiatric patients; medical, behavioural and environmental. In terms of the medical interventions, one common treatment option is the prescription of antidepressant or antipsychotic medication with sedative effects (Becker & Sattar, 2009; Khurshid, 2018). These can increase total sleep time and decrease awakenings (Khurshid, 2018), but side effects (e.g. weight gain, fatigue) can limit usefulness (Becker & Sattar, 2009) and consequences, such as cognitive decline, need consideration (Khurshid, 2018). Similarly, daytime sleepiness in schizophrenia can be improved by one antipsychotic, lurasidone, but worsened by another, quetiapine XR (Loebel et al., 2014).

A non-pharmacological treatment for psychiatric inpatients with self-reported insomnia symptoms is Cognitive Behavioural Therapy for Insomnia (CBTI) which involves psychoeducation, stimulus control and help with coping strategies for night time worries (Sheaves et al., 2018). CBTI is feasible and effective as either a two-week programme for psychiatric inpatients in acute crisis (Sheaves et al., 2018) or as a single session for inmates with acute insomnia, referred to mental health teams in a correctional facility (Randall et al., 2019). Though both of these studies were effective, they only incorporated male participants, reducing the generalisability of results (Randall et al., 2019; Sheaves et al., 2018). Other behavioural interventions, including physical activity such as Tai Chi (Zhu et al., 2018) and yoga (Hegde et al., 2020), can be beneficial for sleep quality in inpatient psychiatric populations. Whilst these are feasible and may act as simple supplementary therapies (Hegde et al., 2020), attention is needed to ensure that intensity level is appropriate and achievable and intervention duration is suitable (Zhu et al., 2018).

Sleeping in a hospital can lead to a reduction in sleep duration and sleep quality, compared to being within domestic surroundings (Delaney et al., 2018). In institutionalised psychiatric health care, patients report that environmental disturbances such as regular safety checking, disruptive light and noise can all negatively impact night-time sleep (Veale et al., 2020). A lack of stimulating and engaging activity (O’Connell et al., 2010), feelings of being imprisoned and mandatory care such as medication (Kamphuis et al., 2013), can have the same impact. Interventions to alleviate such problems include environmental lighting change (Sloane et al., 2007), night-time confinement (Chu et al., 2015) and re-design of patient space (Pyrke et al., 2017). Whilst increased exposure to morning and all-day light can significantly increase night-time sleep statistically, the clinical impact of increasing sleep duration by 11 minutes per night is minimal (Sloane et al., 2007). Similar results are seen in implementing night-time confinement (Chu et al., 2015). A move from dorm-style accommodation to private sleeping quarters can objectively improve sleep efficiency, however self-reported data showed no pre-post difference in sleep quality, suggesting that type of measure used can influence results (Pyrke et al., 2017). Consideration is needed in weighing up costs, benefits and feasibility of environmental interventions (Sloane et al., 2007), given the potential high cost and uncertainty of impact (Chu et al., 2015; Pyrke et al., 2017; Sloane et al., 2007).

It is important to identify which interventions may be most efficacious to improve inpatient sleep and to consider which treatments may be more effective for individual patient groups. Reports consistently indicate a higher incidence of problems with sleep quality and duration for women, compared to men, however, the progressive impact of aging on sleep quality applies to both sexes (Madrid-Valero et al., 2017). Thus, both sex and age need be considered when examining the effectiveness of sleep improvement interventions. In addition, both pharmacological and non-pharmacological treatment approaches for insomnia can be effective but are recommended for different stages of insomnia (Rios et al., 2019). Similarly, there are varying reports on the optimal intervention duration (Rios et al., 2019). Whilst global guidelines encompass similar recommendations for insomnia treatment (Rios et al., 2019), including CBTI (Kyle et al., 2020), access can be limited across healthcare providers internationally (Kyle et al., 2020). It is also important to consider differences between institutional environments as, for example, prisoners requiring mental healthcare can receive a very different standard of care when transferred to a secure hospital (Dyer et al., 2021).

For psychiatric inpatients residing in secure facilities such as forensic psychiatric hospitals, worse sleep quality and higher insomnia scores significantly relate to aggression, hostility and violent incidents (Kamphuis et al., 2014). Thus, targeting inpatient sleep quality can help to reduce aggressive incidents in these populations, which is paramount to continued successful psychiatric treatment (Van Veen et al., 2020). Therefore, the aim of this paper is to examine the effectiveness of behavioural, medical and environmental interventions for improving sleep quality in psychiatric inpatients. Subgroup analysis will be performed to examine differences between set groups based on research studies’ quality score, age, gender, institutional location and type, intervention type, sleep quality measure used and intervention duration.

Method

The review and analysis were reported using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA (Page et al., 2021), Supplementary Material, Appendix 1); and registered in the International Prospective Register of Systematic Reviews (PROSPERO; Reference: CRD42020177944).

Systematic review eligibility criteria & search strategy

Table 1 illustrates the eligibility criteria. A literature search was conducted across four electronic databases: Cochrane Library, Scopus, PubMed and ProQuest, using title, keywords and abstracts, with search terms including truncation (*) where relevant, to accommodate word variants or plurals. Google Scholar was searched following database searching to ensure all relevant articles had been detected. Reference lists of all included articles at the full text level were also scanned. Contact with authors was made where necessary for further data or clarification. The search was conducted in March 2020, updated on October 6^th, 2021. Details of the search strategy are available in Supplementary Material, Appendix 2.

Table 1. Systematic Review Inclusion and Exclusion Criteria.

	Inclusion Criteria	Exclusion Criteria
Date	01.01.2007–06.10.2021.
Language	English language only.
Population	Human participants, adults (aged 18+) residing in a secure, inpatient setting at the time of study. All countries included.	From community and outpatient settings.
Outcome	Primary outcome of the intervention must be a sleep-related outcome.
Design	Quantitative measures of sleep outcomes only. Primary research; randomised control trials, clinical trials, case studies, reports and interventions.	Meta analyses, reviews; systematic, scoping, literature etc.

Identification of relevant studies

All references identified were imported to the data management software Rayyan QCRI (Ouzzani et al., 2016) to identify duplicates, screen articles, record exclusion reasons, link to full texts and store final articles. Software identified duplicates were screened by one reviewer (PG). Study titles and keywords were screened simultaneously for eligibility (Table 1), then screened via abstract, by one reviewer (PG). Two reviewers (PG&IH) then confirmed the studies to be screened via full text, with a third reviewer (FK) available if a mutual decision could not be reached. Figure 1 illustrates the identification process.

Figure 1. PRISMA flow diagram of systematic literature search & study selection (Page et al., 2021)

Data extraction

Data extraction tables were piloted on several studies by one reviewer (PG) and confirmed by a second reviewer (IH). All data were extracted from the included studies into an Excel spreadsheet by one reviewer (PG) with a second reviewer (IH) extracting data in the same manner from a percentage of studies in tandem. Extracted data included: (1) citation, (2) aims, hypotheses, design, (3) participant demographics, diagnoses, admission detail, (4) setting, security level, (5) group method, procedure, (6) outcome measures, scales, (7) summary data for sleep, other outcomes, (8) conclusions, reference to relevant studies.

Risk of bias assessment

The quality of individual included studies was assessed using the Mixed Methods Appraisal Tool (Hong et al., 2018). The tool is designed to ensure that each included study is subject to two screening questions and five questions unique to each study type (qualitative, quantitative, mixed methods, quantitative randomised controlled trials or crossover trials, quantitative non-randomised trials). Based on an earlier MMAT (Pluye et al., 2011), each paper was scored against items with answers, ‘yes’, ‘no’ or ‘can’t tell’. With each of the five criteria items representing 20%, each paper had capacity to score 100%; 0–40% = poor quality, 60% = acceptable, 80–100% = high quality (Pluye et al., 2011). It is not advised that studies are excluded purely due to a low-quality rating, but instead all outcomes are presented and discussed (Hong et al., 2018). This approach was adopted for the present review. The primary author (PG) scored the quality of each article and a co-author (IH) independently scored 20%. A third reviewer (FK) was available for consultation. Throughout conducting the review, risk of bias was minimised by adhering to PRISMA (Page et al., 2021) statements and hand searching Google Scholar and reference lists. Quality assessment of the present review was ensured through following the AMSTAR 2 quality of methodology tool (Shea et al., 2017). The completed assessment is available in Supplementary Material, Appendix 7.

Statistical analysis

Data were prepared in Microsoft Excel 2016, then exported for statistical analyses performed in R, version 4.0.5 (31 March 2021) with RStudio, using the meta and metafor packages. Microsoft Excel 2016 was also used to create Figures 3 and 4 and to conduct the Grubb’s test. Included studies comprised both within-subjects and between-subjects designs, as to best capture all available research and allow for a larger, more detailed overview of existing data. Therefore, the effectiveness of the interventions was based on either the difference between baseline result and follow up result (within) or control group post-result to experimental group post-result (between). Where studies had divided their experimental group into separate groups, included multiple separate experimental populations or conditions, these were considered as individual trials. Therefore, some studies are cited with additional information, for example ‘Moon, 2016: Depressive’ to indicate the individual trial. Only studies reporting the following key sleep outcomes were included in the meta-analysis: sleep quality measured by the Pittsburgh Sleep Quality Index (Buysse et al., 1989), insomnia severity measured by the Insomnia Severity Index (Morin, 1993) or sleep efficiency or total sleep time, measured with actigraphy or polysomnography. These key outcomes were decided upon by two reviewers (PG&IH), as most studies recorded at-least one of these outcomes and they represent both self-reported and instrumental measures of sleep. As a result, 22 studies from the systematic review were included in the meta-analysis.

As we anticipated considerable intra-study heterogeneity, a random-effects model meta-analysis was conducted. This model assumes the effect of interest in all studies is not the same and that observed differences between studies are caused by variability within the studies. The heterogeneity of the effect size (Cohen’s d) between each of the studies was quantified using τ2 and I² statistics, with limits which suggest I² values under 31% are of little concern, whilst those over 56% are likely to have substantial heterogeneity (Higgins & Thompson, 2002). A sensitivity analysis was conducted by way of a diagnostic Baujat plot (Supplementary Material, Appendix 5) and use of Grubb’s statistic, which illustrated the contribution of each study and was used to detect outliers. Publication bias was represented visually using funnel plots (Supplementary Material, Appendix 6) alongside the statistical Egger’s test for asymmetry.

A total pooled effect size using random-effects modelling of all sleep quality improvement interventions was calculated, in order to illustrate the effect of interventions on key sleep outcomes; this was illustrated in a forest plot as a point of Standardised Mean Differences (SMDs; representing Cohen’s d) with 95% confidence intervals (95%CIs). Effect sizes are considered small (d = 0.2), medium (d = 0.5) and large (d = 0.8) (Cohen, 1988). Subgroup analysis was performed to examine the differences between set groups based on categorical variables; studies’ quality score, age, gender, institutional location and type, intervention type, sleep quality measure used and intervention duration.

Results

Systematic review

The initial search was conducted in March 2020 and updated in October 2021 with a total of 72,011 records identified through database searching. Thirty-eight studies were identified at the full text level for inclusion in the systematic review. An overview of included studies, including further detail regarding subgroups, is available in Table 2. Study design detail is also illustrated in Supplementary Material, Appendix 8.

Table 2. Systematic literature review.

Primary Author(s) & Year	Country	Design	Intervention Detail	Intervention Duration	Sleep Outcome(s) of Interest	Direction of Change of Sleep Outcome(s) (Author Reported)	Population	(RoB) Quality Score
Baune et al. (2007)	Germany	Pre-Post	Pharmacological - (venlafaxine, escitalopram, quetiapine.)	≤4 weeks	Sleep Quality (PSQI)	↑	Psychiatric ward, predominantly male, no sleep criteria.	80%
Benedetti et al. (2007)	Italy	Pre-Post	Total Sleep Deprivation (TSD) & Light Therapy (LT)	≤1 week	Actigraphy (Sleep Efficiency & Total Sleep Time]	─	Inpatients with bipolar disorder in depressive episode, predominantly female, no sleep criteria.	80%
Brupbacher et al. (2021)	Switzerland	RCT	One shot aerobic exercise programme (cycling).	≤1 week	Polysomnography (PSG)	→	Rehabilitation inpatients with unipolar depression, predominantly female, those with RLS, mod-severe sleep apnoea excluded.	80%
Cameron et al. (2014)	Canada	Pre-Post Retrospective Evaluation	Pharmacological: Synthetic Cannabinoid	N/R	Chart Review: Sleep Hours & Nightmares (n).	↑	Male inmates with serious mental illness- secure treatment unit (STU), no sleep criteria.	80%
Canazei et al. (2019)	Austria	RCT	Light box therapy.	≤4 weeks	Sleep Quality (PSQI)	↑	Inpatients with burnout in a psychiatric clinic, predominantly female, with evening types excluded.	60%
Chien et al. (2015)	Taiwan	RCT	Breathing relaxation exercise & CBT in group therapy.	≤4 weeks	Sleep Quality (PSQI)	↑	Psychiatric ward, predominantly male, sleep criteria inc. PSQI scores ≥ 5.	60%
Chu et al. (2015)	England	Pre-Post	Night Time Confinement Policy	10 weeks to 3 months	Sleep Quality (PSQI)	→	High secure forensic psychiatric hospital, gender N/R, no sleep criteria.	60%
de Niet et al. (2010)	The Netherlands	Pre-Post Pilot (Quasi-Experimental)	Stimulus Control (CBT) OR Music-Assisted Relaxation (MAR) (MP3) Plus Sleep Education	≤4 weeks	Brief Assessment of Insomnia), Sleep Questionnaire (RCSQ) (Richards et al., 2000)	Insomnia Assessment → RSCQ Total → RSCQ Sleep Quality ↑	Psychiatric hospital inpatients, predominantly female, sleep criteria not sufficiently recorded.	60%
Ellis et al. (2019)	USA	Pre-Post: Open Trial & Case Series	Modified Imagery Rehearsal Therapy (IRT) in small group sessions.	≤4 weeks	Insomnia (ISI) Disturbing Dreams & Nightmare Severity Index (Krakow et al., 2002)	ISI ↑ Disturbing Dreams & Nightmare Severity Index ↑	Psychiatric hospital inpatients, predominantly male, sleep criteria inc. disturbed sleep or self-reported nightmares.	80%
Ferszt et al. (2015)	USA	Pre-Post: Non-Experimental	Mindfulness based programme and meditation homework.	10 weeks to 3 months	Sleep Quality (PSQI)	↑	Female prison inmates, no sleep criteria reported.	60%
Göder et al. (2016)	Germany	Pre-Post	Electroconvulsive Therapy (ECT)	5–10 weeks	Polysomnography (PSG) (Sleep Efficiency & Total Sleep Time)	→	Psychiatric ward inpatients, predominantly female, no sleep criteria reported.	60%
Haynes et al. (2011)	USA	Chart Review (Retrospective Pre-Post)	Behavioural Group Therapy for Insomnia Symptoms	≤1 week	Insomnia (ISI) ‘	↑	Psychiatric ward inpatients, predominantly male, no sleep criteria reported.	60%
Hegde et al. (2020)	India	Two Group Experiment (Randomised Crossover)	“Shirodhara” –yoga with additional ‘oil dropping’ on the head.	≤1 week	Sleep Quality Rating Questionnaire	↑	Inpatient psychiatric, predominantly female, no sleep criteria reported.	100%
Hemmeter et al. (2007)	Switzerland	RCT – Double blind placebo controlled	Pharmacological with flumazenil & Partial Seep Deprivation (PSD)	≤1 week	Microsleep, Mood & Tiredness (Visual Analogue Scale).	→	Inpatients with diagnosis of major depression, predominantly male, no sleep criteria reported.	60%
Henriksen et al. (2020)	Norway	Part of multicentre RCT	Blue blocking glasses.	≤1 week	Actigraphy (Sleep Efficiency & Total Sleep Time).	↑	Hospital inpatients, manic phase BD, predominantly male, no sleep criteria reported.	40%
Holzinger et al. (2020)	Austria	RCT	Lucid dream therapy – inc. self-hypnosis.	5–10 weeks	Sleep Quality (PSQI), ESS & Nightmare Frequency	→	Psychiatric hospital inpatients, predominantly female, with nightmares.	60%
Hsu et al. (2015)	Taiwan	Non-Randomised Trial: Two Groups, Experimental & Control	CBTI - focussed on Spielman’s 3Ps and managing stress that interferes with sleep, plus homework.	5–10 weeks	Sleep Quality (PSQI), DBAS (Morin et al., 2007), PSAS (Nicassio et al., 1985), SHPS (Yang et al., 2010)	─	Psychiatric hospital inpatients, predominantly female with numerous insomnia complaints.	60%
Huang et al. (2019)	China	RCT	“Systematic Psychological Nursing” – emotional, psychologically informed nursing	8–12 weeks	Sleep Quality (PSQI) ‘Sleep Quality’ only, not total score.	↑	Hospitalised inpatients with schizophrenia, predominantly female, with a sleep disorder.	60%
Kluge et al. (2012)	Germany	Randomised Trial: Two Experimental Groups, Double Blind	Pharmacological - Clozapine and olanzapine first at lower doses and then higher doses.	5–10 weeks	Polysomnography (PSG) Daytime sleepiness. Sleep propensity, latency, onset.	─	Inpatients with schizophrenia, schizoaffective or schizophreniform disorder, predominantly female, no sleep criteria reported.	60%
Laguna-Parras et al. (2013)	Spain	A quasi-experimental pre-test-post-test.	“Sleep enhancement” nursing intervention.	3 months+	Oviedo Sleep Questionnaire(Bobes Garcı´a et al., 2000)	↑	Inpatients with mental health disorder(s), predominantly female, with disturbed sleep.	100%
Loebel et al. (2014)	USA, Russia, India, Ukraine, Romania, Columbia.	RCT (Double Blind, Placebo & Active Controlled Study)	Pharmacological: Lurasidone 160mg/d, Lurasidone 80mg/d or Quetiapine XR 600mg/d.	≤4 weeks	Daytime Sleepiness (ESS)	Lurasidone → Quetiapine ↓	Inpatients (DSM-IV-TR schizophrenia), predominantly male, no sleep criteria reported.	100%
Li and Lang (2021)	China	RCT	Psychological nursing.	≤4 weeks	Sleep Quality (PSQI) Sleep Quality’ only, not total score.	↑	Inpatients with schizophrenia, predominantly female, inc: sleep disorder and +12 PSQI score.	40%
Lu et al. (2013)	South Taiwan	RCT	Acupressure applied to the hands and feet.	≤4 weeks	Sleep Quality (PSQI), Actigraphy (Sleep Efficiency & Total Sleep Time)	↑	Psychogeriatric inpatients, predominantly male, inc: +5 PSQI score.	60%
Martin et al. (2018)	UK	Pre-Post: Trial - Quality Improvement Project	Ward Lighting - Red rubber filters on bulbs in bedroom lighting.	10 weeks to 3 months	Chart Review & Staff Questionnaires (Mixed Method)	→	Old age psychiatry inpatient with dementia or cognitive impairment, gender N/R, no sleep criteria recorded.	60%
Moon et al. (2016)	South Korea	Non-Randomised Trial: with Control Group	Treatment for bipolar disorder - psychiatric hospital inpatient.	Continual: 3 months+	Actigraphy (Sleep Efficiency & Total Sleep Time)	─	Psychiatric ward inpatients, mostly male, sleep problem not necessary for inclusion.	80%
Nakamura et al. (2017)	Japan	Pre-Post: Open Label -Prospective	Pharmacological – suvorexant.	≤4 weeks	Sleep Quality (PSQI)	↑	Psychiatric ward inpatients, predominantly male, any sleep disorder included.	80%
Nishida et al. (2017)	Japan	Pre-Post: Open Label Pilot Study	rTMS (Repetitive Transcranial Magnetic Stimulation)	≤4 weeks	Actigraphy (Sleep Efficiency & Total Sleep Time), Sleep Quality (PSQI)	Sleep Efficiency & Total Sleep Time → PSQI ↑	Psychiatric ward inpatients, predominantly female, with sleep disorders excluded.	80%
Okkels et al. (2020)	Denmark	RCT (Effectiveness)	Bedroom lighting (blue lighting, bright light dawn stimulation, white light).	N/R	Sleep Quality (PSQI)	→	Hospitalised psychiatric patients, predominantly female, all 5+ on the PSQI.	100%
Pyrke et al. (2017)	Canada	Pre-Post of an External Intervention.	Room & Hospital Design Change – Reconstruction and redesign of facility.	≤1 week	Sleep Quality (PSQI), Actigraphy (Sleep Efficiency & Total Sleep Time), Sleep Diary	PSQI → Total Sleep Time → Sleep Efficiency ↑	Psychiatric hospital inpatients, 50% female, those with sleep disorders excluded.	80%
Randall et al. (2019)	England	Pre-Post	CBTI on a one-to-one basis, with self help pamphlet, sleep diaries.	≤4 weeks	Insomnia (ISI)	↑	Male prison inmates (Category C), inc: DSM criteria for insomnia.	60%
Reiter and Humphreys (2021)	Australia	Pre-Post Case Series	Exposure, relaxing and rescripting of nightmares.	≤4 weeks	Sleep Quality (PSQI), Insomnia Severity Index (ISI)	─	Hospital inpatients with PTSD, predominantly female.	60%
Schäfer et al. (2019)	Germany	RCT	Olfactory Stimulation (Pleasant odors via tube or nasal clip whilst sleeping).	≤1 week	Actigraphy (Sleep Efficiency), SF-A (Görtelmeyer, 2011), MADRE (Schredl et al., 2014)	→	Psychiatric ward inpatient, predominantly female, no sleep criteria reported.	40%
Sheaves et al. (2018)	England	Pilot RCT (Assessor Blind)	2 week Sleep Treatment at Acute Crisis - Adapted CBTI and lightbox if needed.	≤4 weeks	Insomnia (ISI)	─	Psychiatric hospital inpatient, predominantly male, inc. 3+ insomnia symptoms or diagnosis.	100%
Sloane et al. (2007)	USA	A cluster-unit crossover intervention trial with four conditions, inc a control.	High intensity environmental light: (bright light, minimum standard light).	≤4 weeks	Actigraphy – Hours of night-time sleep, Observed Daytime Sleepiness	─	Older adults with dementia, psychiatric facility, predominantly male, no sleep criteria reported.	80%
Stanton et al. (2016)	Australia	Pilot (Pre-Post)	Exercise (20 minute endurance and 20 minute weights/resistance).	N/R	Richard Campbell Sleep Questionnaire (RCSQ)	→	Inpatients, psychiatric ward, gender N/R, no sleep criteria reported.	80%
Tsekou et al. (2015)	Greece	Pre-Post Pilot Study	Pharmacological – clozapine.	≤4 weeks	Polysomnography (PSG) (Total Sleep Time)	→	Male psychiatric inpatients, no sleep criteria reported.	80%
Zhou et al. (2020)	China	RCT (Double Blind)	tDCS procedure	≤4 weeks	Sleep Quality (PSQI), Polysomnography (PSG) (Sleep Efficiency & Total Sleep Time)	↑	Inpatient in general hospital, psychiatric ward with depression and insomnia, predominantly female.	80%
Zhu et al. (2018)	China	RCT (Assessor Blind)	Tai Chi - modified physical activity to accommodate capabilities	3 months+	Sleep Quality (PSQI)	↑	Detained female inpatients (correctional facility), no sleep criteria reported.	60%

↑ indicates an improvement in sleep quality,↓ a decrease in sleep quality, → no statistically significant effect on sleep quality, ─ data regarding significance was not thoroughly reported.

PSQI=Pittsburgh Sleep Quality Index, TSD=Total Sleep Deprivation, LT=Light Therapy, PSG=Polysomnography, RCT=Randomised Controlled Trial, N/R=Not Recorded, STU=Secure Treatment Unit, PTSD=Post Traumatic Stress Disorder, CBT=Cognitive Behavioural Therapy, MAR=Music Assisted Relaxation, RSCQ=Richards-Campbell Sleep Questionnaire, IRT=Image Rehearsal Therapy, ISI=Insomnia Severity Index, ECT=Electro-Convulsive Therapy, PSD=Partial Sleep Deprivation, CBTI=Cognitive Behavioural Therapy for Insomnia, 3Ps=Spielman’s 3P Model: Predisposing, Precipitating, Perpetuating Factors, ESS=Epworth Sleepiness Scale, DBAS=Dysfunctional Beliefs & Attitudes about Sleep Questionnaire, PSAS=Pre-sleep Arousal Scale, SHPS=Sleep Hygiene Practice Sleep, SF-A=Sleep Questionnaire, MADRE=Mannheim Dream Questionnaire, DSM-IV-TR=Diagnostic and Statistical Manual of Mental Disorders: 4^th Edition - Text Revision, rTMS=repetitive Transcranial Magnetic Stimulation, DSM=Diagnostic and Statistical Manual of Mental Disorders, tDCS=transcranial Direct-Current Stimulation.

Descriptive participant demographics & study design

Fifteen studies reported predominantly (>50% or only) male samples and 19 reported predominantly female samples. Three studies did not record participant gender. One recorded an equally gender split sample. Further demographic data regarding age and location can be summed from Table 2. Of the 38 studies, 17 were quantitative randomised-controlled or crossover trials, 19 were non-randomised quantitative studies and two were quantitative descriptive studies. Further study detail regarding number of studies utilising self-report measures, instrumental measures, various intervention durations and various intervention types (behavioural, medical, environmental) can be summed from Table 2.

Risk of bias

The MMAT (Pluye et al., 2011) was used to give each study a quality score, indicative of risk of bias. Nineteen were of high quality, (scoring 80%–100%), 16 of moderate quality (scoring 60%), and three of low quality and at significant risk of bias (all scoring 40%). See Supplementary Material, Appendix 8 for quality scores.

Meta-analysis of sleep interventions

Identification for meta-analysis from systematic review

Studies from the systematic review only (not in the meta-analysis) are listed in Supplementary Material, Appendix 9. A number of studies from the systematic review met inclusion criteria for the meta-analysis, however, due to missing data, were not included. A list of authors contacted for further data is available in Supplementary Material, Appendix 10.

Sensitivity analysis

Twenty-two studies were identified for the meta-analysis, which offered 38 individual trials which incorporated one of the key sleep outcome variables (Sleep Quality, Insomnia Severity, Total Sleep Time, Sleep Efficiency). To ensure that all these trials were eligible for inclusion in the meta-analysis, in terms of outlying data, both visual and statistical analyses were completed. An initial forest plot (Supplementary Material, Appendix 4) was created with all 38 individual trials; visually, two trials were identified as outliers. A Baujat plot (Supplementary Material, Appendix 5) was created to test for heterogeneity across the 38 trials, and again, these two trials were noted. A Grubbs test was then completed to the value of p = 0.05 and analysed against the critical values chart, which indicated that the two trials were statistically outliers in terms of heterogeneity and these were therefore removed from the data set. With the remaining data set (n = 36 trials), publication bias was assessed visually using a funnel plot (Supplementary Material, Appendix 6) and statistically with an Eggers regression; z = 0.02, p = 0.98, indicating no publication bias. Therefore, analysis continued from here on in with n = 36 trials. Finally, twenty-two studies were identified for the meta-analysis, which comprised a total of 36 individual trials (n = 844 control participants or pre-test participants, n = 880 experimental participants or post-test participants).

Descriptive participant demographics & study design

Within the 36 individual trials, 14 participant samples were predominately (>50%, or only) male and 18 were predominately (>50%, or only) female. Three studies did not record participant gender. One recorded an equally gender split sample. Further demographic data regarding age and location can be summed from Table 2. Of the 22 meta-analysis studies, 12 employed a pre-post, within-groups sample design and ten employed a between-groups design with control groups. Further study detail regarding number of studies utilising self-report measures, instrumental measures, various intervention durations and various intervention types (behavioural, medical, environmental) can be summed from Table 2 and are discussed across subgroup analysis.

Risk of bias

Subgroup analysis (random-effects model) of risk of bias quality score found that 26 trials were scored as high quality and yielded an estimated pooled effect size of d = 0.37 (p < 0.001, 95% CI: 0.16, 0.59) with heterogeneity tests reporting, τ2 = 0.00, I² = 0%. There were seven trials of moderate quality, yielding an estimated pooled effect size of d = 1.08 (p < 0.005, 95% CI: 0.34, 1.83) with τ2 = 0.71, I² = 70%. Three were of low quality and yielded an estimated pooled effect size of d = 0.35 (p = 0.38, 95% CI: −0.43, 1.13) with τ2 = 0.00, I² = 0%. The MMAT (Hong et al., 2018) discourages excluding studies with low methodological quality from analysis, thus, those with a lower quality score were still included in all analyses.

Total pooled effect size

A meta-analysis of the total effect of all interventions on sleep quality (random-effects model) using all applicable trials (n = 36) yielded an estimated pooled effect size of d = 0.54, (p < 0.0001, 95% CI: 0.30, 0.77) with a low to moderate risk of heterogeneity (τ2 = 0.20, I² = 38%). Effect sizes ranged from d=-0.38 (95% CI: −1.38, 0.62) in Moon, 2016: Depressive (Sleep Efficiency) (Moon et al., 2016), to d = 2.59 (95% CI: 2.04, 3.14) in Lu, 2013 (Total Sleep Time) (Lu et al., 2013). Figure 2 illustrates the forest plot of all sleep interventions (d = 0.54). Subgroup analysis was performed to examine the differences in effect size (Cohen’s d) between set groups based on categorical variables; research studies’ quality score, age, gender, institutional location and type, intervention type, sleep quality measure used and intervention duration.

Figure 2. Total pooled effect size (Standardised mean difference; cohen’s d) of all interventions to improve sleep quality in secure settings.

SE of d = Standard Error of Cohen’s d, SMD = Standardised Mean Difference, CI = Confidence Intervals, PSQI = Pittsburgh Sleep Quality Index, SE = Sleep Efficiency, TST = Total Sleep Time, ISI = Insomnia Severity Index.)

Subgroup analysis conducted with random effects modelling

Participant demographics

Subgroup analysis was completed to compare the effect of gender on intervention effect, with trials divided between those with a predominantly (>50% or only) male or female sample. Three trials reported an equally split sample (Pyrke et al., 2017), so these were not included in this sub-analysis. One further trial did not specify gender (Chu et al., 2015) and was excluded from this analysis. Trials with a predominately male sample (n = 14) reported a pooled effect of d = 0.80 (p < 0.005, 95% CI: 0.30, 1.30) with τ2 = 0.57, I² = 63%. Trials with a predominately female sample, (n = 18) reported a pooled effect size of d = 0.39 (p < 0.005, 95% CI: 0.12, 0.66) with τ2 = 0.00, I² = 0%. Participant average age was compared categorially. Trials which encompassed participants of an average age of older than 55 years (n = 6) reported the largest pooled effect size; d = 1.24 (p < 0.005, 95% CI: 0.49, 1.99) with τ2 = 0.45, I² = 52%. Those with an average age of 46–55 years (n = 9) reported d = 0.45 (p < 0.05, 95% CI: 0.04, 0.87) with τ2 = 0.08, I² = 21%, those with an average age of 36–45 years (n = 13) reported d = 0.37 (p < 0.05, 95% CI: 0.08, 0.65) with τ2 = 0.00, I² = 0% and those with an average age of 35 or younger (n = 8) reported d = 0.40 (p = 0.23, 95% CI: −0.25, 1.06) with τ2 = 0.54, I² = 60%. Effect of age can be seen in Figure 3.

Figure 3. Sub-group analysis – effect of age.

Trial location

Geographical location was compared with estimated pooled effect size for interventions conducted in Asia (n = 14) reported as d = 0.41 (p < 0.05, 95% CI: 0.03, 0.80) with τ2 = 0.24, I² = 44%, in USA and Canada (n = 6) as d = 0.41 (p = 0.08, 95% CI: −0.05, 0.88) with τ2 = 0.06, I² = 19%, in mainland Europe (n = 13), as d = 0.55 (p < 0.001, 95% CI: 0.18, 0.93) with τ2 = 0.09, I² = 18%, and the UK (n = 3) as d = 1.13 (p = 0.07, 95% CI: −0.10,2.37) with τ2 = 0.89, I² = 75%. Subgroup analysis compared the type of institution in which the trial was conducted, with psychiatric specific facilities such as forensic hospitals (n = 10) reporting a pooled effect size of d = 0.69 (p < 0.01, 95% CI: 0.17, 1.21) with τ2 = 0.40, I² = 58%; other institutions such as general hospitals with a psychiatric ward (n = 23) reporting d = 0.35 (p < 0.005, 95% CI: 0.12, 0.58) with τ2 = 0.00, I² = 0% and secure correctional facilities with mental health care teams (n = 3) reporting d = 1.22 (p < 0.05, 95% CI: 0.06, 2.38) with τ2 = 0.73, I² = 69%.

Study design

Subgroup analysis was conducted to investigate the effect of different intervention types; behavioural (CBTI, mindfulness, physical activity), medical (pharmacological, electro-convulsive and non-behavioural individual therapies), and environmental (change to hospital design and protocol). Behavioural interventions (n = 10) (Brupbacher et al., 2021; Chien et al., 2015; Ellis et al., 2019; Ferszt et al., 2015; Haynes et al., 2011; Hegde et al., 2020; Holzinger et al., 2020; Randall et al., 2019; Sheaves et al., 2018) yielded the largest estimated pooled effect size; d = 0.65 (p < 0.01, 95% CI: 0.16, 1.14) with τ2 = 0.32, I² = 52%. Medical interventions (n = 21) (Baune et al., 2007; Göder et al., 2016; Henriksen et al., 2020; Lu et al., 2013; Moon et al., 2016; Nakamura & Nagamine, 2017; Nishida et al., 2017; Schäfer et al., 2019; Tsekou et al., 2015; Zhou et al., 2020) yielded an estimated pooled effect size of d = 0.58 (p < 0.001, 95% CI: 0.25, 0.91) with τ2 = 0.22, I² = 38%. Environmental interventions (n = 5) (Canazei et al., 2019; Chu et al., 2015; Pyrke et al., 2017) yielded an estimated pooled effect size of d = 0.19 (p = 0.38, 95% CI: −0.24, 0.62) with τ2 = 0.00, I² = 0%.

Subgroup analysis compared the type of measure recorded; self-reported (n = 15) including the PSQI (Buysse et al., 1989) and the ISI (Morin, 1993) or instrumental (n = 21), including wrist-worn actigraphy and polysomnography to measure sleep efficiency and total sleep time. Interventions utilising self-report measures reported a calculated pooled effect size (random-effects model) of d = 0.72 (p < 0.0001, 95% CI: 0.39, 1.05) with τ2 = 0.12, I² = 30%, and those with instrumental measures reported d = 0.38 (p < 0.05, 95% CI: 0.05,0.71) τ2 = 0.23, I² = 40%.

Interventions with a duration of up to and including one week (n = 10) reported a pooled effect size of d = 0.15 (p = 0.39, 95% CI: −0.19, 0.48) with τ2 = 0.00, I² = 0%. Those lasting eight days to four weeks (n = 18) reported d = 0.65 (p < 0.005, 95% CI: 0.26, 1.04) with τ2 = 0.41, I² = 58%. Those lasting 5–10 weeks(n = 5) reported the largest pooled effect size; d = 0.92 (p < 0.005, 95% CI: 0.29, 1.55) with τ2 = 0.00, I² = 0%. Those lasting >10 weeks (n = 3) reported d = 0.50 (p = 0.12, 95% CI: −0.14, 1.13) with τ2 = 0.00, I² = 0%. This can be seen in Figure 4.

Figure 4. Subgroup analysis - effect of intervention duration.

Discussion

Despite the increasing amount of research that has examined the effect of behavioural, medical and environmental interventions on sleep outcomes for people with a mental health disorder, to the best of our knowledge, no systematic review or meta-analysis has been conducted to synthesise the findings within inpatient settings. After a rigorous selection method, 38 studies were included in the present systematic review (Table 2), and 22 studies comprising 36 trials (Figure 2), were included in the meta-analysis. Overall, the majority of studies in the systematic review − 69% - included a self-reported measure of sleep quality, and 31% reported use of an instrumental measure; four studies used both self-report and instrumental data. Most studies found an improvement in sleep quality outcomes, with authors reporting a positive effect on sleep quality in 47% of cases, 36% finding no effect and 2% finding a negative effect on sleep quality. Direction of results were unclear in approximately 16% of cases. Results indicated that, as a pooled value, interventions for improving sleep quality are generally effective in this population to a medium size (Cohen, 1988). Interventions, of any type, conducted over a period of five to 10 weeks were more effective than those conducted over a longer or shorter duration. Behavioural interventions had the highest effect size for improving sleep quality.

Behavioural interventions included CBT with breathing techniques (Chien et al., 2015), mindfulness practice (Ferszt et al., 2015), physical activity (Brupbacher et al., 2021; Zhu et al., 2018), image rehearsal therapy (Ellis et al., 2019), lucid dream therapy (Holzinger et al., 2020), behavioural group therapy (Haynes et al., 2011) and CBTI either with optional additional light therapy (Sheaves et al., 2018) or alone (Randall et al., 2019). It is noted that the behavioural group encompassed one of the trials (Randall et al., 2019) with the largest effect sizes (d = 2.35), however, the behavioural subset remained the most effective even when modelled without this trial (data not shown). This confirms findings from general population studies which found behavioural interventions, specifically CBTI and physical and mind-body exercises (Xie et al., 2021), to be consistently effective for improving insomnia symptoms (Rios et al., 2019) and improving subjective sleep quality (Xie et al., 2021). Further, it supports the use of behavioural interventions for insomnia in other clinical groups (Johnson et al., 2016) and in psychiatric outpatient populations (Wagley et al., 2013), and justifies the inclusion of behavioural approaches as the first line of treatment for insomnia, as recommended by European clinical guidelines (Riemann et al., 2017).

The behavioural interventions that were identified all encompassed some form of active patient engagement, through attendance and completion of a therapy session [see Sheaves et al., 2018] or actively engaging in physical activity. [See Zhu et al., 2018] In comparison, the medical interventions group, comprising of pharmacological trials, [see Nakamura et al., 2017] and electroconvulsive and transcranial current stimulation trials [see Nishida et al., 2017] were effective, with a moderate, but smaller, effect size. Medical interventions used in these populations are often managed by staff and not controlled independently (Hipp et al., 2021), which could cause some patient distress (Soliman & Reza, 2001), which could in turn hinder the positive effects of such interventions. Previous studies have indicated that stakeholder volitional involvement is key for interventions looking to improve health outcomes (O’Cathain et al., 2019). Thus, our findings support the higher effectiveness of behavioural interventions, which may be in part due to the active volitional involvement of patients.

The interventions which focussed on the patient’s environment in terms of ward and bedroom design (Pyrke et al., 2017), external light exposure (Canazei et al., 2019) and night-time confinement (Chu et al., 2015) were seen to be effective to a smaller degree (Cohen, 1988). It is plausible that the particular trials which were included in this subgroup held too many methodological limitations for a larger effect size to be achieved. Pyrke and colleagues (Pyrke et al., 2017) reported problems with recording self-reported sleep data in this population, and Canazei and colleagues (Canazei et al., 2019) did not use an instrumental measure. Chu and colleagues (Chu et al., 2015) reported overall practical difficulties in recording patient data hospital wide, and Pyrke and colleagues (Pyrke et al., 2017) suggested that institution wide design elements have differential impact depending on the individual’s presenting problems and treatment plans. This would suggest that, by their complex nature, environmental interventions included here may be too broad to be conducive to measurable individual patient effect.

The effect of intervention (any type) increased from those conducted over less than or equal to one week, through to a peak at five to ten weeks before decreasing again among those lasting any longer. This is fitting with a large (n = 1,891) RCT which investigated the effects of CBTI for insomnia on mental health outcomes (Freeman et al., 2017), and found a sleep-focused treatment intervention of ten weeks could significantly reduce insomnia, paranoia symptoms and hallucinations (all p < 0·0001). This is also fitting with a recent meta-analysis on exercise use for improving insomnia and sleep quality, which identified that short-term interventions (≤3 months) had a significantly greater reduction in sleep disturbance compared to long-term interventions (>3 months), though this did vary between exercise type (Xie et al., 2021). Further, cognitive deficits related to mental health disorders can often lead to a reduced attention span, inadequate retention of taught material and poor adherence to scheduling (Kwan et al., 2014), which could explain a drop off in effect, as patients may begin to lose interest in the intervention.

Subgroup analysis revealed that the trials encompassing a predominantly male sample reported a larger pooled effect size than those which were predominantly female. This suggests that men had a significantly greater improvement in sleep outcomes over the course of interventions than women. This may be due to women experiencing poorer sleep quality than men, in that insomnia symptoms are approximately 1.5 times more common in women (Zhang & Wing, 2006) and night-time awakenings are more prevalent for women (Jonasdottir et al., 2021). Women also experience biological and hormonal changes throughout the lifespan, such as menopause, which can increase insomnia risk and sleeping difficulties (Zhang & Wing, 2006). Therefore, as women can be said to have sleep problems which are more multi-factorial in nature than men, their sleep symptomology may be more difficult to resolve. This could explain why the present analysis identified that females report the sleep improvement interventions to be less effective than males.

Trials conducted in the UK reported the largest effect size, followed by those in mainland Europe, then Asia and the USA and Canada. Although the global demand for insomnia self-help information has increased continually since 2004 (Kirsi-Marja et al., 2021), there are large discrepancies between healthcare systems internationally (Rada, 2019). The UK has the lowest rates of unmet healthcare needs, compared to areas of mainland Europe, USA and Canada (Papanicolas et al., 2019) and so this may be why sleep improvement interventions in the UK were more effective than these areas, as the nation is well prepared to provide effective healthcare. It may also be due to the UK having a high prevalence rate of insomnia (Hartescu & Morgan, 2019). A recent international study compared the prevalence rates of insomnia in South Africa, Australia, South Korea and China and found the UK prevalence rate to be the highest at 14.3%, compared with just 4.1% in China (Hartescu & Morgan, 2019). It is therefore possible that the participants in the UK studies had higher prevalence rates of insomnia and associated poor sleep quality at baseline, allowing for larger significant increases post-intervention, compared to participants from other countries or continents.

The strengths of the present paper include the comprehensive and rigorous literature search and data synthesis. Further, both self-reported and instrumental sleep outcomes were included which allows for a more holistic overview of the effects of interventions on sleep. The novel nature of the review itself and the substantial amount of literature which has been covered to span a 15-year period is advantageous. However, there are some limitations which need to be considered. A number of relevant factors associated with sleep in this population, such as psychiatric diagnosis, medication and admission length were not reported in full for a number of trials, and so it was not feasible to examine these factors in subgroup analyses. Further, there were a small number of subgroup analyses in which there was as issue of heterogeneity, however, this was expected and was the reason for conducting a random-effects model meta-analysis. In addition, the total pooled effect of all interventions did not report substantial heterogeneity (Higgins & Thompson, 2002) nor did the behavioural interventions model.

As the need to provide both preventative and treatment options in mental health is greater than ever and with individual, community and societal costs of mental health ever-rising (Wykes et al., 2021), it is increasingly pressing to develop feasible and effective sleep interventions in this population as mental health and sleep are significantly related (Harvey, 2008).

The current work identified that both medical and behavioural interventions were effective in improving inpatient sleep quality, to a moderate degree, though the behavioural interventions were seen to be more effective. Attitude towards medication is an important factor in predicting long term success with treatment in severe mental illness, however, inpatients often have a negative attitude towards taking their psychiatric medication (Kondrátová et al., 2019). It is therefore suggested that healthcare professionals utilise behavioural methods as a first line of treatment for improving sleep quality.

The present review can be utilised by healthcare professionals when making decisions on intervention strategies for improving sleep in this population. The specific design features explored in the review, such as intervention duration, highlight study characteristics which lead to better results for sleep quality and as such, should be implemented in future intervention development. Healthcare professionals may wish to focus on behavioural therapies such as CBTI and physical activity to improve sleep for this population and to focus on patient driven, tailored interventions which align to feasible time periods.

Author contributions

PG contributed to all aspects of the study from conceptualisation to data curation, investigation, methodology, software use and validation, original manuscript writing and editing. FK contributed to conceptualisation, methodology, validation, editing, supervision. KB contributed to interpretation of data, editing, supervision. AG contributed to conceptualisation, methodology, editing and supervision. IH contributed to conceptualisation, data curation, investigation, validation, visualisation, editing, supervision.

All authors revised and approved of the final version of the manuscript.

Data availability

The study protocol, including analytic plan, was pre-registered with PROSPERO prior to data collection (reg: CRD42020177944). Data, analytic code and materials are not available in an a public archive. Data can be made available (upon reasonable request) by emailing the corresponding author. The data does not currently belong in any research depository.

Geolocation

This research was conducted at Loughborough University, Leicestershire, UK.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/14789949.2022.2111320

Additional information

Funding

The project is funded by Loughborough University, Leicestershire, UK and St Andrew’s Healthcare, Northampton, UK.