Pharmacological and non‐pharmacological interventions to enhance sleep in mild cognitive impairment and mild Alzheimer's disease: A systematic review

Abstract Suboptimal sleep causes cognitive decline and probably accelerates Alzheimer's Disease (AD) progression. Several sleep interventions have been tested in established AD dementia cases. However early intervention is needed in the course of AD at Mild Cognitive Impairment (MCI) or mild dementia stages to help prevent decline and maintain good quality of life. This systematic review aims to summarize evidence on sleep interventions in MCI and mild AD dementia. Seven databases were systematically searched for interventional studies where ≥ 75% of participants met diagnostic criteria for MCI/mild AD dementia, with a control group and validated sleep outcome measures. Studies with a majority of participants diagnosed with Moderate to Severe AD were excluded. After removal of duplicates, 22,133 references were returned in two separate searches (August 2019 and September 2020). 325 full papers were reviewed with 18 retained. Included papers reported 16 separate studies, total sample (n = 1,056), mean age 73.5 years. 13 interventions were represented: Cognitive Behavioural Therapy – Insomnia (CBT‐I), A Multi‐Component Group Based Therapy, A Structured Limbs Exercise Programme, Aromatherapy, Phase Locked Loop Acoustic Stimulation, Transcranial Stimulation, Suvorexant, Melatonin, Donepezil, Galantamine, Rivastigmine, Tetrahydroaminoacridine and Continuous Positive Airway Pressure (CPAP). Psychotherapeutic approaches utilising adapted CBT‐I and a Structured Limbs Exercise Programme each achieved statistically significant improvements in the Pittsburgh Sleep Quality Index with one study reporting co‐existent improved actigraphy variables. Suvorexant significantly increased Total Sleep Time and Sleep Efficiency whilst reducing Wake After Sleep Onset time. Transcranial Stimulation enhanced cortical slow oscillations and spindle power during daytime naps. Melatonin significantly reduced sleep latency in two small studies and sleep to wakefulness transitions in a small sample. CPAP demonstrated efficacy in participants with Obstructive Sleep Apnoea. Evidence to support other interventions was limited. Whilst new evidence is emerging, there remains a paucity of evidence for sleep interventions in MCI and mild AD highlighting a pressing need for high quality experimental studies exploring alternative sleep interventions.

The traditional view has been that AD causes sleep impairment and the extent of symptomatic sleep disturbance correlates with the severity of dementia (Benca et al., 1992;Montplaisir et al., 1995;Pat-Horenczyk et al., 1998;Prinz et al., 1982;Weldemichael & Grossberg, 2010). Circadian rhythm disorders also contribute to sleep disturbance and worsen with age and AD, possibly related to alterations to the Suprachiasmatic Nucleus secretion of melatonin (Swaab et al., 1985).
A recent prominent theory postulates a bidirectional relationship between poor sleep and AD (Mander et al., 2016) -as well as AD causing sleep disturbance, sleep disturbance may lead directly to pathological accumulation of proteins that cause neurodegeneration (Fultz et al., 2019;Hahn et al., 2014). Beta amyloid is one of the two pathognomonic changes of AD. Levels of soluble Amyloid Beta fluctuate diurnally in both mice and humans showing the crucial role that sleep plays in its clearance (Huang et al., 2012). Amyloid Beta 42 levels have been shown to be significantly increased after a single night of sleep deprivation in healthy adults (Ooms et al., 2014) and slow oscillations are temporally linked to CSF flow (Fultz et al., 2019). Chronic sleep deprivation accelerates accumulation of soluble beta amyloid into insoluble amyloid plaques in two mouse models (Kang et al., 2009). A cascade is envisaged in which poor sleep disrupts the clearance of soluble amyloid, leading to plaque deposition. This precipitates further plaque formation through recognised positive feedback loops (Mandrekar-Colucci et al., 2012).
Regardless of whether sleep disturbance arises directly from accumulation of AD pathology or through an independent problem such as sleep apnoea (Ancoli-Israel, 2000;Bliwise, 1999;Van Cauter et al., 2000), poor sleep might be a significant contributory factor in the eventual onset and progression of AD.
AD pathology progresses for 1-2 decades before a full dementia diagnosis giving plenty of time for factors such as sleep disturbance to affect the rate of pathological progression (Jack et al., 2010). Therefore, the optimal opportunity to reverse or halt cognitive decline caused by AD pathology, thus maintaining quality of life, will be early in the course of disease -at the Mild Cognitive Impairment or mild AD stage.
Early diagnosis utilising molecular biomarkers to accurately pinpoint the cause of cognitive symptoms is becoming more common, most frequently for identification of early/prodromal AD (Sutphen et al., 2014) and this provides an opportunity for targeted intervention.
Disruption of sleep occurs early in AD (Ju et al., 2013;Lim et al., 2013;Sprecher et al., 2017), even before symptoms arise, however, the focus of sleep intervention reviews to date have largely been at the stage of established AD dementia (Mitolo et al., 2018;O'Caoimh et al., 2019) when pathology may be too advanced to change. Sleep disorders are also common in this population affecting up to 60% of patients in memory clinics (Littlejohn et al., 2014).
Furthermore, sleep deprivation brings a range of other metabolic and cardiovascular impairments that mean improving sleep is likely to be broadly beneficial in the memory clinic population over and above the potential to modify progression of AD. However, there are no standard sleep treatment protocols within memory clinics and other dementia services.
There are multiple theoretical targets for sleep enhancement in AD (summarised in Figure 1). Much of the previous literature has focussed on pharmacological interventions (McCleery et al., 2016). Notable disadvantages of this approach, particularly in older adults, include the potential for side effects, interactions and polypharmacy (Maher et al., 2014).

Hence, interest in non-pharmacological sleep-modifiers is intensifying
with recent technological advances permitting exploration of novel approaches such as closed loop stimulation of slow wave sleep through sound or electrical brain stimulation (Ngo et al., 2013) and glasses to deliver Bright Light Therapy (Sekiguchi et al., 2017). Lifestyle interventions are often cost effective and non-toxic and have the potential to enhance sleep in early stages of AD, but have not previously been the focus of reviews in mild AD and MCI. We include all potential types of intervention, including lifestyle interventions, in the searches for this review.
Here we systematically review the literature on sleep interventions for people meeting established criteria for either Mild Alzheimer's Disease or Mild Cognitive Impairment. We focus specifically on AD rather than other subtypes of dementia as there are specific mechanisms proposed to link sleep disturbance to AD. In the absence of trials of sufficient length or size to establish the efficacy of sleep disturbance treatment on AD disease progression, we first review studies that measure the effect of intervention on sleep as an initial step in the translational pathway to slowing AD progression.
This review does not explore the effect of interventions on specific sleep phenotypes due to the paucity of data. Improving sleep regardless of presence or absence of sleep disorder could plausibly impact on AD. Therefore inclusion and exclusion of any underlying sleep disorders in study design was noted, permitted and discussed.
In order to assess the degree to which different avenues for treating sleep in early AD have been explored, we, a priori, created a logic model to incorporate different possible intervention targets (Figure 1).
We then compare the scope of existing studies with the model to guide where gaps might lie. The 9 targets included; promotion of wakefulness (to build sleep pressure), optimising circadian fluctuations, improving bed time routines (to remove barriers to sleep), removing overnight sleep interventions in MCI and mild AD highlighting a pressing need for high quality experimental studies exploring alternative sleep interventions.

| ME THODS
The review followed recommendations by the Preferred Reporting Items for Systematic Review and Meta-Analyses Statement (PRISMA) (Moher et al., 2009). The protocol was registered in the Prospero Database (registration number CRD42019126320).

| Eligibility criteria
All peer reviewed articles written in English were eligible for inclusion. The following criteria were used:

| Participants
Inclusion criteria: 1. Adults aged greater than 18 (limit set to avoid excluding studies in genetic dementias); and 2. Male or Female; and 3. At least 75% of the studied group meet the following criteria: Exclusion criteria: a. Satisfies established diagnostic criteria for MCI e.g. Albert Criteria (Albert et al., 2011), Peterson Criteria (Petersen et al., 1999) or would be expected to meet these criteria if the study was conducted before 1999.

Or:
b. Satisfies established diagnostic criteria for AD demen-  et al., 1984), or would be expected to meet these criteria if the study was conducted before 1984.

| Exposure
Interventions of interest include any pharmacological or non-pharmacological treatment/technique primarily utilised to improve the duration or quality of sleep.

| Comparator
The studied intervention must be compared with at least one other intervention, non-exposure or placebo.

| Outcome
Sleep outcomes e.g. total sleep time, sleep efficiency or sleep latency measured by an appropriate validated sleep outcome measure.

| Study design
Eligible studies included comparison study designs e.g. randomised controlled trials (RCTs), group trial designs, case-control, cross-sectional and prospective cohort studies as well as multiple base studies. Systematic reviews and meta-analyses were reference tracked. Case reports, abstracts and grey literature were excluded.

| Study selection
Search results were merged using reference management software and duplicate records of the same report removed. A sample of 10% of the total titles were reviewed by each reviewing team member to ensure consensus in decision making. Remaining titles were reviewed individually by a member of the reviewing team in order to exclude studies obviously not meeting criteria.
Abstracts of the retained references were examined against inclusion and exclusion criteria with a further sample of 10% reviewed by each reviewing team member to ensure consensus. The full texts of remaining articles were checked against inclusion/exclusion criteria by two members of the review team. In the process of the search, in more detail to avoid over-representation of data. Quality appraisal and risk of bias assessment identical due to same study design. b 2 separate papers written by Cruz-Aguilar reporting data from the same study. Analysis focussed on combined findings. Quality appraisal and risk of bias assessment identical due to same study design.

TA B L E 1 (Continued)
the reference lists of any identified relevant review articles were hand searched to ensure capture of references not identified on the electronic databases.
Retained studies were recorded in a spreadsheet and included study characteristics, details of the patient/population group, intervention and primary outcomes. Where required, correspondence with investigators was attempted in order to gain further study data or to clarify study eligibility. Discrepancies were resolved through discussion with the first author (JB) in consensus with the reviewing team.

| Data extraction
Data extracted from the retained articles (Table 1)

| Quality assessment
Methodological quality of eligible studies was evaluated using the Joanna Briggs Institute Checklist for Quasi-Experimental and Randomized Controlled Trials (Tufanaru et al., 2017). Independent evaluation of quality was undertaken by two reviewers (SH and JB) and consensus reached jointly.
Risk of bias for all included studies was also assessed using the Cochrane Collaboration's Tool for Assessing Risk of Bias (Higgins et al., 2019). Two reviewers (SH and JB) undertook this analysis independently before reaching joint consensus.

| Data synthesis and analysis
Due to substantial heterogeneity in outcome measures no metaanalysis of quantitative data was appropriate. Therefore a narrative synthesis of included studies was performed focusing on population characteristics, interventions utilised and outcomes. The main focus of analysis was on the performance of the intervention assessed against a comparison intervention or control.

| Included studies
The search strategy identified 27,609 unique records alongside an additional 44 identified through bibliographic search. 790 abstracts and subsequently 325 full text articles were screened for eligibility with a total of 18 articles selected for inclusion which were published between 1998 and 2020 (see PRISMA flowchart - Figure 2).
There were two predominant reasons for exclusion, (a) ineligible study type e.g. intervention not analysed against an alternative or control, (b) ineligible population e.g. due to severity of dementia.  The 18 articles identified reported on data from 16 separate studies. Attributes of individual included studies are listed in Table 1.

| Participants and comparison measures
A summary of combined included study characteristics is shown in

| Outcome measurement and outcomes
As per

| Quality appraisal and risk of bias
Quality appraisal was undertaken independently by two reviewers  however this was partly due to inherent difficulties in concealment and blinding associated with utilising a psychotherapeutic intervention. One further study had 3 limitations identified (Cooke et al., 2006), three studies had two limitations identified (Naismith et al., 2018;Petit et al., 1993;Wang et al., 2020). The remaining eleven studies had 0 or 1 limitation.

| Summary of interventions
Overall we found trials in seven of the nine intervention domains identified in Figure 1. Was true randomisation used for assignment of participants to treatment groups?
Was allocation to treatment groups concealed?
Were treatment groups/participants similar at the baseline?
Were participants blind to treatment assignment?
Were those delivering treatment blind to treatment assignment?
Were outcome assessors blind to treatment assignment?
Were treatment groups treated identically other than the intervention of interest?
Was follow-up complete or differences adequately described/analysed?
Were participants analyzed in the groups to which they were randomised?
Were outcomes measured in the same way for treatment/participant groups?
Were outcomes measured in a reliable way?
Was appropriate statistical analysis used?

Galantamine
Galantamine was found to have little or no impact on sleep architecture by Cooke et al. (2006) 2015) found in a non-randomised trial that galantamine when compared against donepezil, rivastigmine and a healthy control group significantly decreased total PSQI score (p = .028). The highest improvement rate was in those with reported poor sleep quality.

Rivastigmine
Rivastigmine was not associated with any change in sleep architecture according to Cooke et al. (2006) although the authors acknowledge that their small sample size (n = 8) is likely to have reduced the power to detect a significant effect. Naharci et al. (2015) also found that rivastigmine decreased total PSQI score although this did not reach statistical significance (p = .193).

| Enhancing NREM
(t = 4.70, p = .002) and (t = 3.62, p = .008) respectively. There was a significant increase in C3-A1 and C4-A2 interhemispheric coherence of slow bands; delta (t = −3.64, p = .008) and theta (t = −3.93, p = .005) but also of gamma (t = −5.13, p = .001) and a decrease in beta 1 coherence (t = 3.91, p = .005). There was also a decrease in all EEG band F7-T3 and F8-T4 interhemispheric coherence following melatonin administration. The authors conclude that shorter NREMS onset following melatonin administration in participants with mild-moderate AD is associated with the RP and EEG coherence differences described above and that these are likely to have been achieved through GABA-A modulation suggesting that these pathways may be well-preserved in this patient group.

Continuous Positive Airway Pressure (CPAP)
Patients with Obstructive Sleep Apnoea (OSA) but no prior diagno- were randomised to three strengths of aroma oils (0.1%, 0.5% and 1%) with bathing performed nightly after 18:00 hr for 24 weeks. No significant changes in PSQI-J were observed during the study period in all groups.

| Daytime behaviour
A structured limbs-exercise program These adaptations to standard CBT were hypothesised to improve both objective actigraphy variables and self-reported insomnia symptoms through the ISI.
CBT-I was shown to have a highly significant desirable effect on  The largest trial in this review conducted by Herring et al. (2020) showed that the dual orexin receptor antagonist Suvorexant led to significant macro-architectural changes in sleep including an increase in TST, SE and a reduction in WASO. Of note, improvements in TST significantly favoured Suvorexant only in the subgroup with mild (21-26) MMSE scores and not in the moderate group, highlighting potential differences in treatment efficacy between these groups. However, increased rates of daytime somnolence and adverse events may limit its potential adoption in this population.
The most widely studied intervention were the AChEIs likely due to a combination of theoretical, clinical and pragmatic reasons.
Early neurochemical theories of sleep hypothesised Acetylcholine playing a primary role in wakefulness and REM sleep (Jouvet, 1972), further strengthened by later work underlying the muscarinic M2 receptor's role in potentiating REM sleep (Baghdoyan & Lydic, 1999;Watson et al., 2012). Clinically, donepezil has been associated with vivid dreams and insomnia with higher incidences versus placebo reported in multiple studies (Burns et al., 1999;Pratt et al., 2002;Rogers et al., 1998) which alongside their common use in those with early cognitive impairment may have driven further research.
The results of trials included here did not reveal a consistent benefit of AChEIs on sleep and it seems unlikely this class of medications will be pursued further to benefit sleep. The one exception to this could involve new technology for drug delivery to modulate available acetylcholine to better mimic physiological fluctuations in cholinergic activity. Given we know that cholinergic synapses are involved in learning while awake (Huerta & Lisman, 1993) and less so during sleep, perhaps more refined temporal target (e.g. daytime only) could improve nocturnal sleep and memory consolidation. We generally noted that there were no identified studies where a daytime-only pharmacological intervention, promoting wakefulness or attention, was utilised as a tool to enhance night-time sleep, and this could be a useful future approach.
In a population with OSA and mild AD/MCI, CPAP does appear to be an effective intervention with significant positive effects demonstrated both in sleep architecture and key sleep parameters. Sleep apnoea affects a large proportion of people aged over 50 (Heinzer et al., 2015), therefore treating sleep apnoea holds great promise to improve quality of life and delay AD progression. Sleep apnoea is sufficiently common that screening should be a routine part of clinical practice in memory clinics -we already know that patients can benefit through treatment of sleep apnoea regardless of whether they have AD. The possibility of treating AD is an added benefit.
We also propose that screening and, if required, treatment of sleep apnoea should occur before entry into AD clinical trial of any type.
Unless we exclude the impact of sleep apnoea on AD progression, we may be clouding benefit from other therapeutic approaches.
Two studies explored the effect of tailored psychotherapeutic programs designed on principles of Cognitive Behavioural Therapy.
Outcomes were mixed, with Naismith et al. (2018)  A structured exercise program in participants with MCI was associated with a statistically significant reduction in PSQI score although there was a corresponding significant decrease in Global Depression Score raising questions as to causality direction. Further evidence is likely to arise from the currently recruiting Dementia-MOVE randomised controlled trial (Haeger et al., 2020). This intends to explore the effects of a long-term, 6 month, multicomponent exercise program on brain metabolism together with actigraphic and subjective measures of sleep in participants with early AD.
Overall, aside from two larger trials, the studies were of relatively There were multiple interventions such as vitamin B12, risperidone, acupressure, bright light therapy, transcutaneous electric nerve stimulation (TENS) and music interventions that were not included in our review either due to the use of unvalidated sleep outcome measures or because there was no comparison group to allow for meaningful conclusions to be drawn. In four cases the diagnoses of mild AD/MCI could not be confirmed to have been made using recognised criteria.
Lifestyle modification was explored only in the context of psychotherapeutic approaches rather than in isolation. It was with regret that a high quality randomised controlled trial in individuals from a general community setting with reduced MoCA scores, testing the effects of a multimodal tailored lifestyle intervention in improving actigraphic and subjective measures of sleep (Falck et al., 2020) could not be included due to our pre-determined strict methodology regarding diagnostic criteria. The intervention consisted of sleep hygiene measures followed by the formulation of an individually timed Bright Light Therapy and was shown to have improved measures of subjective but not objective sleep quality. Whilst in established mild-moderate AD dementia there are concerns about cognitive limitations that might prevent behavioural change, it is our belief that this type of tailored intervention may well be a fruitful route for further exploration, particularly given that cognitive function at this stage often still allows significant adaptation of lifestyle and motivation can be high. Furthermore, pharmacological interventions are inherently associated with potential toxicity and the burden of polypharmacy.

| Strengths and limitations
This is the first review we know of to focus on sleep intervention in MCI and Mild AD. A comprehensive electronic database search protocol was performed with a wide range of potential sleep interventions sought. Additionally, we summarized findings on multiple dimensions of sleep outcome data. A process for rigorous justification of all subsequent decisions in terms of inclusion/exclusion was utilised. As with the inclusion criteria for AD dementia/MCI, attempts were made to contact authors in cases of ambiguity.
A significant challenge in this review was in formulating inclusion criteria to ensure that studies involving important, clinically relevant interventions were included but also ensuring that studies with an overly heterogeneous population whose findings may not be applicable to our population of interest were excluded. A threshold of 75% was chosen for participants to have an established diagnosis of AD dementia or MCI. It was further decided to exclude any study where the majority of participants have moderate or severe dementia. Naturally these cut-offs could be criticised either for diluting the population of interest or for failing to allow for inclusion of relevant studies. We therefore cannot exclude the possibility that there may be important findings left unreported.
Included studies reported on a range of diverse outcomes, which whilst individually valid, made comparison across studies challenging. Direct comparison was further hindered by variations in study designs permitting or excluding the presence of underlying recognised sleep disorders amongst participants. Ideally, the effect of interventions on specific sleep phenotypes would have been reviewed, however unfortunately this information was not reported on the scale to allow for meaningful comparison. Nonetheless our decision to include studies regardless of phenotype could be argued to increase real-world applicability to the clinical population and is not contrary to the central hypothesis that an improvement in sleep regardless of baseline may positively impact on AD progression. Quality appraisal scores across included studies were variable and with one exception, those included were largely subject to at least one domain of significant bias thereby potentially compromising their findings. This though, remains consistent with our assertion that there is a paucity of reliable, large-scale data for this population. For this present review, the risk of publication and language bias for current literature is also recognised.

| Clinical implications and future direction
The lack of rigorous evidence presents a significant future research demand. Those with MCI and mild AD might not respond identically to interventions with more robust evidence in the general population or in those with severe AD dementia. Sleep disturbance may well play a role in triggering or facilitating pathophysiology in AD and is the subject of current, ongoing research. Finding evidencebased treatments for sleep disturbance in a population before more severe manifestations of AD dementia develop, may therefore even offer the potential to delay the onset of more significant functional impairment. Of note, the included studies had a diversity of aims including symptomatic relief of insomnia, optimisation of macro or micro-architectural sleep parameters and assessing sleep disturbance as an effect mediator for cognitive impairment. In the future, a more standardised set of outcome criteria may be helpful in allowing efficacy of interventions to be more meaningfully and directly compared. As understanding develops of the precise mechanisms by which sleep contributes to the progression of AD pathology, this may occur naturally.

| CON CLUS ION
In the specific population of MCI and mild AD there is a relative With additional thanks to the North Bristol NHS Trust Library

CO N FLI C T O F I NTE R E S T
None.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available on request from the authors.